Are you aware of what stepping actually is? As already stated in the article, they are to fix hardware-level erratas or improve the performance or efficiency or stability, or a combination of them all.
Release of a new processor stepping is usually a non-event, unless it is meant to fix a major errata.
Some errata can be "fixed" with bios updates, but some need a new stepping.
It's okay. Ignorance of an issue isn't bad, it's just lack of information about a given topic. Willful ignorance is the worst thing ever though. Hopefully, they learned a bit from this and are now wiser. Hopefully. A guy can hope, yeah?
I just wonder how much it costs these days to have an article published about a new CPU stepping (!) where you can't even see what's new (!!). I can't seem to find any other such article on AT over the past decade, and what is there is from back when a stepping made a world of difference in OC.
"Intel did not comment on the news story."
Of course not. They issued it.
"it is noteworthy that Intel plans to launch a new stepping of its Coffee Lake processors"
No it's not. Unless it's to keep Intel in the news with such fake articles that make it look like something's still moving at Intel.
In other news, motherboard manufacturer launched new mobo revision using slightly shinier capacitor next to one of the SATA ports. Significance of change still a mystery.
"Unless it's to keep Intel in the news with such fake articles that make it look like something's still moving at Intel." While I won't comment on whether the article is fake or "paid", of course the idea is to "keep Intel in the news".They've run out of + to add to their 14nm node, CPUs based on their 10nm+ node are to be released in high volume (presumably) starting from December 2019 but only for the mainstream desktop market and below, so they've now resorted to issuing press releases for ... new steppings!
By the way, they are apparently canning their 10nm node after just a single 10nm based Core-i3 15W specimen release in 2018, largely for schools in China, which had much worse clocks than its 14nm equivalent *despite* a disabled iGPU. The sole purpose of that pathetic specimen was naturally to report a nominal 10nm 2018 release to their investors and avoid having their stock hammered.
If Intel pull back again the release of Ice Lake CPUs deep into 2020 they will be hurt badly. AMD is already set to have a 7nm node head start of ~6 months. Will they gift them even more? Of course we still don't know how their (fixed) 10nm+ node is going to perform, even if it is not delayed further.
The benefits of a new stepping in the past is generally what transition of 14 nm to 14+ nm would bring: slight tweaks to the design to enable lower thermals or jump up a notch or two in clock speeds. There of course would be some errata fixes with each stepping.
The last time Intel really pushed a new stepping was for the TSX instructions in Haswell which were removed via microcode updates on affected chips. Intel did eventually release some Haswell-EP chips with it enabled but those are harder find (when this bug was discovered, the EP line up had it disabled by default but BIOS could re-enable it for developers).
Correct me if I'm wrong, but I was under the impression that stepping derived performance and/or efficiency improvements were the result of things like fixing slight manufacturing issues that, say, caused unnecessarily high voltages, etc.
I've been reading that even the latest skylake processors couldn't match kaby lake in OC or efficiency, or perhaps I've missed something.
I'm not aware of the exact differences between 14 and 14+, but it's been reported that 14++ actually has a different transistor geometry and has a larger gate pitch.
a stepping could fix a problem. like the security vulnerabilities etc. but im wondering if they are putting out a different die with the igpu cut off and some small improvements.
maybe thats what the kf cpus were about. getting people used to the idea of igpu-less cpus before this new stepping\die came out. that would save a lot of capacity with the 14nm shortage going on. of course the current kf cpus still have the igpu but its disabled.
cutting off the 9900k's igpu would actually make it smaller than the 8700k and you could probably make it for the same price as the 8700k. i expect intels rumored 10 core cpus to not have an igpu. that would make them smaller than the 9900k.
are intel releasing a destkop chip this year? they have to right? so why release a new stepping this late in the year when a new release is coming? im wondering if the rumored 10 core will work on 300 series boards. not making a new chipset would save even more fab space and would give intel a win in the value department if we could use our z370 or z390 boards with the new 10 core. so maybe they will just release a new stepping of the 9th gen chips, lower the price of the 9900k, and then add the 10 core on top? possible?
Well first they need to make a 95w 8 core ... and don't tell me the 9900k or even the 9800k are "95w"
The 9900k pulls the same all threads max power consumption as an AMD Threadripper 2950x - which has twice the number of cores. (each pull ~260w under prime95 stress test) When you hold a 9900k to its 95w tdp - it performs worse than a 2700x in multi-threaded loads so you know there's something fishy there ...
The 9800k pulls over 200w and more or less the same as the Ryzen 2700x, but the Ryzen has 8 more threads and better overall multi-threaded performance.
I saw one of Intel's 14 core 9990x CPUs draw almost 800 watts when overclocked to 5.1 ... yikes!
Short answers: 1) Intel's power consumption is actually more than AMDs overall, but Intel assigns super low TDPs that aren't real (they don't consider ANY boost clocks at all) - its a trick to make people believe that they still have more efficient CPUs - they do not. 2) there's no way they can rectify this issue by any decent amount on 14nm - if this new stepping brings them to a 10 core, expect it to be a heater with an artificial TDP slapped on as we already see with current 9th gen.
