10, 18 and 28. They're 3x4, 5x4 and 5x6 designs, with two segments earmarked for the DRAM controllers. Consumer parts being LCC/HCC is easy - the change in the Xeon line naming has made it harder to determine which is which on the enterprise side.
I was wondering here... AMD finally wakes up, gives 16 cores to consumer, then Intel responds by giving 18. but that TR4 socket is big enough to fit the 32core design from AMD.
So I strongly suspect that AMD will launch new variants of Threadriper based on the recent Intel announcements.
5*4 (20 cores), 6*4 (24), 7*4 (28) seems like a given, as it would still save quite a lot of dies that are not able to go to Epyc. in fact this seems very clever and I have to give credit to AMD again: single die to all segments. not the best at the top and not the best at the bottom. but great value in all mid-range segments.
I'm willing to bet that the TR-1960X (20-core) and TR-1970X (24-core) are already made, packaged and sitting in an AMD warehouse somewhere ready to ship. I would expect AMD to announce them two, maybe three days before the i9-7980XE's launch date. I don't expect to see a 28 or 32-core because I think that they need to have at least some distance from EPYC. Since Intel has decided not to make Skylake-X support ECC RAM, they have that divide between Skylake-X and Xeon. On the other hand, AMD, being AMD, has a significantly more attractive offering that DOES support ECC RAM so they have to try a bit harder to keep that divide between Threadripper and EPYC. It won't matter though because the i9-7980XE would be no match for the (theoretically-named) TR-1970X's extra six cores. Since Intel is still using the conventional monolithic manufacturing process, there's no chance in hell that they'll be able to react quickly enough if AMD pulls a stunt like that. I would personally die laughing if that happened.
fully agreed. 24 core to fight against 18 core would be a great marketing and technical feat, while still keeping a safe distance from epyc (but with the possibility to go crazy and reach 28 or 32).
let's see what happens, those are exciting time - CPU wars again!
"The top Core i9-7980XE SKU will have a base frequency of 2.6 GHz but a turbo of 4.2 GHz, and a Favored Core of 4.4 GHz. I suspect the turbo will be limited to 2-4 cores of load, however Intel has not listed the ‘all-core turbo’ frequencies which are often above the base frequencies, nor the AVX frequencies here."
Both are for 2 core loads, lower turbo boost 2.0 and higher turbo boost 3.0
Just wondering how high those all core freqs will be, Amds 16 core TB 1950x has all core turbo of 3.6GHz. And by looking those intel xeons that could be in same ball park.
High turbo clock announcements are to smear on the reality of the rather lousy 2.6 base clock. Because you know what, the target market won't run single threaded workloads for short amounts of time, such machines are being worked 100% of the cores at 100% of load for many hours a day.
Basically, even this revamped xeon, sold 1500$ cheaper just to save face will not be able to match TR. Well, at least it has more cores, that's gotta count for something.
I mean the 7980XE will have to squeeze out at least 30% more performance than then 18 core 2699 v3 to get a definite edge over TR. It has 13% of clock advantage, which will not translate in less than 13% performance increase, and like 10% IPC advantage, which would put it on par or a tad lower than the top of the line TR, I am taking CB performance here.
Oh, and it costs twice as much as the TR... Oh and no ECC.
Just saying, Intel mentioned that the stock 16 core scores 3200 on Cinebench. That's about 7%+ better than TR 16C. Maybe overclocking Threadripper could put it even. The 18 core is going to be faster than "on par or slightly below" TR.
The 7980X appears to have lower clocks than the 7960X (2.6 vs. 2.8). They all have a 4.4GHz max turbo speed. IPC, at least as concerns floating point work, appears to be with Intel, but power consumption is going to be very interesting.
Someone's already linked the tweet. You really think @intelnews is going to tweet something that will EASILY be disproved wildly false? Sure, I'll grant, it's easy to mislead people, but you're reaching if you think its going to end up scoring 2k and they'll be like "Oops sorry, we meant 3200 in Octal!"
Skylake-X is going to outperform threadripper. Unless something wildly strange happens, that's an easy conclusion to make. 8 Core Skylake-X is notably faster than the 1800X after all. All of the lakes are notably faster, core for core, than Ryzen.
You do have it right that TR offers quite a good value. It's going to draw enthusiasts and new enthusiasts to AMD's platform for sure; enthusiasts are used to paying up to $1k for good CPUs. But there are a lot of businesses who will be happy to pay for the higher performance if they need it, and Skylake-X will probably continue to sell well.
"8 Core Skylake-X is notably faster than the 1800X after all. "
That's a rather short-sighted way to look at it. Because you fail to account to account for one CRUCIALLY IMPORTANT thing - TDP budget.
Granted, both 1800x have 8 cores, and the same 3.6 GHz base clock, and intel has a slight IPC advantage, allowing the 7820X to score above the 1800x, but the advantage is the rather modest 6.7%.
But the 7820X required the rather significant 60 % of power to beat the 1800x by the measly 6.7%.
Granted, the HCC parts will be significantly lower clocked in order to get better performance per watt, but they will still be ultimately limited by TDP.
TR is quite literally two ryzens, there is only 200 MHz clock difference between ryzen and TR. And there is an entire GIGAHERTZ between the 8 core and 18 core skylake x.
So no, it is not as simple as "if the 8 core intel is faster than the 8 core zen, then the 16 core intel will be faster than the 18 core zen".
If that score intel are showing off is indeed true, then it is a product of the turbo, and understandable, as the CB test lasts for like several seconds, so it is not likely to run into any problems.
But prosumer workloads are rarely limited to bursts of several seconds. On the contrary, many of those systems will likely crunch data for hours. Which is where intel's lousy power and thermal performance will reveal itself.
And if anything, intel's HCC chips will actually be even less power efficient, because they have additional core circuitry.
Intel has had a long history of dirty tricks for when it cannot out-compete with performance, and they are back to it since zen arrived. Their goal is to discourage as many people as possible from buying TR now, as it will launch in just 2 days.
And for better or worse, intel's "superior" products won't be available for significantly longer than this, so it will take a while before they can be called on their BS by putting actual products to actual, practical work.
At 8 cores, intel has over 50% lower performance per watt, and even if we are extremely generous towards intel and assume that down-clocking will be able to win them a 30% improvement in that figure, it is still mathematically impossible for them to beat a 180W TR with a 165W SX.
It would take at the very least a 220 watt TDP for skylake X to beat 180 watts of zen. Which means that the number they claimed for the "stock" 16 core SX is actually achieved by means of OVERCLOCK that will MOST DEFINITELY not be possible to sustain for extended periods of time.
Yes, in the benchmarks that came out when Ryzen was launched, Skylake definitely outshined Ryzen in all but multi-threaded applications as single core IPC is much higher on Intel cores. However, when Ryzen was launched, there were some initial bugs that needed to be ironed out before Ryzen's performance could really come through. One such bug is RAM. Most of the initial reviews couldn't seem to get RAM working past 2133MHz in most cases. Sometimes, you'd get 2933MHz, but generally, RAM had to be down-clocked or much slower sticks needed to be used.
The architecture of Zen is such that RAM speed greatly affects performance, unlike in an Intel system. Since the Infinity Fabric is linked to RAM speeds, the advantage of using faster RAM in a Ryzen based system is unique in a remarkable way. Upwards of 40% performance increases can be achieved simply by using 3600MHz RAM over 2133MHz.
Outside of RAM is the fact that most software doesn't have Zen optimizations built in to them. Those will come out increasingly with time and as we've all seen, from a 1st-round version of an optimizationpatch, performance increases of greater than 30% can be had. Who knows how much additional performance can be squeezed out with further optimization patches.
We've also seen some additional tweaks that have been discovered with regards to RAM timings and various other features which, depending on whether or not they are turned on or off, combined, can add up to a 9-10% performance increase. Collectively, that's almost 80% performance that can be added to some of the original benchmark numbers.
