If it has 5960X performance at 400$ it will be a killer combo with Vega 10 that would hopefully arrive in October.But Take a heavy pinch of salt whenever talking about Zen
If it's performance competitive with the 5960X it probably won't be priced at $400. It'll probably be priced between the closest competing Broadwell-E SKUs (6900K-6850K).
This. If AMD can justify a competitive price, they will do so. They won't just sell a Skylake-level 8C16T CPU for $400 if they can sell them for $1,500/ea to a server farm or to enthusiasts.
Then again... the 295X2 sold for $1,500 when the Titan Z was $3,000, despite the 295X2 being the better card overall. Too bad we'll have to wait until 2017, after Kaby Lake is out, to see Zen IRL.
AMD only mentions IPC and never mentioned peak performance. By AMD standards that means only one thing. It will be a RX480 style release. Its top cpu will be a power efficient competitor to the lower end chips from Intel. Likely running at 2.4 ghz max. Or simply matching their current cpus just in a lower power package. Nothing in their GPU manufacturing example shows otherwise.
Yup. AMD needs to do what Intel did with Conroe. When the E6300 came out, it was around $330-350 and beat the FX 62 in almost everything. The FX 62 at the time was something like $800+.
The pure size of the cache is becoming ever less important in the face of more sophisticated cache management algorithms. The question that matters is not the size of the caches, it is how smart their cache management algorithms are. These include issues like - where are new lines placed in the recency chain? - how rapidly are lines promoted up this chain? - do you handle prefetched lines differently? - do you handle I-lines differently? - are your caches inclusive, exclusive, or neither? - how do your outer caches know that inner-cache lines are aggressively in use and so should not be dropped? (Relevant to inclusive caches.) - how do you prevent thrashing/streaming cores from trashing the LLC for everyone else? etc etc Correct answers to these questions is worth 30% or more in performance --- a whole lot more than just adding a MiB of cache here or there.
Intel only gives their cores 64k of L1 cache and 256k of L2 cache. Their L3 cache is usually about 1.5-2.5MB per core (lower end chips have less, like the Core i5 6600k has only 6MB of L3 across 4 cores) with all of it accessible to all cores, unlike the L1 and L2 cores.
AMD gives their cores double the L1 cache and quadruple the L2 cache currently. From what I've seen of rumors of the Zen architecture, it will still be double on both counts with more L3 cache. I don't see any mention of cache in the above article.
I just said "less" not "less than 1.5MB". The lower end chips have less cache, usually 1.5MB, while the higher end chips have more, 2-2.5MB/core. It was more of a qualifying remark to the "1.5-2.5MB per core" statement. Sorry if that got you confused.
Now, on to what I've learned about Zen. From what I've read, AMD has developed Zen cores in a quad core module with a shared 8MB L3 cache between each 4 cores. So a 8 core Zen based chip is said to have 16MB of cache. That's 2MB of L3 per core.
The newest FX and A series chips (Bristol Ridge) with integrated graphics are said to be excavator cores with a shared 2MB L3 cache. I think that's probably where Morawka got the "so little cache" idea. The trick is that AMD uses an exclusive cache scheme (possibly at the cost of a little performance) so that the data in the L1 and L2 caches aren't duplicated in the L3 cache. Intel uses an inclusive cache, and 2MB of their L3 is lost to the L1 and L2 cache from each core. An FX processor with the latest core would have an equal caching scheme to 6.25MB of Intel's version of cache. (4X 64k L1 + 2X 2MB L2 + 2MB L3 = 6.25MB) Whereas a Core i5 6600k has only 6MB of usable cache because the L1 data is stored in the L2 and L3 in duplicate and the L2 is stored in the L3 in duplicate. So the 6MB of L3 is all the Core i5 has.
Exclusive caches can have sizable performance hits in cases where you share data among cores. Exclusive caches have to do a lot of cache snooping, which increases latency and needlessly consumes bandwidth. Less wasted space, but at a cost.
BTW no current AMD apu's have L3 cache including Bristol Ridge. Bristol Ridge high end APU's have 2MB of L2. http://www.amd.com/en-us/products/processors/lapto... Go down to "Model Comparisons and Specs" and it will show details for the APU's..
AMD doesn't have the process node technology nor the fabrication production capability to brute force that much cache on a chip and still remain price competitive with Intel. The smaller cache is why AMD can still survive despite selling chips half the price of intel for so long.
AMD doesn't have the process node technology nor the fabrication production capability to brute force that much cache on a chip and still remain price competitive with Intel. The smaller cache is why AMD can still survive despite selling chips half the price of intel for so long.
If Zen actually has 8 *REAL* cores this time, then it could be faster at highly multi-threaded tasks. It will never beat Broadwell-E, but Skylake which only has 4 cores (still! pathetic) could be beaten by Zen.
BTW, does anyone know if they're real fully individual cores in Zen? None of that shared core crap like their previous CPUs, which AMD are rightly being sued over for false advertising.
Which is good, and I hear they'll have 6-cores with 12 threads too. Which I may end up buying, because if the 8-core is really that good, they'll price it accordingly. So I may have to get a 6-core Zen if/when I upgrade.
There is no reason AMD can't disable one, two, or even three cores within a module - the cores only share L3 and an inter-core bus. You can leave one core in a module enabled and disable the other three, leaving that one core with all of the L3... or you can even slice off part of the L3.
So the Summit Ridge die could be configured with any of the following core configurations:
1, 2, 3, 4, 5, 6, 7, 8 And SMT can be disabled as well...
Of course, AMD will probably only offer six and eight core designs as that should equate to 90%+ effective yields while still commanding good margin and market penetration. An up-charge for SMT would allow them to address even more of the market with the same die.