When they get to 10 or 7nm though - whole new ballgame, but the competitor will be beating them to the node shrink it appears. :)
" With the ASUS Maximus XI Hero that we used for our 9900K, a stricter adherence to Intel’s stock policies means we see a sharp drop-off in power consumption when testing under full stock conditions. The CPU falls to 100W load and stays there, leaving more performance available if we were to remove power targets and limits. Some of the other motherboards shipping today will exit these Intel power specs and draw more power. The 2700X pushes closer to about 120W-125W draw. Intel manages to achieve better overall combined throughput for the player and viewer while maintaining a lower power consumption, for which the 9900K deserves acclaim. It is, of course, significantly more expensive, and that is a big factor to play into our conclusion."
In their game streaming benchmarks the 9900k consumes less power while delivery better performance for both viewer and streamer compared to the 2700X.
Going past TDP is not inherently wrong for a desktop processor, if the cooling and power delivery are up to the task.
It's not surprising that 9900k draws so much power at such extreme frequencies. Once you pass a certain threshold, efficiency goes right out the window and power goes up exponentially.
It's also not surprising that when limited to 95W, it performs worse than a 2700x. At 95W the 9900k stays at its base 3.6 GHz clock speed when fully loaded, while 2700x is free to draw as much as ~105W and reach a higher frequency. It might not look like much, but that 10W is quite enough to make a difference. Nothing fishy here. A 9900k limited to 105W should at least be able to match its performance.
Same story for 9800k's high power draw.
If ryzen CPUs could reach 5.0 GHz, you'd see just as high, if not higher, power consumption numbers.
To properly test and compare the efficiency of two processors' architecture/node, you should run them at the same clock speeds, under the same conditions, to get the real picture.
As a side example, Ian managed to achieve 4.5 GHz with a 113W power draw.
Yes, he did choose the voltages manually (instead of using the Auto mode) which resulted in lower numbers but it does show the CPU is capable of running efficiently. Also, keep in mind that the Auto mode of some motherboards is too aggressive and choose unnecessarily high voltages.
Another interesting thing to note in that table; going from 4.5 to 5.0, an 11% increase, resulted in a 70% increase in power.
It seems 2700x can go well beyond 105W depending on the board; comparing a 95W-locked 9900k against it seems to be even less appropriate than I thought.
A new stepping is not a new node. It is just a slight or (rarely) moderate fix or "clean up" of the exact same node. The size of the transistors, cells and of the BEOL layers are never modified with a new stepping. If you do more than fix or clean up the node, but less than developing a true successor node (in order to increase transistor density and reduce die sizes), that's when you usually add symbols like + to create something halfway between a new stepping and a truly new node. Which is what Intel have been doing since 2014.
Intel only released a handful of Coffee Lake Refresh SKUs in October. The rest of the S stack and all of the H parts are still forthcoming. The second round of Coffee Lake Refresh is stepping R0.
Maybe they want to allow it to soak up more power so it can clock faster - possibly through otherwise-used-for-GPU pins that would normally have different tolerances? Just a thought.
To me, increasingly harder just seems redundant. Harder already implies an increase, whereas "increasingly hard" says that the level of difficulty is on an upward trend line, but without being redundant.
I'm planning on getting an Intel Core i9-9900k for my custom build desktop tower. Should I wait for the new stepping or should I just buy it now? How will I be able to differentiate between the two?
I believe that patel21 was referring to Ryzen 3rd generation chips, which are expected to have very similar clock speeds to the 9900k for less money and possibly better real world performance, even in single threaded applications.
I'd wait until June and verify what the Ryzen 3700X specs are first. If the 3700X is 4.2/5.0 for clock speeds for 12 cores/24 threads and you pay $330 for it, that would probably be a better buy for you.
BITD, Intel would've referred to each architecture with the same marketing name, irrespective of the process technology changes. So, in that sense, Skylake would logically group together with everything since.
But, it's not really accurate to group the 14++ nm processors with Skylake, in any strict sense. Even on a per-core basis, Coffee Lake is more than simply an overclocked Skylake. So, if you call them both Skylake, then you should probably also refer to Ivy Bridge as Sandybridge and Broadwell as Haswell.
Those chips in your last sentence were different architectures and process nodes. It is not the same situation at all.
Intel 6th, 7th, 8th, 9th gen is all the *same* core. Same architecture. Same IPC. Same memory controllers (new chips still support DDR3 with the right board). Same graphics (except updated video decode in 7th gen). The 14+++ improvements are about what you would expect after having multiple years to tweak the process for variants of the same exact architecture. He is not wrong to call it an overclocked skylake.
According to Intel's "Tick-Tock" strategy, the first -Bridge and first -Well were the new uArch points. Sure, you change a few things during a node-shrink - it's not strictly about clock speed and power as you also want to take advantage of more die area and greater timing flexibility. But, in Intel's own words, they were not new architectures. And I think probably not such a difference as you would find between the different generations of Pentium 4 chips, for instance.
Fingers crossed this setting thins out the silicon a bit. The 9700k and 9900k are 10+ degrees hotter than they should be primarily because of how stupidly thick the piece of silicon is. If you delid one of these guys and grind it down you'll gain 10 degrees. Pretty annoying that Intel finally started soldering their CPUs again but they almost doubled the z-height of the silicon at the same time.
I feel like people are confused on what a stepping and revision are.
In the past 12+ years ago both Intel and AMD would regularly come out with new steppings throughout the product cycle of a processor. The stepping would primarily be a retiming improvement of the processor to achieve more speed. For example the P4 130 nm (Northwood) launched at 2.2 GHz and typically overclocked to 2.6 GHz, but the final revision ran at 3.4 GHz stock, and overclocked to 4 GHz. The big increase in speed was just the engineers over time improving the layout of the same chip for higher speed.