That all being said, Ryzen clocked at 3.965GHz using DDR4-3600MHz, was able to outperform the i7-7700K clocked at 5GHz in many games that the 7700K outperformed it and many of which still did not have any form of Ryzen optimizations released.
Again, take note, a 7700K at 5GHz! And in games that the 7700K previously won, meaning games that don't necessarily utilize the additional cores that Ryzen offers.
I think that the Zen architecture is much better than most people give it credit for. Ryzen is a first edition of a brand new arch, and with that, there will obviously be some bugs, unforeseen issues and limitations, and various issues that may not be able to fully unleash the architectures capability, but with time, I feel that we'll all start to see just how good the Zen arch can actually be.
If 100% of your cores are running at 100% load 24-7-365 you likely did not properly size your server (yes there are tasks where this would be acceptable but, IMO, still not optimal).
Odd, that the 6154 (3.7 all-core turbo xeon) has a 205W TDP??? Yet, this is 165W for 18 cores @ 4.2 Turbo? More shenanigans with TDP (SKLX's TDPs have been "suggestions" or "aspirational" so far)? Or has something been turned off?
The Xeon Xeon 6154 is listed as having a 200W TDP with base of 3.0 Ghz and max turbo of 3.7Ghz. Intel also have a Xeon 6150 which is listed with a 165W TDP, 2.7Ghz base and max turbo of 3.7Ghz.
Looking at those two, it seems the reason the 6154 would has a higher TDP than the i9-7980XE is the higher base clock.
Again, odd choice, but you are correct, it appears they've chosen the 6150 as the donor line for this chip when 3.0 would have been a more suitable base block for a desktop chip and 200W entirely reasonable given the target market. I supposed MB VRM limits already in place could have made that choice.
Intel certainly won't be lowering that price. It may sell for lower eventually, but they are more than happy to wait until the next generation is out to adjust prices there rather than flinch in the current generation.
Intel will need to bring the prices down. As at 2000$ they are not competing against ThreadRipper, they are competing with a 32 Core/64 thread EPYC 7551P which is also 2000$. The only advantage Intel holds is in single thread. I wish Intel did something similar and released a single socket 28 core Xeon at around 2000$-2500$ range.
They aren't competing with either. It will match TR at double the cost, so half of the price/performance, and it will lose significantly to an equally priced Epyc.
They are doing it just so they can brag about having 2 more cores in HEDT. I bet they hope they won't sell many of those, they'd rather sell them for a 1500$ extra as xeons, which any professional will have to pay for the ECC support. I mean every professional, dumb enough to buy such a low value product now that there is finally a vastly superior alternative.
Yep, if you feel that badly the need to spend 2000$ on CPU, you will simply get TWO TR chips.
After all, pretty much everything that requires such massive CPU power supports network rendering - be that 3d rendering, video rendering or whatever. So you will be able to do twice the work in the same unit of time. You could also use one system to prep batches while the other renders.
Honestly, rich, silly fanboys are the only target market for intel's top end HEDT.
@Ian When you get to review the cpu, could you please post the 1-2 minute graphs of the cpu load across each core ( the one in the windows task manager) while benchmarking the games? That would give us a fair idea of the kind of cpu threads utilized by the modern day games like GTA V and FIFA 17 demo. I am yet to see any reviewer post these graphs, so it would be kind of you to include them.
The reason why you're not seeing those sorts of graphs is likely because the target market for high core count CPUs simply isn't PC gaming. The costs are too high for a significant number of people to purchase a processor with 18-cores and then play video games with it when a CPU with 4 higher clocked cores (possibly 8 in the next couple of years) will offer just as good or even better gaming performance -- better due to higher clock speeds attained by fewer cores.
Spending $2K on a Skylake-X alone, not even factoring in platform costs, just to steal cars and run away from the popo when a Sandy/Ivy Bridge Core i5 (or a Kaby Lake i3) will give you very good FPS if it has an adequate GPU borders on irrational e-peen waving. I mean really, an i5-3320m and a Quadro NVS 5200m in a 2013-era Dell Latitude e6430 laptop can run GTA V pretty well and the dual core mobile Ivy Bridge isn't the system's performance bottleneck.
For now, 4 cores would seem to be the sweet spot for gaming systems, but if you look at the chart above, the 6 core system is gimped to a lower turbo frequency than the 8 and 10 core CPU's in the X-class systems
Personally, I don't think I will need more than 4 or 6 cores but I wish the lowest core counts had better clocks than the high price / high core cpu's
Lower clock speeds for the 6 core CPU makes no sense
I hope Ice Lake has the best clock speeds on 4-6 core systems Higher PCIe Lane count on 4-6 core systems would be nice too
But other people would like a jack of all trades. I'd happily upgrade to 12 cores if there was a guarantee of a decent increase in single core performance over the 6 core chip I currently own. Many people here don't just workstation or game or video edit, they do everything.
So I suppose that single thread performance below that of an i7 at 4.3 Ghz is too just too little for "everything"?
Top performance really only matters for time staking tasks such as video editing. It doesn't make any difference if you play games at 150 or 200 FPS, it doesn't matter if a web page loads in 200 msec rather than 150 msec.
After a certain point, single thread performance simply doesn't matter, what matters is having enough, because tasks that require only a single thread are not of critical importance in terms of performance.
"After a certain point, single thread performance simply doesn't matter, what matters is having enough, because tasks that require only a single thread are not of critical importance in terms of performance."
That's a rather obtuse statement. If say 30% of my work load is in single threaded tasks, while 70% is in multithread, doubling the single threaded performance maters about as much as 1.5x-ing the number of cores.
And for those of us in high core count systems already, we're used to seeing these low base clock rates. The thing is, you're not often in the lowest clock rate, even with all cores active. I'm not sure what the standard with these, but my 2.0 GHz base clock machine (a pair of ivy bridge E5s) is often reporting more like 2.3+ with all cores active. I'd be interested to see if 2.6 GHz is really something closer to 3.0 when people actually get their hands on these processors.
No matter what you do, the base clocks of the 1950X will always be 3.4Ghz, that's why intel is totally lost in 2017, they don't have an arch to compete.
The turbo charts suggest that the 18-core will turbo all cores to 3.4 GHz, but that might exceed TDP in stress-testing situations - although someone building such a system better include a good cooler so they can exceed TDP and possibly even turbo higher then that.
If that excludes AVX, then those frequencies do not matter for prosumers. Every prosumer workload out there uses the vector units. The ALU is probably not more than 10% of the entire core performance, and is used mostly for program flow rather than actual data computations.
AVX/AVX2 doesn't really need that strong of a downlock anymore. AVX512 does, however, but it also offers quite a lot of throughput, so its usually still a net-gain.
I can only judge my 7900X with OC settings (which many motherboards even apply automatically as long as thermals permit), where AVX2 only needs a -2 offset, and AVX512 needs -7 to -9, depending on what turbo you use as a base.
Until these high-core-count CPUs are properly vetted, its impossible to say which typical clocks one can expect in the different workloads, but its definitely going to be above the "base" frequency at all times.
You can perfectly do all of that with base clocks of the Ryzen 1700X, TR 1950X has that same advantage with twice the cores. There goes your jack of all trades.
The base clock isn't all that relevant - what is relevant is the typical clocks its running at in typical workloads. "Base" is a sort of worst case situation, but no typical Intel CPU runs on their base clock much.
According to the graphs posted above, even the 18 core should turbo to 3.4GHz all core by default, which can be increased with decent cooling. And ST workloads where clock matters the most won't run all the cores and clock higher as well.
I work for a company building systems around Xeon chips. Our 16-core, 2.6 GHz base clock Xeon run at an average of 3.1 GHz with all cores active. Intel has done a good job of boosting the clock rate for workloads which allow it, in the presence of adequate cooling.