I cant wait to see. If I get my facts right. AMD moved from SMC using 4 sections containing two integer clusters and a sharable FPU cluster only able to execute one thread per integer cluster; to SMT with 8 modules containing one integer and FPU cluster with the ability to process multiple threads each (In Intel and presumably AMD's case two threads are interlaced). This is a major gain, so who knows. I remember a company with a product with 8 cores later adding two more in light of news from the competitor.
the bulldozer also had 8 REAL CORES genius! They just grouped 2 cores on a module. The cores were completely 100% functional but they shared resources(like cache) clustered in that module. It is very good for multi-threading(Old FX's are still competitive on that matter) but quite crappy on single-threading. Problem is that most programs and games still run on single or a maximum of 4 threads hence the inefficiency of any cpu with more than 4 threads/cores. Next time do your homework! And the lawsuit failed by the way as there really are 8 x86-64 cores.
" Problem is that most programs and games still run on single or a maximum of 4 threads hence the inefficiency of any cpu with more than 4 threads/cores. Next time do your homework!"
I suggest you do yours because this isn't how OS scheduling and programs work.
8 full integer cores, but only half the number of FPUs, which had to be shared in each module. Heck, if I run Geekbench 3.2 right now on my 4790k (running 4.4 (turbo), but across all cores) and 9590 (oc'd slightly to 5ghz constant), the 9590's integer performance slightly beats the 4790k, which otherwise pretty much stomps the 9590 in any game.
I'd much rather have four blazing fast cores than six or eight slower ones for my desktop. Servers are a different matter, but that's rarely what these chips are used for.
Your opinion is likely to change over the next 3-5 years. All of the new (2015/16 released) game engines manage to get good results from more than 4 cores. Of course it takes a while until those multi-year-development games catch up with engine technology, but it looks like a lot of games in the near future will be able to use 6-cores and 8-cores to hit those 120+X fps relevant for twitch gaming and VR scenarios.
That is why doing that in 2011 was insane. Sometimes AMD looks too far ahead. They need winning designs for the code that is being run now.
Still remember the first 64bit CPUs sold years before 64 bit software (at least in windows) was even thought off. In fact i has a 64 bit CPU that never saw 64bit code, as after 5 years it died before a mainstream 64 OS existed.
Yeah, it was too scary for me, so I bought the i7-950 instead. Clocked it to 4.2. Haven't upgraded my cpu yet. My PC has a Firestrike score of 10,700. Why should I? Too bad I didn't see the future in 2011.
Where are you going to find these "blazing fast cores?" Single thread performance is at a dead end. Increased parallelism is the future it's just that most software makers haven't quite received the memo yet, or they have but they think it's too hard.
Because it is very hard to almost impossibl, no amount of work can change the way algorithms work. In fact majority of code is not parallelizable and will never be.
It's currently couched in there "module" and "core" languaged. Windows reports them as cores and threads respectively. I haven't heard anything, but they did get sued for this.
This is a new ground up architecture, I've been waiting for it the moment bulldozer came out and I finally admitted I had to buy a 2500k cpu. It uses threads, the wording is correct. It's completely different design from the bulldozer and it's revision designs.
They got sued by an opportunistic lawyer who represented himself in the case and doesn't understand CPUs. I never heard of the case going anywhere. If FPU is required for something to be called a "core" then there's a Hell of a lot of older chips out there you'll have to stop defining as CPUs, and some modern chips as well.
"so with 40% IPC increase where does this theoretically put them in regards to Skylake i7?"
That's a good question. I did a bit of digging. Comparing an FX-8320 (3.5-4Ghz) to an i7 6700 (3.4-4Ghz) gives you these results in Anandtech Bench: http://www.anandtech.com/bench/product/698?vs=1554 It's not pretty reading! Admittedly there's only two single threaded tests, both from Cinebench, and I don't claim to know how representative those Cinebench tests are of general single-threaded performance. But you can see, clock for clock in those two tests, Skylake is within a whisker of double the single threaded performance of the AMD Vishera. Ouch.
I'm really hoping Zen does well, but I am worried that 40% IPC just isn't going to cut it, even if they can clock it higher and offer genuine 8 cores.
AMD never built a comparable Excavator chip. Excavator was only ever used in low power APUs. Piledriver/Vishera was as far as they got before they gave up.
Then lets go as close as we can get: http://www.anandtech.com/bench/product/1291?vs=154... Obviously - this is not "fair" in respect of price, but it is as close as it gets when you compare core-for-core with same frequency (=IPC). You can add few percent for improvements excavator brings over steamroller, but it doesn't change the picture. You would need WAY more than 40-50% to match Skylake even if we cherry-pick single-threaded tasks so we don't hurt AMD with their own definition of core.
Actuallly given the zen architecture this would be inaccurate you would need to compete against a comparable skylake i3. So this -> http://www.anandtech.com/bench/product/1291?vs=164... would be a better comparison. Remember 1 Zen core = 1 Vishera Module. A10 7800 would be comparible to a Dual core Zen which would be compared to the i3 not i5. From what I can see Zen has two things going for it. The 40% IPC increase and the fixing of AMD's "Core count" in the view of the consumer. But anyways as seen in the benchmarks a 40% boost should make it comparable.