AFAIK revisions after launch died out during the C2D era. The 65 nm C2D had a major revision (G0 I think), which boosted clocks around 10% over the launch revision. The 45 nm C2D had a similar 10% late clock boost revision, but this also increased L2 cache latencies, so the performance improvement was smaller.
What you ended up with was processors launched in a ever more optimized state, making it not worth the while of Intel and AMD to produce new expensive mask sets for very minor improvements.
When you look at this new R0 revision, according to Tom's review of the 9900KF R0, it clocks 100 to 200 MHz better than the launch 9900K. This marks the first time in over a decade where Intel decided it was worthwhile to do a revision for speed reasons, and the improvement was rather small.
It's interesting to me that Intel even bothered. To me this is a sign that 10 nm is still having problems, and they'll probably use this revision to do a 100-200 MHz speed bump across the Coffee Lake line up sometime soon.
I can't hear you, you didn't say 10nm. 14nm is dead to me :) 7nm at this point from AMD or bust until xmas. If Intel has a magical 10nm 10-12 core at AMD 12-16 core pricing maybe I'll bite, but AMD almost has be by default unless they massively mess up 7nm 12-16 core chips I'm after now.
If Intel still has broken chips the keep slowing down over time still, no point in NOT buying AMD, especially when they'll likely be beating Intel in almost every use case once AMD 7nm hits. Is it a fair contest then? Nope but that is Intel's fault for wasting 4-5yrs of 4B+ a year on mobile instead of the fab down the road in AZ (16-20B wasted on mobile, 10nm would be done AGES ago by Intel for that cash!). Nice work losing the fab race too losers. Can Intel recover before China owns all of semi production? I hope, but it's not looking good. Rather, it keeps looking worse. Intel needs to head to 7nm ASAP no matter the cost of tossing 10nm at this point. Better to take a hit and be back in the game quicker, than try to finally make something that is about to be OLD soon, work finally. Let's face it, Intel's 10nm isn't going to blow away TSMC 7nm. These two will be much closer when launched than previous gens.
Also note, Intel's 7nm is much less aggressive than the 10nm (bit off more than they could chew, so backing off for 7nm), so again, this will be about a tie probably for 5nm TSMC vs. 7nm Intel (weakened). Intel actually has to make a better chip now, as process won't help take down AMD it would seem for this next few rounds at least. Good for consumers, but overall, not a great time for USA fab wise. Still empty down the road.
Even with fab process improvements, if Intel can't come up with significant design improvements, then it is probably time for Intel to fully replace "Core" with an all new architecture. It's only been 20 years since Intel had to really come up with an all new design(Core came from the Pentium 3, which isn't even the first to use that design, it came from the old Pentium Pro as I recall). So, all these years, and Intel is still counting on an antique design that has been improved a LOT, but hasn't gotten a full replacement. Maybe Intel will go back to an updated Netburst and see if they can get it to 9GHz or something.
Intel has supposedly had a new uArch waiting, but it was designed around the 10 nm node. That said, I don't know if it's as big a change as you'd like.
In their recent architecture day, they announced that future architectures will not be so closely tied to a manufacturing node.
If you're going to armchair QB Intel's financial priorities, why not criticize their dividends? Maybe *that* money should've gone into their fabs, instead of investors' pockets.
Returning money to shareholders (who have trusted you with their cash) is not the same as pissing it away like Intel did. But yeah, you can kill the div too if desired, I'd always prefer more R&D PERIOD. I still stand by my point, wasted 16-20B and killed 10nm in the process it seems. You need a KING before you bother here (like dirk said before AMD fired him then tried to do what he said ~7-8yrs later...ROFL). You must protect your base/CORE products, or someone flanks you. You are seeing this now. Intel didn't properly address fabs, so chips will take a hit, followed by earnings drops, analysts downgrades, then more bailing, worse R*D, less engineers (layoffs), etc etc, until they become AMD...LOL. Sucks to lose fabs, hope they pull off something great at 5nm because 7i vs. 5t looks like another tie. Yeah, Intel needs new chips so they can take some time to pay their way back into the FAB race. It's not that they've lost (yet fully), just that now they have no fabvantage so to speak...LOL. Then again maybe AMD takes a huge chunk back (50/50 overall in 3yrs?, easily doable with no more dresden 20% limit, no tsmc, samsung or gf can make chips, though GF kind of out this round), and Intel pours more into fabs then gets back on track before REAL damage is perm. The best part is AMD will be out of debt if that happens with potentially a few billion to fight with in CASH. I'm hoping 64/128 core server chips force margins up high enough to bag a few billion before Intel gets something right again. Nothing says TSMC etc will be on a roll forever, so it can change pretty easily in one gen.
