I can say the same about your comment.Can't tell if that is sarcasm or not. I really hope it is.
And you're basing this conclusion just from the fact that sgemm/dgemm performance seems to be good when enforcing AVX2? You haven't really addressed the situation involving the inability to bypass CPUID checks, which afaik is the case when using commercial software packages like MATLAB and Mathematica.You're justifying MKL running poorly on non-Intel systems because they developed the software. From a legal standpoint it holds up just fine and even from an ethical standpoint, nobody should ask Intel to do differently were there intrinsic optimisations that needed to be done to meaningfully extract more performance. But that's not the case, and the post you just linked shows that perfectly. There were significant gains in performance just by enabling the use of AVX2.
And you're basing this conclusion just from the fact that sgemm/dgemm performance seems to be good when enforcing AVX2? You haven't really addressed the situation involving the inability to bypass CPUID checks, which afaik is the case when using commercial software packages like MATLAB and Mathematica.
Which ties back to the original superfluous point made by @DrMrLordX stating that nobody in their right mind would use a Rocket Lake CPU over a Zen 3 CPU for workstation purposes. The point I'm trying to make is that simply because Intel MKL exists and is an inseparable part of many applications, the fact that it requires a CPUID check to function properly means that Intel CPUs are still viable for those particular use-cases, despite the clamoring from the pro-AMD club that Rocket Lake is going to be a useless product.What's there to discuss? The fact that Intel specifically created this situation to try their best to lock out even the possibility of running AVX2 code on Ryzen completely ignoring the far easier solution of just asking Windows what the processor and OS can support?
What do you want me to say? "Oh Intel shouldn't have to spend any resources to implement a basic check that could be done by half decent programmer in maybe 10 minutes (if I'm being really generous) because they have no responsibility trying to make it run well on their competitor's hardware. In fact - good on Intel for locking out good performance on their competitor's platform, bravo the industry can really learn from you"
Which ties back to the original superfluous point made by @DrMrLordX stating that nobody in their right mind would use a Rocket Lake CPU over a Zen 3 CPU for workstation purposes. The point I'm trying to make is that simply because Intel MKL exists and is an inseparable part of many applications, the fact that it requires a CPUID check to function properly means that Intel CPUs are still viable for those particular use-cases, despite the clamoring from the pro-AMD club that Rocket Lake is going to be a useless product.
Why should Intel ensure that their software also runs well on AMD hardware? What incentive do they have to make that happen? This is different from the CPUID checking Intel compiler shenanigans. Intel spends their own resources to develop high-performance libraries for their own products. They've got no reason to bother optimizing for other products.
Your bias is apparent. You've pronounced something as worthless even before it has released.Stop trying to bring it back to a point I never argued against. I chimed in to address this comment made by you:
Not to discuss whether or not one should buy an AMD or an Intel system for use in Matlab. As things stand - you are basically locked into using an Intel system due to artificial limitations Intel themselves placed to try and prevent Ryzen systems from performing reasonably. It's not a case of Intel having to spend resources trying to make it perform better on Ryzen, they made a decision to lock down MKL and that's it.
Also, for crying out loud what's with the personal attacks you feel you have to make.
Rocket Lake is a worthless product. It's made worthless both by the real TGL-H and ADL-S.
Buying RKL-S is even worse than buying Kaby Lake was, because it's no secret that Alder Lake is coming soon and will be vastly improved in every way. Unless you want AVX512 to work out of the box, because of the whole issue with the littles on Alder Lake.
My bias? Rocket Lake hits the market in March. It has less than 9 months on the market before it's replaced by something better in every way possible, because Alder Lake launches by the end of 2021.Your bias is apparent. You've pronounced something as worthless even before it has released.
Pretty sure it's still monolithic, die size won't be much larger than Tiger Lake 8 cores releasing in half a year. The other innovation beside new architecture is small and big cores.Will Alder Lake be a chiplet design, or still monolithic?
Pretty sure it's still monolithic, die size won't be much larger than Tiger Lake 8 cores releasing in half a year. The other innovation beside new architecture is small and big cores.
Maybe next generation they'll finally push for chiplets (so 16 big cores) or something fancy like multi-layer logic, similar to Lakefield but more desktop oriented with cores count.
Never miss a chance do you?
Can't tell if that is sarcasm or not. I really hope it is.
Why should Intel ensure that their software also runs well on AMD hardware?
What's there to discuss? The fact that Intel specifically created this situation to try their best to lock out even the possibility of running AVX2 code on Ryzen completely ignoring the far easier solution of just asking Windows what the processor and OS can support?
Intel to begin mass production of 11th Gen Core Rocket Lake-S CPUs in January - VideoCardz.com
Intel Rocket Lake mass production starts in January Intel is set to begin mass production of its upcoming 11th Gen desktop processor series in January, a new leak from @OneRaichu seems to confirm. We do not know how recent this roadmap is, but it does appear to be a real roadmap. Intel has begun...videocardz.com
Appears that Rocket Lake's launch may still be in January but will be a megapaper launch.
Someone remind me, when is the first big OEM going to ship Zen 3 again?
Don't know where you got that from. Videocardz catches a lot of interesting tweets, but injects much of its own editorializing.