Your mileage may vary: Heavy FPU usage in an HPC environment might not run so quickly, for instance. (Our workload is primarily integer computation and DMA, with short bursts of vectorized code.)
Ian, please review the overclocking potential and temps of these chips, particularly the 18 core ones, when they come out. There has been a lot of discussion about the the negative impact of Intel's decision to use TIM on these chips, and even how lower end MBs may struggle to deliver the required current these chips draw.
Remember boys, the all core turbo only sustains for some seconds, when the cpu gets maxed with warm temps the cpu will be back to base clocks (intel all core turbo is basically create to look good on benchmarks, which don't take much time), while on the other side the 1950X already has a base clock of 3.4Ghz, no mater how you tax it, that's the minimum frequency you will get.
Remember that the turbo is not a binary on/off toggle, and how long it can be sustained depends entirely on your cooling. It doesn't go Turbo -> Base in one step, it slowly backs down if it has to.
Any high-end cooler will be able to sustain the rated all-core turbos, and with really good cooling you can go beyond that.
That's assuming you can get the heat out of the die, which is not a given when toothpaste TIM is involved. You're also assuming that you can get adequate power to the CPU to provide for those frequencies, a problem which gets worse at higher temperatures.
Anyone else getting the distinct feeling that this is a worse version of the 2006-ish core wars? How many $2000 and 165W Processors could Intel possibly sell in the consumer space? I feel like half of anything sold will be to folks that do not even need that much computing power. What application, in the consumer space, is there for this much silicon? Paired with 128GB+ of RAM and a mega raid, and we're talking about the ballpark of a hefty down payment toward a V6 Sports Car. I love the Anandtech Coverage, but am perhaps jaded about the meaning of it all- does it mean I'll be getting a hoverboard anytime soon? I think this is a surge before the deathknell of microcomputing companies, when they hit the 3-4nm wall and the global economy collapses. With that in mind, add in a solar and wind array with night time storage to the cost of ownership. Nope- just go with the car.
"Consumer" is perhaps slightly misleading. I'm sure some ordinary consumers might buy this out of ignorance of what its really for, though.
The real audience are enthusiasts as well as professionals with actual workflows that use this. Video rendering and encoding, 3D rendering, complex math computations of all sorts - workstation tasks.
Clearly its a much smaller market then the mainstream consumer market, but it does exist.
The same argument really applies to ThreadRipper (albeit at the lower cost, its not that "bad"). "Consumers" don't need more then 8 cores, and they likely don't even need 8. Not yet, anyway.
If the HEDT parts come off the Xeon line, they will provide for ECC.
Support will be up to individual motherboard providers. ECC is a mainstay of the Xeon line and to the best of my knowledge no Xeon is available which does not provide for ECC memory usage.
As the circuitry that provides for ECC is contained entirely within the IMC, chipset selection is rather inconsequential as well. X79 runs ECC just like C602; likewise, X99 runs with ECC as does C612, so long as you drop a Xeon in there (and not an i7 in the case of the consumer chipsets).
I would bet a juicy steak that the HCC HEDT Skylake-X parts work with Unbuffered/Unregistered ECC memory quite nicely.
It can actually be worse: As reported in c't 17/2017 p.98, when you run the Linpack benchmark on a Core i9-7900X on the Gigabyte X299 Aorus Gaming 3 or MSI X299 SLI Plus, it consumes a lot of power (259W CPU package power on the MSI board) all the time, but delivers only half the performance compared to the same CPU on the Asrock X299 Taichi or Asus Prime X299-A. The explanation is that the BIOSes on the Gigabyte and MSI boards let the CPU run at a high clock, and it gets so hot that thermal throttling sets in. The other boards apparently have a saner Turbo policy by default (you can also get that on the Gigabyte and MSI boards by manually adjusting the power limit with the Intel XTU.
Anyone spending this kind of money on a CPU should probably take some care to properly set it up - you don't even need XTU for that, the relevant settings are available in the BIOS as well.
Defaults greatly differ between boards, that is true.
people who are using these kind of cpus are often artist and other professionals who are not computer nerds. a computeer should run fine with default settings. the need to "properly setup the bios" is something 80% of the customers will probably never do. the world is bigger than a few nerd websites.
Well those people are also unlikely to self-assemble a PC, aren't they? Whoever does the assembly of such an expensive system should know how to set it up.
When you buy a pre-assembled system, I would expect such a setup to have been performed indeed.
I find it interesting that Intel is making this move. I mean, it makes sense for them too.
On the other hand, looking at the charts, the $999 part for Intel gets you 10 cores, 20 threads, 3.3GHz base, 4.3GHz turbo and 40 PCI-e lanes. The $999 part for AMD gets you 60% more cores and threads (16/32), 3.4GHz base, 4GHz turbo and 60 PCI-e lanes.
I am sure the Intel part will be better in single thread, but it looks like the AMD part likely promises to crush the same price Intel part in most all multithreaded workloads.
Which is nice. For the last few years the number of workloads where you could say Intel was the price-performance leader were very, very, very tiny. Now it looks like AMD really does promise to be the price-performance leader, and in some cases outright performance leader in a number of categories.
If I had the spare change for an upgrade I'd definitely be looking at the "low end" Ryzen 8/16 part as an upgrade of my Ivy Bridge quad core desktop. Then again, maybe next year AMD will have something even better in the offing.
These CPUs and the motherboard platform is a joke. I'm not paying 2K when i can get basically the same performance for half the price, which i will be doing.
Incredible how long we've been stuck with 4 cores processors and out of sudden they are able to throw 18 core at us in a few month span!!! Shows how bad a monopoly can be. Whatever it is Intel can come up with, i'll buy AMD stuff for my next rig just to help them STAY ALIVE for a while longer and continue to bring competition to the market. The desktop market has been boring for a decade! I was getting sick to come to Anandtech just to read crappy phone reviews ;)
Well said. And AMD has made it compelling to buy their processors, finally. One will really need to stretch logic and reality to continue with a near-monopoly situation like the past at least FIVE years.
Pretty nice. I wish in a way that I had waited to buy my workstation build for this - but it's always going to be the case that something faster comes out later.
Is the 4.2GHz number the all-cores turbo boost? That's pretty high for an 18-core part. Especially since the 2687Wv4 maxes out at 3/3.1.
Unfortunately for Intel, his greed really shows now. Although core still has about 5-10% more IPC compared to Ryzen, the power consumption per core is about 5-15% higher and with 18 cores this really shows. They had a very competitive tick-tock strategy and now after more than three years they are still stuck in 14nm. If they were smarter and created by now just one fab on 8 or 10 nm for the CPU's with many cores things were more simple for them today. In 8 nm, skylake x would had allowed 18 cores on 3.2-3.6Ghz, and not 2.6 as they are doing now. So they save 3-4 billions dollars not building a 8nm fab but will lose more than this when enthusiast marked will side with AMD. Please be smarter in future Intel, Samsung and TSMC already have 8 nm FABS while you...
Intel's 18c 2.6GHz base clock with a 165 watt TDP vs AMD's 3.4GHz base clock with a 180 watt TDP doesn't look like much of a win, especially with less PCI-E lanes. Especially with new test done comparing the Ryzen R5 1600 vs the i7-7800x and both at stock the Intel chip clearly won, but when both were overclocked (AMD 4.0GHz, Intel 4.7GHz) both running 3200MHz ram they were nearly identical within 1%. Which supports people saying that AMD scales better and has better multi-threading (SMT vs HT).
I'm VERY curious to see the results for Ryzen running on quad channel 3200 memory.
I got the FX-8320 the month it came out for about $180 together with a great motherboard, so it was a no brainer.. kinda lol. Had 4 sticks of 1333MHz ram laying around totaling 8GB and quickly upgraded to 16GB 2400MHz CAS-11 (running at 2133MHz CAS-9) and did some testing. Set it to 1333MHz and matched the old ram, then OCed it to find about a 15%+ difference. I did the same thing with a Q9550 i had before and only found about a 5% gain from 4GB 800MHz CAS-5 to 8GB 1150MHz CAS 5 and made sure that there wasn't any page filing.