This is precisely why NV passed on console (and AMD should have too). It robs from CORE R&D as Jen noted. He almost go caught chasing mobile like Intel but moved to a different market with it. I really think he needs to come back to mobile now that gaming is big on mobile and finally make a REAL ARM desktop 500-1000w box with 2080ti etc as options! That will steal some sales as cheap as NV could sell these just to get it off the ground. As long as they break even on the chips, pump them out and sell cards in them. Xbox proves they could sell a ~450mm^2 soc for $115 or so (ps4 etc also). That would be a BIG chip compared to Intel correct? It would be an ARM powerhouse with decent igpu (think about Intel's, gpu side could be huge here). You could sell that chip for $200-225 and laugh at perf. It only takes 10mil to get devs to chase a console. 2-3yrs of selling powerful arm boxes that can add a discrete NV card later at peanut pricing (still with great margin) vs Intel would start a gaming revolution on ARM/Vulkan IMHO :) Not sure why they haven't started pouring 100mil a year into making 10-20 small dev games a year for this. The hardware is there for more complex games (NV wouldn't need to change much from current socs to go to a desktop). Yank the car specific crap and up everything else, run at 85w, and see what happens for a few hundred mill (peanuts to NV now). My guess is worst case you get your money back on the games and fail. Best case, the games draw a mass crowd that want those cheaper PC's by a MSFT lic + discount off Intel. A 450mm^2 ARM on 7nm at 85w would be VERY potent for $200-250 and no windows fee either. You are talking dual/tri boot boxes here if vendors desired also (linux, android, chromeos?, whatever). Very interesting and covering most moms/dads/internet basic users and great as a gaming box if you like what is out there, plus expandable as Unreal engine etc new games keep getting better and better by all Devs since tools from all sides are advancing so fast. Single people can make quite a compelling game today in a few years alone.
NV should be working with google to make a desktop OS for real. Google doesn't seem to be aiming at desktop OS stuff fast enough IMHO (at all yet?). Linux just doesn't seem to be getting off the gaming ground (that is needed, then apps, then pro apps at some point), and google can't seem to make a full OS or anything for more than a phone/facebook crap.
Why would new steppings alone of otherwise identical processors require BIOS updates? Is that really a thing? I frankly cannot recall any BIOS updates being issued in the past, let alone *announced* via a press release, for just new steppings.
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
60 Comments
Back to Article
Marlin1975 - Tuesday, March 26, 2019 - link
So this is 14++++++ or 14+++++++?And it adds what of real value/change? Higher power usage to keep up with AMDs new chips?
eddman - Tuesday, March 26, 2019 - link
Are you aware of what stepping actually is? As already stated in the article, they are to fix hardware-level erratas or improve the performance or efficiency or stability, or a combination of them all.Release of a new processor stepping is usually a non-event, unless it is meant to fix a major errata.
Some errata can be "fixed" with bios updates, but some need a new stepping.
Some examples:
https://www.theregister.co.uk/2017/04/27/intel_red...
https://techreport.com/news/13721/chip-problem-lim...
https://techreport.com/news/26911/errata-prompts-i...
goatfajitas - Tuesday, March 26, 2019 - link
I am prettttttttty sure he was being sarcastic.wut - Tuesday, March 26, 2019 - link
I am prettttttty sure he was just ignorant.Samus - Wednesday, March 27, 2019 - link
If it was sarcasm, my detector didn't go of. It just seemed stupid.LordSojar - Wednesday, March 27, 2019 - link
It's okay. Ignorance of an issue isn't bad, it's just lack of information about a given topic. Willful ignorance is the worst thing ever though. Hopefully, they learned a bit from this and are now wiser. Hopefully. A guy can hope, yeah?close - Wednesday, March 27, 2019 - link
I just wonder how much it costs these days to have an article published about a new CPU stepping (!) where you can't even see what's new (!!). I can't seem to find any other such article on AT over the past decade, and what is there is from back when a stepping made a world of difference in OC."Intel did not comment on the news story."
Of course not. They issued it.
"it is noteworthy that Intel plans to launch a new stepping of its Coffee Lake processors"
No it's not. Unless it's to keep Intel in the news with such fake articles that make it look like something's still moving at Intel.
In other news, motherboard manufacturer launched new mobo revision using slightly shinier capacitor next to one of the SATA ports. Significance of change still a mystery.
Santoval - Thursday, March 28, 2019 - link
"Unless it's to keep Intel in the news with such fake articles that make it look like something's still moving at Intel."While I won't comment on whether the article is fake or "paid", of course the idea is to "keep Intel in the news".They've run out of + to add to their 14nm node, CPUs based on their 10nm+ node are to be released in high volume (presumably) starting from December 2019 but only for the mainstream desktop market and below, so they've now resorted to issuing press releases for ... new steppings!
By the way, they are apparently canning their 10nm node after just a single 10nm based Core-i3 15W specimen release in 2018, largely for schools in China, which had much worse clocks than its 14nm equivalent *despite* a disabled iGPU. The sole purpose of that pathetic specimen was naturally to report a nominal 10nm 2018 release to their investors and avoid having their stock hammered.
If Intel pull back again the release of Ice Lake CPUs deep into 2020 they will be hurt badly. AMD is already set to have a 7nm node head start of ~6 months. Will they gift them even more? Of course we still don't know how their (fixed) 10nm+ node is going to perform, even if it is not delayed further.
shusse82 - Wednesday, March 27, 2019 - link
You can't really tell if someones being sarcastic in print unless they use exaggeration or <s></s>.Kevin G - Tuesday, March 26, 2019 - link
The benefits of a new stepping in the past is generally what transition of 14 nm to 14+ nm would bring: slight tweaks to the design to enable lower thermals or jump up a notch or two in clock speeds. There of course would be some errata fixes with each stepping.The last time Intel really pushed a new stepping was for the TSX instructions in Haswell which were removed via microcode updates on affected chips. Intel did eventually release some Haswell-EP chips with it enabled but those are harder find (when this bug was discovered, the EP line up had it disabled by default but BIOS could re-enable it for developers).
eddman - Tuesday, March 26, 2019 - link
Correct me if I'm wrong, but I was under the impression that stepping derived performance and/or efficiency improvements were the result of things like fixing slight manufacturing issues that, say, caused unnecessarily high voltages, etc.I've been reading that even the latest skylake processors couldn't match kaby lake in OC or efficiency, or perhaps I've missed something.
