Those numbers make no sense. Is 14900K ~17% faster than 14700K ? Also the 14600K is 6% faster than 14700K which is absurd.In gaming it s just 3% below the 14900K and 14% faster than the 14700K, FTR 21 games are tested.
Why ST negates possible advantage in private caches? Also I am not exactly sure which Intel generation you mean specifically as Skylake was at cache disadvantage and Raptor Lake at advantage. To truly judge what is the reason we would need someone to do a profiling of R23 on specific architectures to compare what the performance counters say.CB 2024 is no more using Embree, it use Maxon s renderer, so it s not comparable to R23, one more time, how did Intel gain more than 10% in ST from R15 to R23, the fact that we re talking of ST eliminate the cache possibility
You are rendering tile by tile, and tiles mostly do not share data, since on Zen the L3 is a victim cache it is only fed the data that were discarded from L2, but if you do not render the same tile twice, from data side it will not give you large advantage. Now from instruction side it might be able to contain the whole program, but if there is no gain from x3d in Zen4 vs Zen4x3d that simply means the program footprint was small enough to fit in L3 of vanilla Zen4. Also for ST test, Zen4x3d will be at frequency disadvantage.and the X3D is no better than the regular chip
Until we know the option used for ICC it's really hard to say. Y-Cruncher for a long time was also ICC compiled and there Zen4 did not have issues with ranking high. So while ICC is historically controversial compiler it's usage doesn't have to equate the Zen -50%.because Cinebench R20/23 are ICC compiled and are the only renderering tests where the 12900K is ahead of the 5950X, actualy it s about the only Computerbase MT benches where it was ahead.
Seriously? Who even uses ICC anymore?Cinebench R20/23 are ICC compiled
NBC published their review of the 9700X.
Overall very strong ST perf, FI in GB 5.0 and 5.5, Kracken, WebXPRT 3.
In gaming it s just 3% below the 14900K and 14% faster than the 14700K, FTR 21 games are tested.
The only drawback according to NBC is the strong competition price wise by AMD s own previous gen, much less by Intel given their current degradation problems.
From CB R15 to CB R20 Intel got a 10% ST uplift in respect of AMD, and another few % with R23 that was released just one year after R20, check in the review i just linked, there s CB 11.5 as well as R15, R20 and R23.
In the Maxon video i linked we can see that they endorsed Intel surely in exchange of all their software suite for free, guess that there must be some pay back as a consequence
I mean by selecting a convenient scene, because all renderings using a same soft wont yield the same results between 2 CPU as it depend of the exact arithmetic ops distributions that are used, and independently of the fact that R15, R20 and R23 all use Embree wich is Intel s in house renderer and wich is also used by Corona, now compare those renderers scores to say Blender and Vray.
Maybe the AIO cooler is helping? I think others did their tests with air coolers.Gaming performance on par with 14900k? Why's there discrepancy between NBC and other review outlets like HUB?
http://www.numberworld.org/y-cruncher/news/2024.html#2024_5_10 at least for Y-cruncher there are regressions from switching to ICX from ICC. Still it (clang) is my compiler of choice lately.I don't get why people don't simply use Clang/GCC. It's not like they're some amateur hobby projects...
They don't care since their real cash cow Turin isn't out yet.And AMD is extremely poor here seeing upstream CLANG still doesn't have Zen5 target...
Clearly, Mike Clark works in a wrong department. Rather than Chief Architect he would do better at marketing, as he fooled us all At very least they should rename his position to something like Chief MarketectI must say I am not impressed with Mike Clark between his lousy answer on the decoders and then later telling Dr. Cutress saying how software will catch up and the Zen 6/7 guys will get credit for the work Zen 5 did.
Intel is usually year ahead of the cash cow launch. If you want to have software that is using your nice and dandy features you better make it possible to target it. Especially if you claim the software might benefit from recompilation... But AMD instead is sending misleading patches to GCC and ignoring CLANG completely. How nice software company it isThey don't care since their real cash cow Turin isn't out yet.
Why ST negates possible advantage in private caches? Also I am not exactly sure
You are rendering tile by tile, and tiles mostly do not share data, since on Zen the L3 is a victim cache it is only fed the data that were discarded from L2, but if you do not render the same tile twice, from data side it will not give you large advantage. Now from instruction side it might be able to contain the whole program, but if there is no gain from x3d in Zen4 vs Zen4x3d that simply means the program footprint was small enough to fit in L3 of vanilla Zen4. Also for ST test, Zen4x3d will be at frequency disadvantage.
Until we know the option used for ICC it's really hard to say. Y-Cruncher for a long time was also ICC compiled and there Zen4 did not have issues with ranking high. So while ICC is historically controversial compiler it's usage doesn't have to equate the Zen -50%.
As to why 12900k is ahead of 5950x once again I would say we would need to have a nice profiling traces to be able to tell. I have seen people jump to conclusions only to be proven wrong by the actual measurements. While it's not a satisfactory answer most of the time, it's the correct one. That is why in software world when you listen to performance focused talks they will always repeat ad nausea "measure, measure". The CPU execution pipeline is so complex that it's really hard to make correct assumptions. It's another thing that it is also hard to measure correctly
Gaming performance on par with 14900k? Why's there discrepancy between NBC and other review outlets like HUB?