Point is! lol, I'll wait until i see a RAM speed wall because i don't want to buy some ram then find out 4000MHz is the sweet spot then feel neutered lol. Also, it looks like Ryzen is massively affected by RAM because of CCX interconnecting the cores. It also seems that Ryzen scales very well in frequency.
In case nobody has noticed (very slim chance but there's a lot of comments), you have Threadripper listed as having only sixty PCI-Express 3.0 lanes when your own article about it from May says sixty-four which is the correct number. You even have the slide on your site. You might want to correct that because some people might get the wrong idea. I'm sure it's just a typo or something but you know how these things can get. http://images.anandtech.com/doci/11482/threadrippe...
7960X and 7980XE make no sense at all. Base clock decrease combined with less than linear scalability of even multi-threaded applications combined will give the same real life performance as 7940X. 3.1 * 14= 43.4 2.8 * 16= 44.8 2.6 * 18= 46.8
Non-linear scalability, 3rd level cache contention, memory bus contention, TDP limit will do the rest. And lower ST performance will seal the deal. Don't waste your money.
Guys, base clock rate is really just a worst-case scenario. You see the Turbo boosts by number of active cores, right? For the most part, with good cooling that is, those will be achieved and sustained. Or at the very least its going to be closer to that than to the base clocks. So...
3.8 * 14 = 53.2 3.6 * 16 = 57.6 3.4 * 18 = 61.2
If you already have intel systems, download the Intel power gadget, open up activity monitor and run something that maxes all your cores. You'll probably see that so long as temps aren't going crazy, you'll be running above your base clock rate. Even on my MacBook Pro, I can do this with temps around 98C and clock rates sustaining about .4GHz over my base clock for as long as I cared to let the thing run (about 5 minutes, and maybe 2 minutes in it seem to reach a steady state). It state that the 3-4 core turbo on my 4980HQ is 3.6, but I'm really only able to sustain 3.2. This is a laptop though...
Back when AMD came out with the original (Sledgehammer) Opterons there was one feature that didn't show up on any spec sheet, but was known by the people in the trenches who wrote server code. (Or in my case, put in compiler optimizations used by the HPC crowd.) Opterons were predictable. If a given job took thirty minutes to run today, it would be within a minute of that (both wall clock and CPU times) tomorrow. I never had a case where the Opteron code path was not pretty much common to all chips. Yes, occasionally you did have to put in "ifdefs" to distinguish between single chip and multiple chip configurations, but no with and without Hyperthreading, with and without SSE, or the latest feature of the month that marketing said we just HAD to support.
Now with Zen, AMD is getting back to that part of their roots. Zen code runs on all Zen chips. Yes, I am writing code that determines how many chips, cores, and threads are present, but that's about it.
Oh, while I am grousing in Intel's general direction, why do they need three subsets of AVX-512. Yes, that is subset, no current processor supports all of AVX-512, and no future processor is expected to do so....
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
100 Comments
Back to Article
ImSpartacus - Monday, August 7, 2017 - link
It's cool that Anandtech is marking the actual die used.Do we know the physical cores present on the LCC, HCC and XCC die?
I know that it's not always straightforward since some die might always disable a core or two for yield purposes.
Ian Cutress - Monday, August 7, 2017 - link
10, 18 and 28. They're 3x4, 5x4 and 5x6 designs, with two segments earmarked for the DRAM controllers. Consumer parts being LCC/HCC is easy - the change in the Xeon line naming has made it harder to determine which is which on the enterprise side.ImSpartacus - Monday, August 7, 2017 - link
It's interesting how two "segments" get occupied by memory controllers. That explains the "odd" core counts.Thanks for sharing that.
PixyMisa - Wednesday, August 9, 2017 - link
It's quite visible in die photos - they have a neat grid of cores, then two odd things that look completely different.marc1000 - Wednesday, August 9, 2017 - link
I was wondering here... AMD finally wakes up, gives 16 cores to consumer, then Intel responds by giving 18. but that TR4 socket is big enough to fit the 32core design from AMD.So I strongly suspect that AMD will launch new variants of Threadriper based on the recent Intel announcements.
5*4 (20 cores), 6*4 (24), 7*4 (28) seems like a given, as it would still save quite a lot of dies that are not able to go to Epyc. in fact this seems very clever and I have to give credit to AMD again: single die to all segments. not the best at the top and not the best at the bottom. but great value in all mid-range segments.
Avro Arrow - Thursday, August 10, 2017 - link
I'm willing to bet that the TR-1960X (20-core) and TR-1970X (24-core) are already made, packaged and sitting in an AMD warehouse somewhere ready to ship. I would expect AMD to announce them two, maybe three days before the i9-7980XE's launch date. I don't expect to see a 28 or 32-core because I think that they need to have at least some distance from EPYC. Since Intel has decided not to make Skylake-X support ECC RAM, they have that divide between Skylake-X and Xeon. On the other hand, AMD, being AMD, has a significantly more attractive offering that DOES support ECC RAM so they have to try a bit harder to keep that divide between Threadripper and EPYC. It won't matter though because the i9-7980XE would be no match for the (theoretically-named) TR-1970X's extra six cores. Since Intel is still using the conventional monolithic manufacturing process, there's no chance in hell that they'll be able to react quickly enough if AMD pulls a stunt like that. I would personally die laughing if that happened.marc1000 - Friday, August 11, 2017 - link
fully agreed. 24 core to fight against 18 core would be a great marketing and technical feat, while still keeping a safe distance from epyc (but with the possibility to go crazy and reach 28 or 32).let's see what happens, those are exciting time - CPU wars again!
T1beriu - Monday, August 7, 2017 - link
10, 18, 28 cores on LCC, HCC and XCC die.jabbadap - Monday, August 7, 2017 - link
"The top Core i9-7980XE SKU will have a base frequency of 2.6 GHz but a turbo of 4.2 GHz, and a Favored Core of 4.4 GHz. I suspect the turbo will be limited to 2-4 cores of load, however Intel has not listed the ‘all-core turbo’ frequencies which are often above the base frequencies, nor the AVX frequencies here."You don't have to suspect anything it is given information from intel all ready.
https://simplecore.intel.com/newsroom/wp-content/u...
Both are for 2 core loads, lower turbo boost 2.0 and higher turbo boost 3.0
Just wondering how high those all core freqs will be, Amds 16 core TB 1950x has all core turbo of 3.6GHz. And by looking those intel xeons that could be in same ball park.
ddriver - Monday, August 7, 2017 - link
High turbo clock announcements are to smear on the reality of the rather lousy 2.6 base clock. Because you know what, the target market won't run single threaded workloads for short amounts of time, such machines are being worked 100% of the cores at 100% of load for many hours a day.Basically, even this revamped xeon, sold 1500$ cheaper just to save face will not be able to match TR. Well, at least it has more cores, that's gotta count for something.
I mean the 7980XE will have to squeeze out at least 30% more performance than then 18 core 2699 v3 to get a definite edge over TR. It has 13% of clock advantage, which will not translate in less than 13% performance increase, and like 10% IPC advantage, which would put it on par or a tad lower than the top of the line TR, I am taking CB performance here.