I'm not aware of the exact differences between 14 and 14+, but it's been reported that 14++ actually has a different transistor geometry and has a larger gate pitch.
MrSpadge - Tuesday, March 26, 2019 - link
You're right, 14 / 14+ / 14++ use slightly different transistor geometries. Which is a bigger change than just a new stepping.bobhumplick - Wednesday, March 27, 2019 - link
a stepping could fix a problem. like the security vulnerabilities etc. but im wondering if they are putting out a different die with the igpu cut off and some small improvements.maybe thats what the kf cpus were about. getting people used to the idea of igpu-less cpus before this new stepping\die came out. that would save a lot of capacity with the 14nm shortage going on. of course the current kf cpus still have the igpu but its disabled.
cutting off the 9900k's igpu would actually make it smaller than the 8700k and you could probably make it for the same price as the 8700k. i expect intels rumored 10 core cpus to not have an igpu. that would make them smaller than the 9900k.
are intel releasing a destkop chip this year? they have to right? so why release a new stepping this late in the year when a new release is coming? im wondering if the rumored 10 core will work on 300 series boards. not making a new chipset would save even more fab space and would give intel a win in the value department if we could use our z370 or z390 boards with the new 10 core. so maybe they will just release a new stepping of the 9th gen chips, lower the price of the 9900k, and then add the 10 core on top? possible?
mode_13h - Tuesday, March 26, 2019 - link
If anyone remembers the C3 stepping of AMD's Phenom II, that was pretty significant. Maybe on par with Skylake -> Kaby Lake.mode_13h - Tuesday, March 26, 2019 - link
I mean in terms of impact - not in terms of the scale and scope of the underlying changes.LiviuTM - Tuesday, March 26, 2019 - link
No, this time I'm sure it's 14##.BigMamaInHouse - Tuesday, March 26, 2019 - link
Maybe 95W TDP 10C/20T is coming?repoman27 - Tuesday, March 26, 2019 - link
That would be Comet Lake-S which is due in Q4.Stepping R0 is the rest of the Coffee Lake Refresh-S parts (65W, 35W) which should be available shortly (April). Anandtech already posted about these here: https://www.anandtech.com/show/14058/complete-coff...
Jimbo Jones - Tuesday, March 26, 2019 - link
Well first they need to make a 95w 8 core ... and don't tell me the 9900k or even the 9800k are "95w"The 9900k pulls the same all threads max power consumption as an AMD Threadripper 2950x - which has twice the number of cores. (each pull ~260w under prime95 stress test)
When you hold a 9900k to its 95w tdp - it performs worse than a 2700x in multi-threaded loads so you know there's something fishy there ...
The 9800k pulls over 200w and more or less the same as the Ryzen 2700x, but the Ryzen has 8 more threads and better overall multi-threaded performance.
I saw one of Intel's 14 core 9990x CPUs draw almost 800 watts when overclocked to 5.1 ... yikes!
Short answers: 1) Intel's power consumption is actually more than AMDs overall, but Intel assigns super low TDPs that aren't real (they don't consider ANY boost clocks at all) - its a trick to make people believe that they still have more efficient CPUs - they do not. 2) there's no way they can rectify this issue by any decent amount on 14nm - if this new stepping brings them to a 10 core, expect it to be a heater with an artificial TDP slapped on as we already see with current 9th gen.
When they get to 10 or 7nm though - whole new ballgame, but the competitor will be beating them to the node shrink it appears. :)
BigMamaInHouse - Wednesday, March 27, 2019 - link
I know- I was joking with "95W", same as "95W" 9900K :-).Jimbo Jones - Wednesday, March 27, 2019 - link
BigMama - someone needs to invent a sarcasm font, lol ...JlHADJOE - Wednesday, March 27, 2019 - link
> When you hold a 9900k to its 95w tdp - it performs worse than a 2700x in multi-threaded loads so you know there's something fishy there ...GamersNexus' test says otherwise:
https://www.gamersnexus.net/hwreviews/3378-intel-9...
" With the ASUS Maximus XI Hero that we used for our 9900K, a stricter adherence to Intel’s stock policies means we see a sharp drop-off in power consumption when testing under full stock conditions. The CPU falls to 100W load and stays there, leaving more performance available if we were to remove power targets and limits. Some of the other motherboards shipping today will exit these Intel power specs and draw more power. The 2700X pushes closer to about 120W-125W draw. Intel manages to achieve better overall combined throughput for the player and viewer while maintaining a lower power consumption, for which the 9900K deserves acclaim. It is, of course, significantly more expensive, and that is a big factor to play into our conclusion."
In their game streaming benchmarks the 9900k consumes less power while delivery better performance for both viewer and streamer compared to the 2700X.