You are mistaking IPC with performance. AMD was talking about IPC and in that case there is NO equivalence 1 Zen Core = 1 Buldozer Module. My comparison was about IPC. You can take single-threaded (and also arguably dual-threaded) benchmarks from my link and add 40-50% and you get what AMD promises. And assuming AMD will achieve similar frequency as intel, it will not be enough to match performance per core (but it will take AMD from "pathetic" to "ok"). This assessment will apply to all single-threaded applications and also to all applications, that actively use less threads, than competing intel CPUs have cores (2 for i3 and 4 for i5/i7). We can only guess how it will scale beyond this because we have no idea about SMT implementation, thermal limits and other variables. It is probably good guess, that in massively multi-threaded apps like Cinebench 8-core Zen will come on top (v 4-core i7) ... but its only a guess.
Cinebench? I'm not sure it's a particularly fair benchmark. Even so, the Bulldozer architecture was hamstrung by having a much narrower FPU interface. Zen is twice as wide. Some benchmarks will be above the 40% value, some will be less.
40% IPC + some gains from others stuff, like moving from the module architecture to the "classic" one. Add HT, add better cache and more stuff like that and you have a much faster than 40% CPU
40% should be enough in order to have a GTX 1070 / Radeon Fury X with no CPU limitations in gaming scenarios, I think..
Even Haswell has problems with GTX 1070 according to reviews I've seen - check Ashes Of singularity DX11 vs DX12 performance with an I7-5970X or something like that
It depends. A 40% IPC increase would bring a kraken 1.1 score of 1200 mS, which is still garbage considering my G3258 scores under 1000 and a 6700K can easily hit 700. Hell, the next iphone is likely to do better than 1200mS.
This could be really cool. I hope its not some lawyer marketing nonsense like 40% IPC per core, but its split across two threads, so its just 20% increase per thread in real benchmarks and less if they had to reduce the frequencies.
If the frequencies jump as they should for the newer process, it could be the new budget superchip!
Given the SMT nature of the new cores the two threads will never benefit the same from whatever IPC increase so even if is really 40% split among the two threads (both of which need to be running on the same core) the IPC gain will weigh heavily in primary thread's favor ...
The typical split of 70-30 to 85-15 between primary and secondary thread means at least 30-35% IPC increase for a single thread running on the new core.
"Wider sampling to their larger OEM base " Does AMD still have a OEM base for desktop CPUs? I can't remember the last time I saw a AMD based desktop in a store. It's like the single runt AMD laptop on the shelf surrounded by Intel ones.
HP, Toshiba, Lenovo, even Dell sometimes still make plenty of AMD based machines. Unfortunately they usually take decent parts and surround them with crap. 15" laptops with 1366 x 768 screens, 4gb of single channel memory, and the slowest hard drives you can find with an AMD A10 or FX APU. Even with the 15 and 19w chips they basically dump them into crap commodity cases with crap everything else, even though they could make some decent and fairly cheap thin and light designs that would probably sell quite well. Examples below.
Yeah we all know the runty AMD laptops Intel asks the OEMs to build to save them from monopoly investigations, but I'm talking...desktop. Also specifically anything with a full size FX/AM3+ chip in it.
The initial launch of Zen is aimed at OEMs that are more open to them, like HP, as well as enthusiasts. If they can prove themselves with Zen CPUs, that will boost adoption of the platform, and help the Zen APUs nab design wins across the market. But that's not going to be easy, and it won't happen overnight.
Ryan Smith and/or Ian Cutress, I don't know if you will see this, but I found this picture of the back of the Zen CPU: http://www.jagatreview.com/wp-content/uploads/2016... In the past, usually the size of the die is slightly smaller than the cut out on the back without connections/pins. Do you now if either of you could possibly approximate the size of the Zen die using that image?
Nevermind don't think that will work... Just realized old AMD CPUs had pins over the whole socket and that for Haswell a lot of the die passed over an area with pins
"shows silicone" ... Their Heatspreaders are made from silicone? Doesn't showing silicon usually mean holding up a processed waiver, so that media representatives with a quick hand and a very good optical zoom can figure out some details about the chip in question?
Not to hate on AMD, but showing off something that looks exactly like its predecessor is a bit sad.
The 5960X was the processor that could convince me to upgrade my i7 920, for the number of cores more than the single thread performance.
But his price tag killed the plan. If AMD offers something competitive, I'm going to purchase it, and since processors progress so slowly these days, it looks like I will keep it for a looong time, and Intel will not get any money from me for ages.
But it really needs to deliver performance at a reasonable price.
All the upgrades I skipped for avoiding those "5% faster than older generation CPU" piled a huge amount of money, and I want to spend it. But I will no waste 1700$ on what should be a mid range CPU.
AMD has been testing engineering samples for a considerable time by now. One thing I'm sure of is if this new architecture was as promising as it has been hyped, AMD would have already provided a figure or a real comparison demo to show it.
I have lost the last specks of trust in AMD people since the Bulldozer was released. I don't think this time is any different.
Man... Size matters. They just released Polaris. Do you want AMD being able to develop simultaneously completely new CPU and GPU architectures?
Be realistic... AMD can spend just over 1b$ R&D for BOTH CPU and GPU while Intel is spending about 4b$ (almost) only for CPU and Nvidia 2b$ (almost) only for GPU...
Furthermore, and they know it, Zen is make or break. If it goes wrong, AMD will probably be over. They will finetune as much as possible. This time, they really can't screw up.
Oppositely, I find it AMAZING that within a calendar year AMD is able to launch both CPU and GPU. And when you see the R9 480 being a 390-equivalent for about twice less TDP and for 200$... That's just awesome when you compare it with a GTX970 that costs the double, is more power-intensive and is slower...
Yes but as we have seen several times in the past AMD are fantastic at hyping up pre-launch tech demos etc. etc. Promising revolutionary leaps of performance and efficiency. Only to find on release day it's slower than the previous tech or just 5% faster. With AMD its best to expect disappointment on the CPU front, as that has been the result the past three to four releases.