Oh, and it costs twice as much as the TR... Oh and no ECC.
ddriver - Monday, August 7, 2017 - link
I meant "which will not translate in MORE than 13% performance increase"Drumsticks - Tuesday, August 8, 2017 - link
Just saying, Intel mentioned that the stock 16 core scores 3200 on Cinebench. That's about 7%+ better than TR 16C. Maybe overclocking Threadripper could put it even. The 18 core is going to be faster than "on par or slightly below" TR.ddriver - Tuesday, August 8, 2017 - link
Where did they mention it? In a dream of yours? Got a source? Because this sounds quite implausible.mkaibear - Tuesday, August 8, 2017 - link
Really ddriver? Can't be bothered to even google for "7960X cinebench"?https://twitter.com/intelnews/status/8946200915468...
ddriver - Tuesday, August 8, 2017 - link
Dunno honestly, it is an intel tweet that directs to an engaged article that claims that intel claimed that score.But even if true, TR would still be a WHOOPING 85% better value. So... kind of a moot point.
tamalero - Tuesday, August 8, 2017 - link
Sounds like an echo chamber technique to try way too hard to look real.tamalero - Tuesday, August 8, 2017 - link
Nevermind, the whole article looks like a paid PR announcement by Intel.silverblue - Tuesday, August 8, 2017 - link
The 7980X appears to have lower clocks than the 7960X (2.6 vs. 2.8). They all have a 4.4GHz max turbo speed. IPC, at least as concerns floating point work, appears to be with Intel, but power consumption is going to be very interesting.Drumsticks - Tuesday, August 8, 2017 - link
Hahahahaha okay.Someone's already linked the tweet. You really think @intelnews is going to tweet something that will EASILY be disproved wildly false? Sure, I'll grant, it's easy to mislead people, but you're reaching if you think its going to end up scoring 2k and they'll be like "Oops sorry, we meant 3200 in Octal!"
Skylake-X is going to outperform threadripper. Unless something wildly strange happens, that's an easy conclusion to make. 8 Core Skylake-X is notably faster than the 1800X after all. All of the lakes are notably faster, core for core, than Ryzen.
You do have it right that TR offers quite a good value. It's going to draw enthusiasts and new enthusiasts to AMD's platform for sure; enthusiasts are used to paying up to $1k for good CPUs. But there are a lot of businesses who will be happy to pay for the higher performance if they need it, and Skylake-X will probably continue to sell well.
ddriver - Tuesday, August 8, 2017 - link
"8 Core Skylake-X is notably faster than the 1800X after all. "That's a rather short-sighted way to look at it. Because you fail to account to account for one CRUCIALLY IMPORTANT thing - TDP budget.
Granted, both 1800x have 8 cores, and the same 3.6 GHz base clock, and intel has a slight IPC advantage, allowing the 7820X to score above the 1800x, but the advantage is the rather modest 6.7%.
But the 7820X required the rather significant 60 % of power to beat the 1800x by the measly 6.7%.
Granted, the HCC parts will be significantly lower clocked in order to get better performance per watt, but they will still be ultimately limited by TDP.
TR is quite literally two ryzens, there is only 200 MHz clock difference between ryzen and TR. And there is an entire GIGAHERTZ between the 8 core and 18 core skylake x.
So no, it is not as simple as "if the 8 core intel is faster than the 8 core zen, then the 16 core intel will be faster than the 18 core zen".
If that score intel are showing off is indeed true, then it is a product of the turbo, and understandable, as the CB test lasts for like several seconds, so it is not likely to run into any problems.
But prosumer workloads are rarely limited to bursts of several seconds. On the contrary, many of those systems will likely crunch data for hours. Which is where intel's lousy power and thermal performance will reveal itself.
And if anything, intel's HCC chips will actually be even less power efficient, because they have additional core circuitry.
Intel has had a long history of dirty tricks for when it cannot out-compete with performance, and they are back to it since zen arrived. Their goal is to discourage as many people as possible from buying TR now, as it will launch in just 2 days.
And for better or worse, intel's "superior" products won't be available for significantly longer than this, so it will take a while before they can be called on their BS by putting actual products to actual, practical work.
At 8 cores, intel has over 50% lower performance per watt, and even if we are extremely generous towards intel and assume that down-clocking will be able to win them a 30% improvement in that figure, it is still mathematically impossible for them to beat a 180W TR with a 165W SX.
It would take at the very least a 220 watt TDP for skylake X to beat 180 watts of zen. Which means that the number they claimed for the "stock" 16 core SX is actually achieved by means of OVERCLOCK that will MOST DEFINITELY not be possible to sustain for extended periods of time.
ddriver - Tuesday, August 8, 2017 - link
"But the 7820X required the rather significant 60 % of EXTRA POWER power to beat the 1800x by the measly 6.7%."MLSCrow - Wednesday, August 9, 2017 - link
Yes, in the benchmarks that came out when Ryzen was launched, Skylake definitely outshined Ryzen in all but multi-threaded applications as single core IPC is much higher on Intel cores. However, when Ryzen was launched, there were some initial bugs that needed to be ironed out before Ryzen's performance could really come through. One such bug is RAM. Most of the initial reviews couldn't seem to get RAM working past 2133MHz in most cases. Sometimes, you'd get 2933MHz, but generally, RAM had to be down-clocked or much slower sticks needed to be used.The architecture of Zen is such that RAM speed greatly affects performance, unlike in an Intel system. Since the Infinity Fabric is linked to RAM speeds, the advantage of using faster RAM in a Ryzen based system is unique in a remarkable way. Upwards of 40% performance increases can be achieved simply by using 3600MHz RAM over 2133MHz.
Outside of RAM is the fact that most software doesn't have Zen optimizations built in to them. Those will come out increasingly with time and as we've all seen, from a 1st-round version of an optimizationpatch, performance increases of greater than 30% can be had. Who knows how much additional performance can be squeezed out with further optimization patches.
We've also seen some additional tweaks that have been discovered with regards to RAM timings and various other features which, depending on whether or not they are turned on or off, combined, can add up to a 9-10% performance increase. Collectively, that's almost 80% performance that can be added to some of the original benchmark numbers.
That all being said, Ryzen clocked at 3.965GHz using DDR4-3600MHz, was able to outperform the i7-7700K clocked at 5GHz in many games that the 7700K outperformed it and many of which still did not have any form of Ryzen optimizations released.
Again, take note, a 7700K at 5GHz! And in games that the 7700K previously won, meaning games that don't necessarily utilize the additional cores that Ryzen offers.
https://www.youtube.com/watch?v=RZS2XHcQdqA
I think that the Zen architecture is much better than most people give it credit for. Ryzen is a first edition of a brand new arch, and with that, there will obviously be some bugs, unforeseen issues and limitations, and various issues that may not be able to fully unleash the architectures capability, but with time, I feel that we'll all start to see just how good the Zen arch can actually be.
Total Meltdowner - Thursday, August 10, 2017 - link
u a faggot nig g3rIcehawk - Tuesday, August 8, 2017 - link
If 100% of your cores are running at 100% load 24-7-365 you likely did not properly size your server (yes there are tasks where this would be acceptable but, IMO, still not optimal).Total Meltdowner - Thursday, August 10, 2017 - link
This is a very good point.twtech - Friday, August 11, 2017 - link
With sufficient cooling, you can achieve the all-core turbo frequency all of the time. That's why it's important to find out what it is.dgingeri - Monday, August 7, 2017 - link
I'd be willing to bet they release a WS version of Xeon using much the same specs as here, clock locked, and socket 2066.cekim - Monday, August 7, 2017 - link
Odd, that the 6154 (3.7 all-core turbo xeon) has a 205W TDP??? Yet, this is 165W for 18 cores @ 4.2 Turbo? More shenanigans with TDP (SKLX's TDPs have been "suggestions" or "aspirational" so far)? Or has something been turned off?ilt24 - Monday, August 7, 2017 - link
The Xeon Xeon 6154 is listed as having a 200W TDP with base of 3.0 Ghz and max turbo of 3.7Ghz. Intel also have a Xeon 6150 which is listed with a 165W TDP, 2.7Ghz base and max turbo of 3.7Ghz.Looking at those two, it seems the reason the 6154 would has a higher TDP than the i9-7980XE is the higher base clock.