Notmyusualid - Friday, March 29, 2019 - link
Joe - agreed.eddman - Wednesday, March 27, 2019 - link
Going past TDP is not inherently wrong for a desktop processor, if the cooling and power delivery are up to the task.It's not surprising that 9900k draws so much power at such extreme frequencies. Once you pass a certain threshold, efficiency goes right out the window and power goes up exponentially.
It's also not surprising that when limited to 95W, it performs worse than a 2700x. At 95W the 9900k stays at its base 3.6 GHz clock speed when fully loaded, while 2700x is free to draw as much as ~105W and reach a higher frequency. It might not look like much, but that 10W is quite enough to make a difference. Nothing fishy here. A 9900k limited to 105W should at least be able to match its performance.
Same story for 9800k's high power draw.
If ryzen CPUs could reach 5.0 GHz, you'd see just as high, if not higher, power consumption numbers.
To properly test and compare the efficiency of two processors' architecture/node, you should run them at the same clock speeds, under the same conditions, to get the real picture.
As a side example, Ian managed to achieve 4.5 GHz with a 113W power draw.
https://images.anandtech.com/doci/13400/9900K%20OC...
Yes, he did choose the voltages manually (instead of using the Auto mode) which resulted in lower numbers but it does show the CPU is capable of running efficiently. Also, keep in mind that the Auto mode of some motherboards is too aggressive and choose unnecessarily high voltages.
Another interesting thing to note in that table; going from 4.5 to 5.0, an 11% increase, resulted in a 70% increase in power.
eddman - Wednesday, March 27, 2019 - link
It seems 2700x can go well beyond 105W depending on the board; comparing a 95W-locked 9900k against it seems to be even less appropriate than I thought.eek2121 - Tuesday, March 26, 2019 - link
My money is on this being a silent update to compete with zen 2.bobhumplick - Wednesday, March 27, 2019 - link
wow 14++++++++++ i havent heard that a million frickin time. honestly. thats the first time ive heard that jokeIronchef3500 - Wednesday, March 27, 2019 - link
HeheSahrin - Wednesday, March 27, 2019 - link
A stepping is a change in the mask set, not the manufacturing process. (Same process building slightly different design).Santoval - Thursday, March 28, 2019 - link
A new stepping is not a new node. It is just a slight or (rarely) moderate fix or "clean up" of the exact same node. The size of the transistors, cells and of the BEOL layers are never modified with a new stepping. If you do more than fix or clean up the node, but less than developing a true successor node (in order to increase transistor density and reduce die sizes), that's when you usually add symbols like + to create something halfway between a new stepping and a truly new node. Which is what Intel have been doing since 2014.TristanSDX - Tuesday, March 26, 2019 - link
If these changes require new BIOS and they are silent, then some big hole has been patchedCellar Door - Tuesday, March 26, 2019 - link
Could be silicon that is completely lacking igpu(not just disabled) or like you said spectre etc related fixes.repoman27 - Tuesday, March 26, 2019 - link
Stepping R0 has been plan of record since at least August of 2018. It was on Intel's slide for 16 Gb DDR4 device support: https://benchlife.info/wp-content/uploads/2018/10/...Intel only released a handful of Coffee Lake Refresh SKUs in October. The rest of the S stack and all of the H parts are still forthcoming. The second round of Coffee Lake Refresh is stepping R0.
brakdoo - Tuesday, March 26, 2019 - link
Comet Lake? Just a stepping?GreenReaper - Tuesday, March 26, 2019 - link
Tom's Hardware reported that the 9900KF had the R0 stepping:https://www.tomshardware.co.uk/intel-9900kf-vs-990...
Maybe they want to allow it to soak up more power so it can clock faster - possibly through otherwise-used-for-GPU pins that would normally have different tolerances? Just a thought.
PeachNCream - Tuesday, March 26, 2019 - link
"...as it gets increasingly hard to make any alterations..."I'd suggest "harder" in this statement.
mode_13h - Tuesday, March 26, 2019 - link
No, "increasingly hard" seems right.mode_13h - Tuesday, March 26, 2019 - link
To me, increasingly harder just seems redundant. Harder already implies an increase, whereas "increasingly hard" says that the level of difficulty is on an upward trend line, but without being redundant.Targon - Tuesday, March 26, 2019 - link
It may be a partial fix for Spoiler, or more Meltdown mitigations.EphraimB - Tuesday, March 26, 2019 - link
I'm planning on getting an Intel Core i9-9900k for my custom build desktop tower. Should I wait for the new stepping or should I just buy it now? How will I be able to differentiate between the two?mode_13h - Tuesday, March 26, 2019 - link
Depends on how badly you need it. If there's a real cost to waiting, then I would just pull the trigger. Most often, steppings aren't huge changes.But, if you can afford to wait, then why not?