A bigger problem with AMD APUs is not just the slow progress but also the fact that AMD doesn't seem to understand that if there are no decent computers using AMD chips, AMD will forever be relegated to picking up the scraps off the table. Google created the Nexus hardware brand in order to showcase the Android tech, and Microsoft created the Surface line in order to show of the Windows tech. AMD should take a page out of that book, and design a family or two of slick professional laptops, one for gaming/performance, kind of like the Skylake Dell XPS machines, and another line of ultra-portables in the same niche as say Lenovo Yoga. AMD chips have a lot of fans, and I suspect such products could at least break even while improving the AMD name.
Except that both of those companies have to look *hard* for places to shove all that excess money that is falling out of each company. AMD really doesn't have that problem. I'm probably even more annoyed that AMD never seemed to make the 8+ CPU servers that the original Sledgehammers suggested (from later analysis the whole thing was BS: the bandwidth needed exploded after 4 CPUs).
I think google loses money on Nexus, and I'm sure MS loses money on Surface: it doesn't help that NFL commentators call them ipads or use them to prop up ipads. This is all pretty much the penalty for paying cash for ATI (a stock swap might have saved both companies).
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
89 Comments
Back to Article
Eden-K121D - Tuesday, May 31, 2016 - link
If it has 5960X performance at 400$ it will be a killer combo with Vega 10 that would hopefully arrive in October.But Take a heavy pinch of salt whenever talking about Zendragonsqrrl - Wednesday, June 1, 2016 - link
If it's performance competitive with the 5960X it probably won't be priced at $400. It'll probably be priced between the closest competing Broadwell-E SKUs (6900K-6850K).nathanddrews - Wednesday, June 1, 2016 - link
This. If AMD can justify a competitive price, they will do so. They won't just sell a Skylake-level 8C16T CPU for $400 if they can sell them for $1,500/ea to a server farm or to enthusiasts.Then again... the 295X2 sold for $1,500 when the Titan Z was $3,000, despite the 295X2 being the better card overall. Too bad we'll have to wait until 2017, after Kaby Lake is out, to see Zen IRL.
SeanJ76 - Wednesday, June 8, 2016 - link
AMD has never been competitive......acme64 - Wednesday, June 8, 2016 - link
you got a short term memorysc14s - Thursday, June 9, 2016 - link
err they actually used to be competitive not all that long ago all things considered. Though I guess its ancient times by tech standardsredraider89 - Thursday, June 30, 2016 - link
No, you just have never been paying attention to what's going on.sharath.naik - Saturday, July 9, 2016 - link
AMD only mentions IPC and never mentioned peak performance. By AMD standards that means only one thing. It will be a RX480 style release. Its top cpu will be a power efficient competitor to the lower end chips from Intel. Likely running at 2.4 ghz max.Or simply matching their current cpus just in a lower power package. Nothing in their GPU manufacturing example shows otherwise.
Shadow7037932 - Wednesday, June 1, 2016 - link
Yup. AMD needs to do what Intel did with Conroe. When the E6300 came out, it was around $330-350 and beat the FX 62 in almost everything. The FX 62 at the time was something like $800+.III-V - Tuesday, May 31, 2016 - link
I'm calling wood screws on this.Just kidding.
tamalero - Wednesday, June 8, 2016 - link
I understood that reference!0ldman79 - Thursday, June 9, 2016 - link
I didn't. Need more coffee.Morawka - Wednesday, June 1, 2016 - link
so with 40% IPC increase where does this theoretically put them in regards to Skylake i7?Also, is hyperthreading new to current AMD cpu's?
Morawka - Wednesday, June 1, 2016 - link
Also, why so little Cache? when intel gives each core 1.5 or 2MBdarkvader75 - Wednesday, June 1, 2016 - link
AMD uses L3 cache as well so be cautious in comparing cache sizes until after the full specification is announced.name99 - Wednesday, June 1, 2016 - link
The pure size of the cache is becoming ever less important in the face of more sophisticated cache management algorithms. The question that matters is not the size of the caches, it is how smart their cache management algorithms are. These include issues like- where are new lines placed in the recency chain?
- how rapidly are lines promoted up this chain?
- do you handle prefetched lines differently?
- do you handle I-lines differently?
- are your caches inclusive, exclusive, or neither?
- how do your outer caches know that inner-cache lines are aggressively in use and so should not be dropped? (Relevant to inclusive caches.)
- how do you prevent thrashing/streaming cores from trashing the LLC for everyone else?
etc etc
Correct answers to these questions is worth 30% or more in performance --- a whole lot more than just adding a MiB of cache here or there.
dgingeri - Wednesday, June 1, 2016 - link
Intel only gives their cores 64k of L1 cache and 256k of L2 cache. Their L3 cache is usually about 1.5-2.5MB per core (lower end chips have less, like the Core i5 6600k has only 6MB of L3 across 4 cores) with all of it accessible to all cores, unlike the L1 and L2 cores.AMD gives their cores double the L1 cache and quadruple the L2 cache currently. From what I've seen of rumors of the Zen architecture, it will still be double on both counts with more L3 cache. I don't see any mention of cache in the above article.
So, why do you say "why so little cache?"?