cekim - Monday, August 7, 2017 - link
Again, odd choice, but you are correct, it appears they've chosen the 6150 as the donor line for this chip when 3.0 would have been a more suitable base block for a desktop chip and 200W entirely reasonable given the target market. I supposed MB VRM limits already in place could have made that choice.colonelclaw - Monday, August 7, 2017 - link
I suspect the 18-core will either launch at a lower price, or come down shortly after launch.Or stay at $2k and bomb.
cekim - Monday, August 7, 2017 - link
Intel certainly won't be lowering that price. It may sell for lower eventually, but they are more than happy to wait until the next generation is out to adjust prices there rather than flinch in the current generation.sharath.naik - Monday, August 7, 2017 - link
Intel will need to bring the prices down. As at 2000$ they are not competing against ThreadRipper, they are competing with a 32 Core/64 thread EPYC 7551P which is also 2000$. The only advantage Intel holds is in single thread. I wish Intel did something similar and released a single socket 28 core Xeon at around 2000$-2500$ range.ddriver - Monday, August 7, 2017 - link
They aren't competing with either. It will match TR at double the cost, so half of the price/performance, and it will lose significantly to an equally priced Epyc.They are doing it just so they can brag about having 2 more cores in HEDT. I bet they hope they won't sell many of those, they'd rather sell them for a 1500$ extra as xeons, which any professional will have to pay for the ECC support. I mean every professional, dumb enough to buy such a low value product now that there is finally a vastly superior alternative.
Lolimaster - Monday, August 7, 2017 - link
But any professional will get TR that actually supports ECC :Dddriver - Tuesday, August 8, 2017 - link
Yep, if you feel that badly the need to spend 2000$ on CPU, you will simply get TWO TR chips.After all, pretty much everything that requires such massive CPU power supports network rendering - be that 3d rendering, video rendering or whatever. So you will be able to do twice the work in the same unit of time. You could also use one system to prep batches while the other renders.
Honestly, rich, silly fanboys are the only target market for intel's top end HEDT.
prateekprakash - Monday, August 7, 2017 - link
@Ian When you get to review the cpu, could you please post the 1-2 minute graphs of the cpu load across each core ( the one in the windows task manager) while benchmarking the games?That would give us a fair idea of the kind of cpu threads utilized by the modern day games like GTA V and FIFA 17 demo.
I am yet to see any reviewer post these graphs, so it would be kind of you to include them.
BrokenCrayons - Monday, August 7, 2017 - link
The reason why you're not seeing those sorts of graphs is likely because the target market for high core count CPUs simply isn't PC gaming. The costs are too high for a significant number of people to purchase a processor with 18-cores and then play video games with it when a CPU with 4 higher clocked cores (possibly 8 in the next couple of years) will offer just as good or even better gaming performance -- better due to higher clock speeds attained by fewer cores.Spending $2K on a Skylake-X alone, not even factoring in platform costs, just to steal cars and run away from the popo when a Sandy/Ivy Bridge Core i5 (or a Kaby Lake i3) will give you very good FPS if it has an adequate GPU borders on irrational e-peen waving. I mean really, an i5-3320m and a Quadro NVS 5200m in a 2013-era Dell Latitude e6430 laptop can run GTA V pretty well and the dual core mobile Ivy Bridge isn't the system's performance bottleneck.
Glock24 - Monday, August 7, 2017 - link
180W TDP.... maybe per core?Glock24 - Monday, August 7, 2017 - link
I meant 165W, but still...Bullwinkle J Moose - Monday, August 7, 2017 - link
4 cores or 6 for gaming >https://www.techspot.com/review/1445-core-i7-7800x...
For now, 4 cores would seem to be the sweet spot for gaming systems, but if you look at the chart above, the 6 core system is gimped to a lower turbo frequency than the 8 and 10 core CPU's in the X-class systems
Personally, I don't think I will need more than 4 or 6 cores but I wish the lowest core counts had better clocks than the high price / high core cpu's
Lower clock speeds for the 6 core CPU makes no sense
I hope Ice Lake has the best clock speeds on 4-6 core systems
Higher PCIe Lane count on 4-6 core systems would be nice too
Gothmoth - Monday, August 7, 2017 - link
when i buy a 16 core system i don´t buy it for single core performance. ROTFLthe base clock of 2.6 GHz is a joke. and no professional worth a dime will be confused by the turbo clocks.
this is just a intel CPU for bragging rights...
HomeworldFound - Monday, August 7, 2017 - link
But other people would like a jack of all trades. I'd happily upgrade to 12 cores if there was a guarantee of a decent increase in single core performance over the 6 core chip I currently own. Many people here don't just workstation or game or video edit, they do everything.ddriver - Monday, August 7, 2017 - link
So I suppose that single thread performance below that of an i7 at 4.3 Ghz is too just too little for "everything"?Top performance really only matters for time staking tasks such as video editing. It doesn't make any difference if you play games at 150 or 200 FPS, it doesn't matter if a web page loads in 200 msec rather than 150 msec.
After a certain point, single thread performance simply doesn't matter, what matters is having enough, because tasks that require only a single thread are not of critical importance in terms of performance.
IanHagen - Monday, August 7, 2017 - link
There are some niche areas in which more single threaded performance always matters, but even so... Someone pay this man a pint!wallysb01 - Monday, August 7, 2017 - link
"After a certain point, single thread performance simply doesn't matter, what matters is having enough, because tasks that require only a single thread are not of critical importance in terms of performance."That's a rather obtuse statement. If say 30% of my work load is in single threaded tasks, while 70% is in multithread, doubling the single threaded performance maters about as much as 1.5x-ing the number of cores.
And for those of us in high core count systems already, we're used to seeing these low base clock rates. The thing is, you're not often in the lowest clock rate, even with all cores active. I'm not sure what the standard with these, but my 2.0 GHz base clock machine (a pair of ivy bridge E5s) is often reporting more like 2.3+ with all cores active. I'd be interested to see if 2.6 GHz is really something closer to 3.0 when people actually get their hands on these processors.
Lolimaster - Monday, August 7, 2017 - link
No matter what you do, the base clocks of the 1950X will always be 3.4Ghz, that's why intel is totally lost in 2017, they don't have an arch to compete.Lolimaster - Monday, August 7, 2017 - link
With intel all thise year, buying a high core count xeon meant having way sub 3Ghz cpu's, not anymore.nevcairiel - Tuesday, August 8, 2017 - link
The turbo charts suggest that the 18-core will turbo all cores to 3.4 GHz, but that might exceed TDP in stress-testing situations - although someone building such a system better include a good cooler so they can exceed TDP and possibly even turbo higher then that.ddriver - Tuesday, August 8, 2017 - link
If that excludes AVX, then those frequencies do not matter for prosumers. Every prosumer workload out there uses the vector units. The ALU is probably not more than 10% of the entire core performance, and is used mostly for program flow rather than actual data computations.nevcairiel - Tuesday, August 8, 2017 - link
AVX/AVX2 doesn't really need that strong of a downlock anymore. AVX512 does, however, but it also offers quite a lot of throughput, so its usually still a net-gain.I can only judge my 7900X with OC settings (which many motherboards even apply automatically as long as thermals permit), where AVX2 only needs a -2 offset, and AVX512 needs -7 to -9, depending on what turbo you use as a base.
Until these high-core-count CPUs are properly vetted, its impossible to say which typical clocks one can expect in the different workloads, but its definitely going to be above the "base" frequency at all times.