Personally, I'm waiting to see how 7 nm Ryzen turns out. If not for that and the high price point of the i9-9900K, I'd probably be right with you.
patel21 - Wednesday, March 27, 2019 - link
If you can, then please wait for Ryzen 3 chips to land. Even if you want to buy 9900K, as it will have an impact on the prices of intel chips.eddman - Wednesday, March 27, 2019 - link
We've had ryzen 3 since 2017.Targon - Wednesday, March 27, 2019 - link
I believe that patel21 was referring to Ryzen 3rd generation chips, which are expected to have very similar clock speeds to the 9900k for less money and possibly better real world performance, even in single threaded applications.eddman - Thursday, March 28, 2019 - link
I'm aware of all that. I was making fun of wrongly calling ryzen 3rd gen or 3000 as "3".Targon - Wednesday, March 27, 2019 - link
I'd wait until June and verify what the Ryzen 3700X specs are first. If the 3700X is 4.2/5.0 for clock speeds for 12 cores/24 threads and you pay $330 for it, that would probably be a better buy for you.benzosaurus - Tuesday, March 26, 2019 - link
Cool, a new Sky Lake stepping!mode_13h - Tuesday, March 26, 2019 - link
BITD, Intel would've referred to each architecture with the same marketing name, irrespective of the process technology changes. So, in that sense, Skylake would logically group together with everything since.But, it's not really accurate to group the 14++ nm processors with Skylake, in any strict sense. Even on a per-core basis, Coffee Lake is more than simply an overclocked Skylake. So, if you call them both Skylake, then you should probably also refer to Ivy Bridge as Sandybridge and Broadwell as Haswell.
garbagedisposal - Wednesday, March 27, 2019 - link
Those chips in your last sentence were different architectures and process nodes. It is not the same situation at all.Intel 6th, 7th, 8th, 9th gen is all the *same* core. Same architecture. Same IPC. Same memory controllers (new chips still support DDR3 with the right board). Same graphics (except updated video decode in 7th gen). The 14+++ improvements are about what you would expect after having multiple years to tweak the process for variants of the same exact architecture. He is not wrong to call it an overclocked skylake.
eddman - Wednesday, March 27, 2019 - link
"He is not wrong to call it an overclocked skylake."CFL might not be different from its predecessors from a microarchitecture standpoint, but it's not an OCed skylake either.
mode_13h - Wednesday, March 27, 2019 - link
According to Intel's "Tick-Tock" strategy, the first -Bridge and first -Well were the new uArch points. Sure, you change a few things during a node-shrink - it's not strictly about clock speed and power as you also want to take advantage of more die area and greater timing flexibility. But, in Intel's own words, they were not new architectures. And I think probably not such a difference as you would find between the different generations of Pentium 4 chips, for instance.althaz - Tuesday, March 26, 2019 - link
Fingers crossed this setting thins out the silicon a bit. The 9700k and 9900k are 10+ degrees hotter than they should be primarily because of how stupidly thick the piece of silicon is. If you delid one of these guys and grind it down you'll gain 10 degrees. Pretty annoying that Intel finally started soldering their CPUs again but they almost doubled the z-height of the silicon at the same time.Khenglish - Wednesday, March 27, 2019 - link
I feel like people are confused on what a stepping and revision are.In the past 12+ years ago both Intel and AMD would regularly come out with new steppings throughout the product cycle of a processor. The stepping would primarily be a retiming improvement of the processor to achieve more speed. For example the P4 130 nm (Northwood) launched at 2.2 GHz and typically overclocked to 2.6 GHz, but the final revision ran at 3.4 GHz stock, and overclocked to 4 GHz. The big increase in speed was just the engineers over time improving the layout of the same chip for higher speed.
AFAIK revisions after launch died out during the C2D era. The 65 nm C2D had a major revision (G0 I think), which boosted clocks around 10% over the launch revision. The 45 nm C2D had a similar 10% late clock boost revision, but this also increased L2 cache latencies, so the performance improvement was smaller.
What you ended up with was processors launched in a ever more optimized state, making it not worth the while of Intel and AMD to produce new expensive mask sets for very minor improvements.
When you look at this new R0 revision, according to Tom's review of the 9900KF R0, it clocks 100 to 200 MHz better than the launch 9900K. This marks the first time in over a decade where Intel decided it was worthwhile to do a revision for speed reasons, and the improvement was rather small.
It's interesting to me that Intel even bothered. To me this is a sign that 10 nm is still having problems, and they'll probably use this revision to do a 100-200 MHz speed bump across the Coffee Lake line up sometime soon.
TheJian - Wednesday, March 27, 2019 - link
I can't hear you, you didn't say 10nm. 14nm is dead to me :) 7nm at this point from AMD or bust until xmas. If Intel has a magical 10nm 10-12 core at AMD 12-16 core pricing maybe I'll bite, but AMD almost has be by default unless they massively mess up 7nm 12-16 core chips I'm after now.If Intel still has broken chips the keep slowing down over time still, no point in NOT buying AMD, especially when they'll likely be beating Intel in almost every use case once AMD 7nm hits. Is it a fair contest then? Nope but that is Intel's fault for wasting 4-5yrs of 4B+ a year on mobile instead of the fab down the road in AZ (16-20B wasted on mobile, 10nm would be done AGES ago by Intel for that cash!). Nice work losing the fab race too losers. Can Intel recover before China owns all of semi production? I hope, but it's not looking good. Rather, it keeps looking worse. Intel needs to head to 7nm ASAP no matter the cost of tossing 10nm at this point. Better to take a hit and be back in the game quicker, than try to finally make something that is about to be OLD soon, work finally. Let's face it, Intel's 10nm isn't going to blow away TSMC 7nm. These two will be much closer when launched than previous gens.