JoeMonco - Wednesday, June 1, 2016 - link
"Their L3 cache is usually about 1.5-2.5MB per core (lower end chips have less, like the Core i5 6600k has only 6MB of L3 across 4 cores) "Your parenthetical statement doesn't make sense. 6MB / 4 = 1.5 MB. How is that less than the 1.5-2MB range you state just prior?
dgingeri - Wednesday, June 1, 2016 - link
I just said "less" not "less than 1.5MB". The lower end chips have less cache, usually 1.5MB, while the higher end chips have more, 2-2.5MB/core. It was more of a qualifying remark to the "1.5-2.5MB per core" statement. Sorry if that got you confused.Now, on to what I've learned about Zen. From what I've read, AMD has developed Zen cores in a quad core module with a shared 8MB L3 cache between each 4 cores. So a 8 core Zen based chip is said to have 16MB of cache. That's 2MB of L3 per core.
The newest FX and A series chips (Bristol Ridge) with integrated graphics are said to be excavator cores with a shared 2MB L3 cache. I think that's probably where Morawka got the "so little cache" idea. The trick is that AMD uses an exclusive cache scheme (possibly at the cost of a little performance) so that the data in the L1 and L2 caches aren't duplicated in the L3 cache. Intel uses an inclusive cache, and 2MB of their L3 is lost to the L1 and L2 cache from each core. An FX processor with the latest core would have an equal caching scheme to 6.25MB of Intel's version of cache. (4X 64k L1 + 2X 2MB L2 + 2MB L3 = 6.25MB) Whereas a Core i5 6600k has only 6MB of usable cache because the L1 data is stored in the L2 and L3 in duplicate and the L2 is stored in the L3 in duplicate. So the 6MB of L3 is all the Core i5 has.
bcronce - Thursday, June 2, 2016 - link
Exclusive caches can have sizable performance hits in cases where you share data among cores. Exclusive caches have to do a lot of cache snooping, which increases latency and needlessly consumes bandwidth. Less wasted space, but at a cost.utroz - Sunday, June 5, 2016 - link
BTW no current AMD apu's have L3 cache including Bristol Ridge. Bristol Ridge high end APU's have 2MB of L2.http://www.amd.com/en-us/products/processors/lapto... Go down to "Model Comparisons and Specs" and it will show details for the APU's..
MobiusPizza - Thursday, June 9, 2016 - link
AMD doesn't have the process node technology nor the fabrication production capability to brute force that much cache on a chip and still remain price competitive with Intel. The smaller cache is why AMD can still survive despite selling chips half the price of intel for so long.MobiusPizza - Thursday, June 9, 2016 - link
AMD doesn't have the process node technology nor the fabrication production capability to brute force that much cache on a chip and still remain price competitive with Intel. The smaller cache is why AMD can still survive despite selling chips half the price of intel for so long.killerbunnies - Wednesday, June 1, 2016 - link
With a 40% IPC increase, it is still looking way up to Skylake i7 desktop platforms.B3an - Wednesday, June 1, 2016 - link
If Zen actually has 8 *REAL* cores this time, then it could be faster at highly multi-threaded tasks. It will never beat Broadwell-E, but Skylake which only has 4 cores (still! pathetic) could be beaten by Zen.BTW, does anyone know if they're real fully individual cores in Zen? None of that shared core crap like their previous CPUs, which AMD are rightly being sued over for false advertising.
Arnulf - Wednesday, June 1, 2016 - link
Yes, they are full cores with SMT, very much like Intel's cores.8 core Zen chip means 16 threads.
AS118 - Wednesday, June 1, 2016 - link
Which is good, and I hear they'll have 6-cores with 12 threads too. Which I may end up buying, because if the 8-core is really that good, they'll price it accordingly. So I may have to get a 6-core Zen if/when I upgrade.davegraham - Wednesday, June 1, 2016 - link
no, 4c/8t; 8c/16t will be the APU low-end. each set of cores is built into a 4 core module. so, your multipliers for cores will be in blocks of 4.each core will have 512k of L2 with a minimum of 2mb of L3 per block of 4 cores. (last time i checked)
subject to change, as it were.
looncraz - Wednesday, June 1, 2016 - link
There is no reason AMD can't disable one, two, or even three cores within a module - the cores only share L3 and an inter-core bus. You can leave one core in a module enabled and disable the other three, leaving that one core with all of the L3... or you can even slice off part of the L3.So the Summit Ridge die could be configured with any of the following core configurations:
1, 2, 3, 4, 5, 6, 7, 8
And SMT can be disabled as well...
Of course, AMD will probably only offer six and eight core designs as that should equate to 90%+ effective yields while still commanding good margin and market penetration. An up-charge for SMT would allow them to address even more of the market with the same die.
lunchbox4k - Wednesday, June 1, 2016 - link
I cant wait to see. If I get my facts right. AMD moved from SMC using 4 sections containing two integer clusters and a sharable FPU cluster only able to execute one thread per integer cluster; to SMT with 8 modules containing one integer and FPU cluster with the ability to process multiple threads each (In Intel and presumably AMD's case two threads are interlaced). This is a major gain, so who knows. I remember a company with a product with 8 cores later adding two more in light of news from the competitor.Spunjji - Wednesday, June 1, 2016 - link
If you think the way they marketed their cores was false advertising then I'm surprised to find you on a site like Anandtech.uzishan - Wednesday, June 1, 2016 - link
the bulldozer also had 8 REAL CORES genius! They just grouped 2 cores on a module. The cores were completely 100% functional but they shared resources(like cache) clustered in that module. It is very good for multi-threading(Old FX's are still competitive on that matter) but quite crappy on single-threading. Problem is that most programs and games still run on single or a maximum of 4 threads hence the inefficiency of any cpu with more than 4 threads/cores. Next time do your homework! And the lawsuit failed by the way as there really are 8 x86-64 cores.xenol - Wednesday, June 1, 2016 - link
" Problem is that most programs and games still run on single or a maximum of 4 threads hence the inefficiency of any cpu with more than 4 threads/cores. Next time do your homework!"I suggest you do yours because this isn't how OS scheduling and programs work.