Spec_FP usage of SIMD depends on the quality of the compiler's autovectorizer and the instruction set you allow it to use. David Huang, Anandtech and Geekerwan are all using different settings here. You cannot therefore assume that it will translate to CB because CB is not making use of anything more than SSE 4.2. Not to mention Cinebench instruction and data flow might be different that SPEC average. Maybe you should try to find spec subset that correlates best.All tiles are not equal quantity of computations wise, that s a given, beside CB use hardly more than SSE 4.2, so it should have about the same improvement as what is measured by Spec_FP, yet the ST improvement for Zen 5 is only 10-12% fo CB R20/R23 while it s 15% for CB 2024.
Nah, just profile first. Analyze and then if something smells fishy show the traces with argumentation why they are fishyIt s impossible to know without being in the know...
Zen3 was also doing quite well against contemporary Skylake derivatives.As for y-Cruncher this soft use AVX512 and Zen 4 half rate AVX512 will still have a big impact even if ICC compiled.
This is probably just a not so eloquent figure of speech, what he meant is that with Zen CPUs there was less of a gap (or bigger advantage) vis a vis Intel for more SIMD-heavy calculations (I guess no one really did cb24 with ln2 on intel properly on hwbot, or maybe the gap was narrowed somehow) :Once again if you claim with 100% certainty the benchmark is flawed towards Intel, it would be nice if you could back this up with some relevant metrics
I'm not sure I follow you. You can't expect two different benchmarks to get the same speedup. In particular when one is the geomean of several tests with different profiles and can make use of AVX2/AVX-512 depending on how it's compiled.All tiles are not equal quantity of computations wise, that s a given, beside CB use hardly more than SSE 4.2, so it should have about the same improvement as what is measured by Spec_FP, yet the ST improvement for Zen 5 is only 10-12% fo CB R20/R23 while it s 15% for CB 2024.
Spec_FP usage of SIMD depends on the quality of the compiler's autovectorizer and the instruction set you allow it to use. David Huang, Anandtech and Geekerwan are all using different settings here. You cannot therefore assume that it will translate to CB because CB is not making use of anything more than SSE 4.2. Not to mention Cinebench instruction and data flow might be different that SPEC average. Maybe you should try to find spec subset that correlates best.
Nah, just profile first. Analyze and then if something smells fishy show the traces with argumentation why they are fishy
Zen3 was also doing quite well against contemporary Skylake derivatives.
And care to explain how exactly its flawed in Intel's favour? Here is R24 review by C&C https://chipsandcheese.com/2023/10/22/cinebench-2024-reviewing-the-benchmark/ it contains profiling data, while not exactly related to Zen5 or RaptorLake it should be able to help you along the way.
Btw, for people worrying about Core to Core latency regression and it's effect on performance:
Exploring CPU Core to Core Latency and the Role that Locks Play
This article has been a LONG time coming since our article on Rocket Lake, where we talked about core to core latency for the first time here on Chips and Cheese. This is a follow up article exclus…chipsandcheese.com
NBC isn't really the go to Website for gaming tests. 3% slower than 14900K and 14% faster than 14700K is already impossible, normally there isn't such a gap between 14900K and 14700K.NBC published their review of the 9700X.
Overall very strong ST perf, FI in GB 5.0 and 5.5, Kracken, WebXPRT 3.
In gaming it s just 3% below the 14900K and 14% faster than the 14700K, FTR 21 games are tested.
It’s not his fault they woke him up from his dream too early.Clearly, Mike Clark works in a wrong department. Rather than Chief Architect he would do better at marketing, as he fooled us all At very least they should rename his position to something like Chief Marketect
NBC isn't really the go to Website for gaming tests. 3% slower than 14900K and 14% faster than 14700K is already impossible, normally there isn't such a gap between 14900K and 14700K.
That s an argument thrown out of total ignorance, those e cores only flip the marketing driven core count, in Blender, Corona, Vray, Mental ray and any other renderer a 7950X is largely ahead, and even in CB R15, the only exception is precisely CB R20/R23 and of course the AVX2 crippling Povray that produce the same result as R20/23 -3%.are E / compact cores actually worth it in real world? or just bench number crunching to ignite flamewars in threads like this 😂
For gaming, definitely. For office/productivity, I'm not so sure. For battery life, I would imagine that two CCXs is better.unpopular opinion HX 370 probably would be better off with 8 full zen5 instead of 4 full 8c
hi everyone, i have a ryzen 9 9950x
with unlocked PPT it reaches 280 watts with 95 degrees, i saw in the post of @igor_kavinski
that it is much colder. what am i doing wrong?
sorry I didn't understand, can you explain it to me?Listening to that guy.
You don't need to worry. The colder temps were due to custom watercooling. If you touched that water, your finger would probably get frostbitesorry I didn't understand, can you explain it to me?