Also note, Intel's 7nm is much less aggressive than the 10nm (bit off more than they could chew, so backing off for 7nm), so again, this will be about a tie probably for 5nm TSMC vs. 7nm Intel (weakened). Intel actually has to make a better chip now, as process won't help take down AMD it would seem for this next few rounds at least. Good for consumers, but overall, not a great time for USA fab wise. Still empty down the road.
Targon - Wednesday, March 27, 2019 - link
Even with fab process improvements, if Intel can't come up with significant design improvements, then it is probably time for Intel to fully replace "Core" with an all new architecture. It's only been 20 years since Intel had to really come up with an all new design(Core came from the Pentium 3, which isn't even the first to use that design, it came from the old Pentium Pro as I recall). So, all these years, and Intel is still counting on an antique design that has been improved a LOT, but hasn't gotten a full replacement. Maybe Intel will go back to an updated Netburst and see if they can get it to 9GHz or something.mode_13h - Wednesday, March 27, 2019 - link
Intel has supposedly had a new uArch waiting, but it was designed around the 10 nm node. That said, I don't know if it's as big a change as you'd like.In their recent architecture day, they announced that future architectures will not be so closely tied to a manufacturing node.
mode_13h - Wednesday, March 27, 2019 - link
If you're going to armchair QB Intel's financial priorities, why not criticize their dividends? Maybe *that* money should've gone into their fabs, instead of investors' pockets.TheJian - Sunday, April 7, 2019 - link
Returning money to shareholders (who have trusted you with their cash) is not the same as pissing it away like Intel did. But yeah, you can kill the div too if desired, I'd always prefer more R&D PERIOD. I still stand by my point, wasted 16-20B and killed 10nm in the process it seems. You need a KING before you bother here (like dirk said before AMD fired him then tried to do what he said ~7-8yrs later...ROFL). You must protect your base/CORE products, or someone flanks you. You are seeing this now. Intel didn't properly address fabs, so chips will take a hit, followed by earnings drops, analysts downgrades, then more bailing, worse R*D, less engineers (layoffs), etc etc, until they become AMD...LOL. Sucks to lose fabs, hope they pull off something great at 5nm because 7i vs. 5t looks like another tie. Yeah, Intel needs new chips so they can take some time to pay their way back into the FAB race. It's not that they've lost (yet fully), just that now they have no fabvantage so to speak...LOL. Then again maybe AMD takes a huge chunk back (50/50 overall in 3yrs?, easily doable with no more dresden 20% limit, no tsmc, samsung or gf can make chips, though GF kind of out this round), and Intel pours more into fabs then gets back on track before REAL damage is perm. The best part is AMD will be out of debt if that happens with potentially a few billion to fight with in CASH. I'm hoping 64/128 core server chips force margins up high enough to bag a few billion before Intel gets something right again. Nothing says TSMC etc will be on a roll forever, so it can change pretty easily in one gen.This is precisely why NV passed on console (and AMD should have too). It robs from CORE R&D as Jen noted. He almost go caught chasing mobile like Intel but moved to a different market with it. I really think he needs to come back to mobile now that gaming is big on mobile and finally make a REAL ARM desktop 500-1000w box with 2080ti etc as options! That will steal some sales as cheap as NV could sell these just to get it off the ground. As long as they break even on the chips, pump them out and sell cards in them. Xbox proves they could sell a ~450mm^2 soc for $115 or so (ps4 etc also). That would be a BIG chip compared to Intel correct? It would be an ARM powerhouse with decent igpu (think about Intel's, gpu side could be huge here). You could sell that chip for $200-225 and laugh at perf. It only takes 10mil to get devs to chase a console. 2-3yrs of selling powerful arm boxes that can add a discrete NV card later at peanut pricing (still with great margin) vs Intel would start a gaming revolution on ARM/Vulkan IMHO :) Not sure why they haven't started pouring 100mil a year into making 10-20 small dev games a year for this. The hardware is there for more complex games (NV wouldn't need to change much from current socs to go to a desktop). Yank the car specific crap and up everything else, run at 85w, and see what happens for a few hundred mill (peanuts to NV now). My guess is worst case you get your money back on the games and fail. Best case, the games draw a mass crowd that want those cheaper PC's by a MSFT lic + discount off Intel. A 450mm^2 ARM on 7nm at 85w would be VERY potent for $200-250 and no windows fee either. You are talking dual/tri boot boxes here if vendors desired also (linux, android, chromeos?, whatever). Very interesting and covering most moms/dads/internet basic users and great as a gaming box if you like what is out there, plus expandable as Unreal engine etc new games keep getting better and better by all Devs since tools from all sides are advancing so fast. Single people can make quite a compelling game today in a few years alone.
NV should be working with google to make a desktop OS for real. Google doesn't seem to be aiming at desktop OS stuff fast enough IMHO (at all yet?). Linux just doesn't seem to be getting off the gaming ground (that is needed, then apps, then pro apps at some point), and google can't seem to make a full OS or anything for more than a phone/facebook crap.
Santoval - Thursday, March 28, 2019 - link
Why would new steppings alone of otherwise identical processors require BIOS updates? Is that really a thing? I frankly cannot recall any BIOS updates being issued in the past, let alone *announced* via a press release, for just new steppings.Khenglish - Thursday, March 28, 2019 - link
The BIOS update is just adding microcode updates. Each processor revision has its own microcode version. This is normal.It would be interesting to see if R0 could run without a microcode update though. The last Intel CPU I am aware of that could was Sandy Bridge.