blppt - Thursday, June 2, 2016 - link
8 full integer cores, but only half the number of FPUs, which had to be shared in each module. Heck, if I run Geekbench 3.2 right now on my 4790k (running 4.4 (turbo), but across all cores) and 9590 (oc'd slightly to 5ghz constant), the 9590's integer performance slightly beats the 4790k, which otherwise pretty much stomps the 9590 in any game.monsted - Wednesday, June 1, 2016 - link
I'd much rather have four blazing fast cores than six or eight slower ones for my desktop. Servers are a different matter, but that's rarely what these chips are used for.ShieTar - Wednesday, June 1, 2016 - link
Your opinion is likely to change over the next 3-5 years. All of the new (2015/16 released) game engines manage to get good results from more than 4 cores. Of course it takes a while until those multi-year-development games catch up with engine technology, but it looks like a lot of games in the near future will be able to use 6-cores and 8-cores to hit those 120+X fps relevant for twitch gaming and VR scenarios.valinor89 - Wednesday, June 1, 2016 - link
That is why doing that in 2011 was insane. Sometimes AMD looks too far ahead. They need winning designs for the code that is being run now.Still remember the first 64bit CPUs sold years before 64 bit software (at least in windows) was even thought off. In fact i has a 64 bit CPU that never saw 64bit code, as after 5 years it died before a mainstream 64 OS existed.
WhisperingEye - Saturday, June 4, 2016 - link
Yeah, it was too scary for me, so I bought the i7-950 instead. Clocked it to 4.2. Haven't upgraded my cpu yet. My PC has a Firestrike score of 10,700. Why should I? Too bad I didn't see the future in 2011.prisonerX - Wednesday, June 1, 2016 - link
Where are you going to find these "blazing fast cores?" Single thread performance is at a dead end. Increased parallelism is the future it's just that most software makers haven't quite received the memo yet, or they have but they think it's too hard.vladx - Thursday, June 2, 2016 - link
Because it is very hard to almost impossibl, no amount of work can change the way algorithms work. In fact majority of code is not parallelizable and will never be.stardude82 - Wednesday, June 1, 2016 - link
It's currently couched in there "module" and "core" languaged. Windows reports them as cores and threads respectively. I haven't heard anything, but they did get sued for this.ANobody - Wednesday, June 1, 2016 - link
This is a new ground up architecture, I've been waiting for it the moment bulldozer came out and I finally admitted I had to buy a 2500k cpu. It uses threads, the wording is correct. It's completely different design from the bulldozer and it's revision designs.h4rm0ny - Wednesday, June 1, 2016 - link
They got sued by an opportunistic lawyer who represented himself in the case and doesn't understand CPUs. I never heard of the case going anywhere. If FPU is required for something to be called a "core" then there's a Hell of a lot of older chips out there you'll have to stop defining as CPUs, and some modern chips as well.rhysiam - Wednesday, June 1, 2016 - link
"so with 40% IPC increase where does this theoretically put them in regards to Skylake i7?"That's a good question. I did a bit of digging. Comparing an FX-8320 (3.5-4Ghz) to an i7 6700 (3.4-4Ghz) gives you these results in Anandtech Bench: http://www.anandtech.com/bench/product/698?vs=1554
It's not pretty reading! Admittedly there's only two single threaded tests, both from Cinebench, and I don't claim to know how representative those Cinebench tests are of general single-threaded performance. But you can see, clock for clock in those two tests, Skylake is within a whisker of double the single threaded performance of the AMD Vishera. Ouch.
I'm really hoping Zen does well, but I am worried that 40% IPC just isn't going to cut it, even if they can clock it higher and offer genuine 8 cores.
mdriftmeyer - Wednesday, June 1, 2016 - link
Why are you comparing it with Vishera? This will be 40% IPC increase over Excavator. God people are dense around here.Gigaplex - Wednesday, June 1, 2016 - link
AMD never built a comparable Excavator chip. Excavator was only ever used in low power APUs. Piledriver/Vishera was as far as they got before they gave up.Spunjji - Wednesday, June 1, 2016 - link
Doesn't matter, it's an IPC comparison and Excavator's IPC is significantly higher than Vishera.Galatian - Wednesday, June 1, 2016 - link
Does AMD actually state to what Bulldozer Derivative they compare their "up to 40% IPC increase"?psychobriggsy - Wednesday, June 1, 2016 - link
Yes, Excavator.Some people suspect that over Bulldozer it's 60%+ higher IPC.
What really matters is what the clocks will be.
Meteor2 - Wednesday, June 1, 2016 - link
I don't believe that's the case. I've never seen AMD qualify the 40% statement.qap - Wednesday, June 1, 2016 - link
Then lets go as close as we can get:http://www.anandtech.com/bench/product/1291?vs=154...