Only boutique builders - the likes of PCSpecialist and similar companies overseas - actually try to use regular Zen CPUs. Most major OEMs prefer to stick with APUs for the iGPU. For gaming systems you can get away with iGPU-less CPUs, for anything non-gaming Cezanne is the one to wait for.
Whenever AMD launches AM4 Cezanne.
Only boutique builders - the likes of PCSpecialist and similar companies overseas - actually try to use regular Zen CPUs. Most major OEMs prefer to stick with APUs for the iGPU. For gaming systems you can get away with iGPU-less CPUs, for anything non-gaming Cezanne is the one to wait for.
Also you should probably wait and see if Rocket Lake has meaningful volume in February. From what I've heard, I'm very, very doubtful.
I remember new CPUs being introduced back in the 80s and 90s, the last generation almost instantly became junk. Wait 2 generations for an upgrade and your old $2000 PC was relegated to a job as a doorstop or footstool.
Now we wait 5 years and argue endlessly about minutia over a ~20% IPC gain. And we don't even get the full 20%, because the new chips can't clock as high as the old chips. Edit And, that's not even areal IPC increase, it's mostly due to cache re-work. To wit, on TPU's 1080P aggregate performance with a 2080 Ti, the difference between a 5800X and an i3-10300 is 5.4%. That is not noticeable to most humans, and all of it can be attributed to the clock speed difference between those 2 chips (i3 @ 4.4Ghz single turbo vs 5800X at 4.7Ghz single turbo).
It's real clear to me that the future of desktop performance increases is going to be just like it has been for mobile phone SoC's. Specialized circuitry with developer libraries & compilers to make use of them for common algorithms. Apple gets that, and Intel is quickly moving that way.
General purpose compute is pretty much done, it's a very 1990s concept at this point.
So, it's not about hardware but violates the cross-licensing agreement? LOLThere really isn't anything to discuss, and I wasn't the one who brought up MKL as if to say that it was somehow a hardware advantage for Intel's Rocket Lake-S (it isn't).
I started writing this reply in disagreement only to realize mid-sentence that the math checks out. The jump from KBL to CFL in terms of perf/dollar was so massive I was almost tempted to think ADL-S could not reproduce it, when in fact it has all the ingredients to do so.Buying RKL-S is even worse than buying Kaby Lake was, because it's no secret that Alder Lake is coming soon and will be vastly improved in every way.
Exactly.I started writing this reply in disagreement only to realize mid-sentence that the math checks out. The jump from KBL to CFL in terms of perf/dollar was so massive I was almost tempted to think ADL-S could not reproduce it, when in fact it has all the ingredients to do so.
All true - if waiting for Alder is possible, buying Rocket Lake makes no sense. Still, if you had to buy a PC in H1 2021, I don't see how Zen 3 is much different and why Rocket would have to be terrible, as long as you don't need the MT performance that the 5900X and 5950X offer. 11700KF will probably have good performance and performance/$ in any workload and especially gaming. Budget stuff like the 11400 should be good too, especially now that Intel won't gimp memory OCs on budget boards. AVX-512, as niche as it is, is also a plus for some.Exactly.
ADL-S will bring significant 1T perf uplift, vastly improved mT and improved power efficiency all at the same time with the only downsides being that DDR5 will probably be expensive at launch. And for all of those upsides, it's actually a larger improvement than Kaby -> Coffee was. TGL-H35 has a 5GHz 1T SKU, by the time Alder launches I'd expect Intel to be pushing 5.1GHz+ again, so it's not even like there's any clock degredation either.
You're looking at a straight 15%+ 1T improvement and much more in mT.
So like I said before, buying Rocket Lake is like buying Kaby Lake once was. Maybe even worse.
Obviously, RL will come down to pricing. If they get a mid double digit IPC increase, and price it correctly it will be a decent product. Too bad they did not get it out earlier, as there would be a good opportunity to move product now, since AMD doesnt really have "budget" 5xxx series products yet, and supply is terrible. (Dont think I have ever seen 5xxx series in stock at Microcenter, although I am not in the market and only check occasionally.)All true - if waiting for Alder is possible, buying Rocket Lake makes no sense. Still, if you had to buy a PC in H1 2021, I don't see how Zen 3 is much different and why Rocket would have to be terrible, as long as you don't need the MT performance that the 5900X and 5950X offer. 11700KF will probably have good performance and performance/$ in any workload and especially gaming. Budget stuff like the 11400 should be good too, especially now that Intel won't gimp memory OCs on budget boards. AVX-512, as niche as it is, is also a plus for some.
Those must be quad core though. Single thread looks great, but they desperately need the 8 core products, even for a gaming laptop.New benchmarks for Tiger Lake H showing up on Geekbench. The single thread scores are close to 5950X, the multi-core scores are in spitting distance of a non-K 10600 stock. A power unlocked 10400 can still beat this, marginally.
View attachment 36387
By comparison, to the 6 core 4600H:
View attachment 36388
View attachment 36389
And the 6 core 10750H :
View attachment 36395
View attachment 36396
And the 3300X, closest thing we have to a 4-core Zen part :
View attachment 36393
View attachment 36394
5950X single core, current single core champ :
View attachment 36397
Those must be quad core though. Single thread looks great, but they desperately need the 8 core products, even for a gaming laptop.