Lolimaster - Monday, August 7, 2017 - link
You can perfectly do all of that with base clocks of the Ryzen 1700X, TR 1950X has that same advantage with twice the cores. There goes your jack of all trades.drothgery - Monday, August 7, 2017 - link
None of Intel's current 16-core (or greater) offerings have a base clock > 2.6 GHz. In fact, nothing with over 12 cores does.nevcairiel - Tuesday, August 8, 2017 - link
The base clock isn't all that relevant - what is relevant is the typical clocks its running at in typical workloads. "Base" is a sort of worst case situation, but no typical Intel CPU runs on their base clock much.According to the graphs posted above, even the 18 core should turbo to 3.4GHz all core by default, which can be increased with decent cooling. And ST workloads where clock matters the most won't run all the cores and clock higher as well.
nevcairiel - Tuesday, August 8, 2017 - link
.. also, you are wrong. The 16-core itself even has a base of 2.8 GHz and the 13 core even 3.1 GHz - only the 18 core has a base of 2.6nevcairiel - Tuesday, August 8, 2017 - link
*14 core. Edit button where are thou.drothgery - Tuesday, August 8, 2017 - link
Eh, I looked at the latest-version Xeon family pages at ark.intel.com when I posted that comment.Ian Cutress - Wednesday, August 9, 2017 - link
Check our Skylake-SP vs EPYC review for a full breakdown of all the Sky-SP CPUsGothmoth - Tuesday, August 8, 2017 - link
lol..... base clock is most likely what these CPU´s will run at... when used by people who actually need them.so it´s most relevant!!
my systems run at full load 90% of the time.
i don´t buy a 16+ core cpu to have 80% of the cores idling.
sorry but you really have no clue about what you are talking here..
abr2 - Tuesday, August 8, 2017 - link
I work for a company building systems around Xeon chips. Our 16-core, 2.6 GHz base clock Xeon run at an average of 3.1 GHz with all cores active. Intel has done a good job of boosting the clock rate for workloads which allow it, in the presence of adequate cooling.Your mileage may vary: Heavy FPU usage in an HPC environment might not run so quickly, for instance. (Our workload is primarily integer computation and DMA, with short bursts of vectorized code.)
palladium - Monday, August 7, 2017 - link
Ian, please review the overclocking potential and temps of these chips, particularly the 18 core ones, when they come out. There has been a lot of discussion about the the negative impact of Intel's decision to use TIM on these chips, and even how lower end MBs may struggle to deliver the required current these chips draw.Lolimaster - Monday, August 7, 2017 - link
Remember boys, the all core turbo only sustains for some seconds, when the cpu gets maxed with warm temps the cpu will be back to base clocks (intel all core turbo is basically create to look good on benchmarks, which don't take much time), while on the other side the 1950X already has a base clock of 3.4Ghz, no mater how you tax it, that's the minimum frequency you will get.nevcairiel - Tuesday, August 8, 2017 - link
Remember that the turbo is not a binary on/off toggle, and how long it can be sustained depends entirely on your cooling. It doesn't go Turbo -> Base in one step, it slowly backs down if it has to.Any high-end cooler will be able to sustain the rated all-core turbos, and with really good cooling you can go beyond that.
Spunjji - Tuesday, August 8, 2017 - link
That's assuming you can get the heat out of the die, which is not a given when toothpaste TIM is involved. You're also assuming that you can get adequate power to the CPU to provide for those frequencies, a problem which gets worse at higher temperatures.Comdrpopnfresh - Monday, August 7, 2017 - link
Anyone else getting the distinct feeling that this is a worse version of the 2006-ish core wars? How many $2000 and 165W Processors could Intel possibly sell in the consumer space? I feel like half of anything sold will be to folks that do not even need that much computing power. What application, in the consumer space, is there for this much silicon? Paired with 128GB+ of RAM and a mega raid, and we're talking about the ballpark of a hefty down payment toward a V6 Sports Car. I love the Anandtech Coverage, but am perhaps jaded about the meaning of it all- does it mean I'll be getting a hoverboard anytime soon? I think this is a surge before the deathknell of microcomputing companies, when they hit the 3-4nm wall and the global economy collapses. With that in mind, add in a solar and wind array with night time storage to the cost of ownership. Nope- just go with the car.nevcairiel - Tuesday, August 8, 2017 - link
"Consumer" is perhaps slightly misleading. I'm sure some ordinary consumers might buy this out of ignorance of what its really for, though.The real audience are enthusiasts as well as professionals with actual workflows that use this. Video rendering and encoding, 3D rendering, complex math computations of all sorts - workstation tasks.
Clearly its a much smaller market then the mainstream consumer market, but it does exist.
The same argument really applies to ThreadRipper (albeit at the lower cost, its not that "bad"). "Consumers" don't need more then 8 cores, and they likely don't even need 8. Not yet, anyway.
tamalero - Tuesday, August 8, 2017 - link
Agree, This multicore war is gold for those who are now entering the 4k or 8k production of video or rendering.jabber - Tuesday, August 8, 2017 - link
A good proportion will be sold to those that just do benchmarks all day on their parents electric bill.jwcalla - Monday, August 7, 2017 - link
For $2,000 it better support ECC RAM.Spunjji - Tuesday, August 8, 2017 - link
For $2000? It won't.kjboughton - Wednesday, August 9, 2017 - link
If the HEDT parts come off the Xeon line, they will provide for ECC.Support will be up to individual motherboard providers. ECC is a mainstay of the Xeon line and to the best of my knowledge no Xeon is available which does not provide for ECC memory usage.
As the circuitry that provides for ECC is contained entirely within the IMC, chipset selection is rather inconsequential as well. X79 runs ECC just like C602; likewise, X99 runs with ECC as does C612, so long as you drop a Xeon in there (and not an i7 in the case of the consumer chipsets).
I would bet a juicy steak that the HCC HEDT Skylake-X parts work with Unbuffered/Unregistered ECC memory quite nicely.
AntonErtl - Tuesday, August 8, 2017 - link
It can actually be worse: As reported in c't 17/2017 p.98, when you run the Linpack benchmark on a Core i9-7900X on the Gigabyte X299 Aorus Gaming 3 or MSI X299 SLI Plus, it consumes a lot of power (259W CPU package power on the MSI board) all the time, but delivers only half the performance compared to the same CPU on the Asrock X299 Taichi or Asus Prime X299-A. The explanation is that the BIOSes on the Gigabyte and MSI boards let the CPU run at a high clock, and it gets so hot that thermal throttling sets in. The other boards apparently have a saner Turbo policy by default (you can also get that on the Gigabyte and MSI boards by manually adjusting the power limit with the Intel XTU.nevcairiel - Tuesday, August 8, 2017 - link
Anyone spending this kind of money on a CPU should probably take some care to properly set it up - you don't even need XTU for that, the relevant settings are available in the BIOS as well.Defaults greatly differ between boards, that is true.
Gothmoth - Tuesday, August 8, 2017 - link
@nevcairiel .... what a complete BS.people who are using these kind of cpus are often artist and other professionals who are not computer nerds. a computeer should run fine with default settings. the need to "properly setup the bios" is something 80% of the customers will probably never do. the world is bigger than a few nerd websites.
nevcairiel - Tuesday, August 8, 2017 - link
Well those people are also unlikely to self-assemble a PC, aren't they?Whoever does the assembly of such an expensive system should know how to set it up.
When you buy a pre-assembled system, I would expect such a setup to have been performed indeed.
Ian Cutress - Wednesday, August 9, 2017 - link
We posted about MB manufacturers doing their own thing YEARS ago. www.anandtech.com/show/6214/multicore-enhancement-the-debate-about-free-mhzGothmoth - Tuesday, August 8, 2017 - link
what is this list at the bottom exactly?in the text you mention a max turbo of 4.4 for the 18 core.... in the list the max turbo is 4.2 GHz.
Elsote - Tuesday, August 8, 2017 - link
In the table there is a "turbo clock (4.2)" and a "turbo max clock(4.4)"Ian Cutress - Wednesday, August 9, 2017 - link
TB2 is 4.2 GHzTB3 (favored core) is 4.4 GHz, requires OS software.
azazel1024 - Tuesday, August 8, 2017 - link
I find it interesting that Intel is making this move. I mean, it makes sense for them too.On the other hand, looking at the charts, the $999 part for Intel gets you 10 cores, 20 threads, 3.3GHz base, 4.3GHz turbo and 40 PCI-e lanes. The $999 part for AMD gets you 60% more cores and threads (16/32), 3.4GHz base, 4GHz turbo and 60 PCI-e lanes.