Obviously - this is not "fair" in respect of price, but it is as close as it gets when you compare core-for-core with same frequency (=IPC). You can add few percent for improvements excavator brings over steamroller, but it doesn't change the picture. You would need WAY more than 40-50% to match Skylake even if we cherry-pick single-threaded tasks so we don't hurt AMD with their own definition of core.
caqde - Wednesday, June 1, 2016 - link
Actuallly given the zen architecture this would be inaccurate you would need to compete against a comparable skylake i3. So this -> http://www.anandtech.com/bench/product/1291?vs=164... would be a better comparison. Remember 1 Zen core = 1 Vishera Module. A10 7800 would be comparible to a Dual core Zen which would be compared to the i3 not i5. From what I can see Zen has two things going for it. The 40% IPC increase and the fixing of AMD's "Core count" in the view of the consumer. But anyways as seen in the benchmarks a 40% boost should make it comparable.JoeMonco - Wednesday, June 1, 2016 - link
Why would you compare to an i3 when the original question was in comparison to i7s?qap - Wednesday, June 1, 2016 - link
You are mistaking IPC with performance. AMD was talking about IPC and in that case there is NO equivalence 1 Zen Core = 1 Buldozer Module.My comparison was about IPC. You can take single-threaded (and also arguably dual-threaded) benchmarks from my link and add 40-50% and you get what AMD promises. And assuming AMD will achieve similar frequency as intel, it will not be enough to match performance per core (but it will take AMD from "pathetic" to "ok").
This assessment will apply to all single-threaded applications and also to all applications, that actively use less threads, than competing intel CPUs have cores (2 for i3 and 4 for i5/i7). We can only guess how it will scale beyond this because we have no idea about SMT implementation, thermal limits and other variables. It is probably good guess, that in massively multi-threaded apps like Cinebench 8-core Zen will come on top (v 4-core i7) ... but its only a guess.
jimjamjamie - Wednesday, June 1, 2016 - link
It's refreshing to see that not everyone is blindly perpetuating the AMD hype. The TDP hasn't even been mentioned yet.silverblue - Wednesday, June 1, 2016 - link
Cinebench? I'm not sure it's a particularly fair benchmark. Even so, the Bulldozer architecture was hamstrung by having a much narrower FPU interface. Zen is twice as wide. Some benchmarks will be above the 40% value, some will be less.psychobriggsy - Wednesday, June 1, 2016 - link
Well, floating point is not integer. Cinebench is probably a poor choice to compare CPU performance - you should be doing the work on a GPU anyway.Secondly, Zen will have 8 FPUs, compared to Bulldozer's 4, so there's your 50% in FP.
cotak - Thursday, June 2, 2016 - link
40% increase likely doesn't cut it like you suspect.More cores = more chip area. And chip area cost money. If you can deliver a competitive product with 4 cores you'd do that rather than shove out 8.
junky77 - Wednesday, June 1, 2016 - link
40% IPC + some gains from others stuff, like moving from the module architecture to the "classic" one. Add HT, add better cache and more stuff like that and you have a much faster than 40% CPU40% should be enough in order to have a GTX 1070 / Radeon Fury X with no CPU limitations in gaming scenarios, I think..
Spunjji - Wednesday, June 1, 2016 - link
Given that Sandy Bridge with an OC is still "good enough" in games this should have no problems :)junky77 - Wednesday, June 1, 2016 - link
Even Haswell has problems with GTX 1070 according to reviews I've seen - check Ashes Of singularity DX11 vs DX12 performance with an I7-5970X or something like thatDuckeenie - Wednesday, June 1, 2016 - link
Because manufacturers are known to be conservative about their claims?junky77 - Wednesday, June 1, 2016 - link
lol, off with their heads!!Shadowmaster625 - Wednesday, June 1, 2016 - link
It depends. A 40% IPC increase would bring a kraken 1.1 score of 1200 mS, which is still garbage considering my G3258 scores under 1000 and a 6700K can easily hit 700. Hell, the next iphone is likely to do better than 1200mS.webdoctors - Wednesday, June 1, 2016 - link
This could be really cool. I hope its not some lawyer marketing nonsense like 40% IPC per core, but its split across two threads, so its just 20% increase per thread in real benchmarks and less if they had to reduce the frequencies.If the frequencies jump as they should for the newer process, it could be the new budget superchip!
Arnulf - Wednesday, June 1, 2016 - link
Given the SMT nature of the new cores the two threads will never benefit the same from whatever IPC increase so even if is really 40% split among the two threads (both of which need to be running on the same core) the IPC gain will weigh heavily in primary thread's favor ...The typical split of 70-30 to 85-15 between primary and secondary thread means at least 30-35% IPC increase for a single thread running on the new core.
T1beriu - Wednesday, June 1, 2016 - link
How long does the chip bring-up faze take? How many months?milli - Wednesday, June 1, 2016 - link
Where's that Excavator review Anand? You promised it more than three months ago.monsted - Wednesday, June 1, 2016 - link
Good luck to AMD! While i probably won't buy this stuff, i really look forward to someone snapping at Intels heels and making them move forward again.jabber - Wednesday, June 1, 2016 - link
"Wider sampling to their larger OEM base " Does AMD still have a OEM base for desktop CPUs? I can't remember the last time I saw a AMD based desktop in a store. It's like the single runt AMD laptop on the shelf surrounded by Intel ones.artk2219 - Wednesday, June 1, 2016 - link
HP, Toshiba, Lenovo, even Dell sometimes still make plenty of AMD based machines. Unfortunately they usually take decent parts and surround them with crap. 15" laptops with 1366 x 768 screens, 4gb of single channel memory, and the slowest hard drives you can find with an AMD A10 or FX APU. Even with the 15 and 19w chips they basically dump them into crap commodity cases with crap everything else, even though they could make some decent and fairly cheap thin and light designs that would probably sell quite well. Examples below.http://www.cpu-world.com/CPUs/Bulldozer/AMD-FX-Ser...
http://www.newegg.com/Product/ProductList.aspx?Sub...
jabber - Wednesday, June 1, 2016 - link
Yeah we all know the runty AMD laptops Intel asks the OEMs to build to save them from monopoly investigations, but I'm talking...desktop. Also specifically anything with a full size FX/AM3+ chip in it.artk2219 - Wednesday, June 1, 2016 - link
There really isnt much even on newegg, which means i doubt there would be much at a standard best buy.http://www.newegg.com/Product/ProductList.aspx?Sub...
http://www.newegg.com/Product/ProductList.aspx?Sub...
jabber - Thursday, June 2, 2016 - link
Yeah...AMD are so screwed at retail.Not good.Alexvrb - Thursday, June 2, 2016 - link
The initial launch of Zen is aimed at OEMs that are more open to them, like HP, as well as enthusiasts. If they can prove themselves with Zen CPUs, that will boost adoption of the platform, and help the Zen APUs nab design wins across the market. But that's not going to be easy, and it won't happen overnight.WhisperingEye - Saturday, June 4, 2016 - link
"...sometimes still make plenty.." Can someone quantify this sentence for me?Dr. Swag - Wednesday, June 1, 2016 - link
Ryan Smith and/or Ian Cutress, I don't know if you will see this, but I found this picture of the back of the Zen CPU: http://www.jagatreview.com/wp-content/uploads/2016...In the past, usually the size of the die is slightly smaller than the cut out on the back without connections/pins. Do you now if either of you could possibly approximate the size of the Zen die using that image?
Dr. Swag - Wednesday, June 1, 2016 - link
Nevermind don't think that will work... Just realized old AMD CPUs had pins over the whole socket and that for Haswell a lot of the die passed over an area with pinsShieTar - Wednesday, June 1, 2016 - link
"shows silicone" ... Their Heatspreaders are made from silicone? Doesn't showing silicon usually mean holding up a processed waiver, so that media representatives with a quick hand and a very good optical zoom can figure out some details about the chip in question?Not to hate on AMD, but showing off something that looks exactly like its predecessor is a bit sad.
pavag - Wednesday, June 1, 2016 - link
The 5960X was the processor that could convince me to upgrade my i7 920, for the number of cores more than the single thread performance.But his price tag killed the plan. If AMD offers something competitive, I'm going to purchase it, and since processors progress so slowly these days, it looks like I will keep it for a looong time, and Intel will not get any money from me for ages.
But it really needs to deliver performance at a reasonable price.
All the upgrades I skipped for avoiding those "5% faster than older generation CPU" piled a huge amount of money, and I want to spend it. But I will no waste 1700$ on what should be a mid range CPU.
texasti89 - Wednesday, June 1, 2016 - link
AMD has been testing engineering samples for a considerable time by now. One thing I'm sure of is if this new architecture was as promising as it has been hyped, AMD would have already provided a figure or a real comparison demo to show it.I have lost the last specks of trust in AMD people since the Bulldozer was released. I don't think this time is any different.
dinin70 - Thursday, June 2, 2016 - link
Man... Size matters. They just released Polaris. Do you want AMD being able to develop simultaneously completely new CPU and GPU architectures?Be realistic... AMD can spend just over 1b$ R&D for BOTH CPU and GPU while Intel is spending about 4b$ (almost) only for CPU and Nvidia 2b$ (almost) only for GPU...
Furthermore, and they know it, Zen is make or break. If it goes wrong, AMD will probably be over. They will finetune as much as possible. This time, they really can't screw up.
Oppositely, I find it AMAZING that within a calendar year AMD is able to launch both CPU and GPU. And when you see the R9 480 being a 390-equivalent for about twice less TDP and for 200$... That's just awesome when you compare it with a GTX970 that costs the double, is more power-intensive and is slower...
Kudos to AMD.
jabber - Thursday, June 2, 2016 - link
Yes but as we have seen several times in the past AMD are fantastic at hyping up pre-launch tech demos etc. etc. Promising revolutionary leaps of performance and efficiency. Only to find on release day it's slower than the previous tech or just 5% faster. With AMD its best to expect disappointment on the CPU front, as that has been the result the past three to four releases.UtilityMax - Sunday, June 5, 2016 - link
A bigger problem with AMD APUs is not just the slow progress but also the fact that AMD doesn't seem to understand that if there are no decent computers using AMD chips, AMD will forever be relegated to picking up the scraps off the table. Google created the Nexus hardware brand in order to showcase the Android tech, and Microsoft created the Surface line in order to show of the Windows tech. AMD should take a page out of that book, and design a family or two of slick professional laptops, one for gaming/performance, kind of like the Skylake Dell XPS machines, and another line of ultra-portables in the same niche as say Lenovo Yoga. AMD chips have a lot of fans, and I suspect such products could at least break even while improving the AMD name.wumpus - Tuesday, June 7, 2016 - link
Except that both of those companies have to look *hard* for places to shove all that excess money that is falling out of each company. AMD really doesn't have that problem. I'm probably even more annoyed that AMD never seemed to make the 8+ CPU servers that the original Sledgehammers suggested (from later analysis the whole thing was BS: the bandwidth needed exploded after 4 CPUs).I think google loses money on Nexus, and I'm sure MS loses money on Surface: it doesn't help that NFL commentators call them ipads or use them to prop up ipads. This is all pretty much the penalty for paying cash for ATI (a stock swap might have saved both companies).
aamartin - Saturday, June 11, 2016 - link
Why is it amazing? It's not like it's a single design team doing both designs.SeanJ76 - Wednesday, June 8, 2016 - link
LOL@AMD!More garbage releasing soon to collect dust on shelves!
SeanJ76 - Thursday, June 9, 2016 - link
AMD is GARBAGE!fanofanand - Thursday, June 9, 2016 - link
Insightful comment, thank you for contributing enormously to the comments for this article. Pat yourself on the back big guy.