I am sure the Intel part will be better in single thread, but it looks like the AMD part likely promises to crush the same price Intel part in most all multithreaded workloads.
Which is nice. For the last few years the number of workloads where you could say Intel was the price-performance leader were very, very, very tiny. Now it looks like AMD really does promise to be the price-performance leader, and in some cases outright performance leader in a number of categories.
If I had the spare change for an upgrade I'd definitely be looking at the "low end" Ryzen 8/16 part as an upgrade of my Ivy Bridge quad core desktop. Then again, maybe next year AMD will have something even better in the offing.
B3an - Tuesday, August 8, 2017 - link
These CPUs and the motherboard platform is a joke. I'm not paying 2K when i can get basically the same performance for half the price, which i will be doing.beck2050 - Tuesday, August 8, 2017 - link
Love 18 core 7980xe in a new Mac Pro.Da W - Tuesday, August 8, 2017 - link
Incredible how long we've been stuck with 4 cores processors and out of sudden they are able to throw 18 core at us in a few month span!!! Shows how bad a monopoly can be.Whatever it is Intel can come up with, i'll buy AMD stuff for my next rig just to help them STAY ALIVE for a while longer and continue to bring competition to the market. The desktop market has been boring for a decade! I was getting sick to come to Anandtech just to read crappy phone reviews ;)
sld - Wednesday, August 9, 2017 - link
Well said. And AMD has made it compelling to buy their processors, finally. One will really need to stretch logic and reality to continue with a near-monopoly situation like the past at least FIVE years.Gothmoth - Wednesday, August 9, 2017 - link
i guess a lot of sandy bridges are retired this year.but from what i see, in my surrounding, not with INTEL cpu´s.
i was ready to buy x299 ...but it is a big disappointment.
i rather spend my money on a nice threadripper build.
Gothmoth - Tuesday, August 8, 2017 - link
will the error in the last slide be fixed?turbo slows down i5??
Ian Cutress - Wednesday, August 9, 2017 - link
FixedManoa - Tuesday, August 8, 2017 - link
+1twtech - Wednesday, August 9, 2017 - link
Pretty nice. I wish in a way that I had waited to buy my workstation build for this - but it's always going to be the case that something faster comes out later.Is the 4.2GHz number the all-cores turbo boost? That's pretty high for an 18-core part. Especially since the 2687Wv4 maxes out at 3/3.1.
Gothmoth - Wednesday, August 9, 2017 - link
depends on how long it can run at 4.2 GHz before throtteling.....Ian Cutress - Wednesday, August 9, 2017 - link
4.2 is 2-core boost, as per the image at the bottom of the piece. All-core is 3.4G.BOBOSTRUMF - Wednesday, August 9, 2017 - link
Unfortunately for Intel, his greed really shows now. Although core still has about 5-10% more IPC compared to Ryzen, the power consumption per core is about 5-15% higher and with 18 cores this really shows. They had a very competitive tick-tock strategy and now after more than three years they are still stuck in 14nm. If they were smarter and created by now just one fab on 8 or 10 nm for the CPU's with many cores things were more simple for them today. In 8 nm, skylake x would had allowed 18 cores on 3.2-3.6Ghz, and not 2.6 as they are doing now.So they save 3-4 billions dollars not building a 8nm fab but will lose more than this when enthusiast marked will side with AMD.
Please be smarter in future Intel, Samsung and TSMC already have 8 nm FABS while you...
NeatOman - Thursday, August 10, 2017 - link
Intel's 18c 2.6GHz base clock with a 165 watt TDP vs AMD's 3.4GHz base clock with a 180 watt TDP doesn't look like much of a win, especially with less PCI-E lanes. Especially with new test done comparing the Ryzen R5 1600 vs the i7-7800x and both at stock the Intel chip clearly won, but when both were overclocked (AMD 4.0GHz, Intel 4.7GHz) both running 3200MHz ram they were nearly identical within 1%. Which supports people saying that AMD scales better and has better multi-threading (SMT vs HT).I'm VERY curious to see the results for Ryzen running on quad channel 3200 memory.
I got the FX-8320 the month it came out for about $180 together with a great motherboard, so it was a no brainer.. kinda lol. Had 4 sticks of 1333MHz ram laying around totaling 8GB and quickly upgraded to 16GB 2400MHz CAS-11 (running at 2133MHz CAS-9) and did some testing. Set it to 1333MHz and matched the old ram, then OCed it to find about a 15%+ difference. I did the same thing with a Q9550 i had before and only found about a 5% gain from 4GB 800MHz CAS-5 to 8GB 1150MHz CAS 5 and made sure that there wasn't any page filing.
Point is! lol, I'll wait until i see a RAM speed wall because i don't want to buy some ram then find out 4000MHz is the sweet spot then feel neutered lol. Also, it looks like Ryzen is massively affected by RAM because of CCX interconnecting the cores. It also seems that Ryzen scales very well in frequency.
Avro Arrow - Thursday, August 10, 2017 - link
In case nobody has noticed (very slim chance but there's a lot of comments), you have Threadripper listed as having only sixty PCI-Express 3.0 lanes when your own article about it from May says sixty-four which is the correct number. You even have the slide on your site. You might want to correct that because some people might get the wrong idea. I'm sure it's just a typo or something but you know how these things can get.http://images.anandtech.com/doci/11482/threadrippe...
peevee - Thursday, August 10, 2017 - link
7960X and 7980XE make no sense at all. Base clock decrease combined with less than linear scalability of even multi-threaded applications combined will give the same real life performance as 7940X.3.1 * 14= 43.4
2.8 * 16= 44.8
2.6 * 18= 46.8
Non-linear scalability, 3rd level cache contention, memory bus contention, TDP limit will do the rest. And lower ST performance will seal the deal.
Don't waste your money.
wallysb01 - Friday, August 11, 2017 - link
Guys, base clock rate is really just a worst-case scenario. You see the Turbo boosts by number of active cores, right? For the most part, with good cooling that is, those will be achieved and sustained. Or at the very least its going to be closer to that than to the base clocks. So...3.8 * 14 = 53.2
3.6 * 16 = 57.6
3.4 * 18 = 61.2
If you already have intel systems, download the Intel power gadget, open up activity monitor and run something that maxes all your cores. You'll probably see that so long as temps aren't going crazy, you'll be running above your base clock rate. Even on my MacBook Pro, I can do this with temps around 98C and clock rates sustaining about .4GHz over my base clock for as long as I cared to let the thing run (about 5 minutes, and maybe 2 minutes in it seem to reach a steady state). It state that the 3-4 core turbo on my 4980HQ is 3.6, but I'm really only able to sustain 3.2. This is a laptop though...
eachus - Sunday, August 20, 2017 - link
Back when AMD came out with the original (Sledgehammer) Opterons there was one feature that didn't show up on any spec sheet, but was known by the people in the trenches who wrote server code. (Or in my case, put in compiler optimizations used by the HPC crowd.) Opterons were predictable. If a given job took thirty minutes to run today, it would be within a minute of that (both wall clock and CPU times) tomorrow. I never had a case where the Opteron code path was not pretty much common to all chips. Yes, occasionally you did have to put in "ifdefs" to distinguish between single chip and multiple chip configurations, but no with and without Hyperthreading, with and without SSE, or the latest feature of the month that marketing said we just HAD to support.Now with Zen, AMD is getting back to that part of their roots. Zen code runs on all Zen chips. Yes, I am writing code that determines how many chips, cores, and threads are present, but that's about it.
Oh, while I am grousing in Intel's general direction, why do they need three subsets of AVX-512. Yes, that is subset, no current processor supports all of AVX-512, and no future processor is expected to do so....
SloppySlim - Wednesday, August 23, 2017 - link
and still no FPGAs ?