It has got noting to do with Cinebench being FP and SPECint being INT, just data which is available if you try to find it.so explain why a FPU based workload , corelates with integer based workloads to the point of it invalidating a consistent rumour of integer performance.
| Cinebench R23 | SPECint 2017 | |
| Zen+ -> Zen 2 | 19.3% | 16.8% |
| Zen 2 -> Zen 3 | 14.9% | 19.7% |
| Zen 3 -> Zen 4 | 5.1% | 5.6% |
| Cinebench R23 | SPECint 2017 | |
| Comet Lake -> Cypress Cove | 18.1% | 17.9% |
| Cypress Cove -> Golden Cove | 19.8% | 18.5% |
| Golden Cove -> Raptor Cove | 3.6% | 2.9% |
It has got noting to do with Cinebench being FP and SPECint being INT, just data which is available if you try to find it.
All SPEC data is from Anandtech reviews; all Cinebench R23 data is from here.
AMD
Cinebench R23 SPECint 2017 Zen+ -> Zen 2 19.3% 16.8% Zen 2 -> Zen 3 14.9% 19.7% Zen 3 -> Zen 4 5.1% 5.6%
Intel
Cinebench R23 SPECint 2017 Comet Lake -> Cypress Cove 18.1% 17.9% Cypress Cove -> Golden Cove 19.8% 18.5% Golden Cove -> Raptor Cove 3.6% 2.9%
I rest my case. It is up to you to believe in hopium.
Calculated the geomean from individual scores in Excel. That's the standard way - like Andrei used to do.How did you get the Zen3 -> Zen4 SPECint number from the Anandtech review?
Calculated the geomean from individual scores in Excel. That's the standard way - like Andrei used to do.
fMax in practice in case of AMD is just the boost clock as advertised, not the limit that is shown by the ucode.I get a geomean of 22.79% overall improvement which means a calculated IPC of 7.8% (5.75GHz Zen 4 vs 5.05 GHz Zen 3). I believe others found a slightly higher SPECint IPC increase, most likely due to using faster RAM compared to Anandtech who only uses officially supported speeds.
fMax in practice in case of AMD is just the boost clock as advertised, not the limit that is shown by the ucode.
For a new architecture, it seems that Cinebench tracks well with SPEC.
so explain why a FPU based workload , corelates with integer based workloads to the point of it invalidating a consistent rumour of integer performance.
7950x has fMax at 5.85 GHz. Anyway, using that number I get 1.2279*5.05/5.85 = 1.0599. So the ccaluclations agree.Fmax for the 7950x is 5.8GHz, but I'm using the boost clocks as measured and reported in the Anandtech reviews.
7950x has fMax at 5.85 GHz. Anyway, using that number I get 1.2279*5.05/5.85 = 1.0599. So the ccaluclations agree.
If that is what you're arguing, then you stand guilty of the same.The Fmax is not guaranteed and Anandtech’s sample does not boost that high. They measured the single core boost of their sample at 5.75 GHz so I don’t know why you would use anything else except to try and get the result you want. . . .
If that is what you're arguing, then you stand guilty of the same.
I go by what AMD advertises, and here you are nitpicking an obvious non-issue.
Doesn't make an iota of difference in my analysis of correlation between SPECint and Cinebench.
You can clutch at straws all you want on whether it is 5.7 or 5.75 or 5.85 - doesn't make any difference.AMD advertises 5.7 GHz, anything above that is not guaranteed and varies by sample. Anandtech actually measured their sample at 5.75 GHz boost so taking any other number makes no sense.
View attachment 97322
You can clutch at straws all you want on whether it is 5.7 or 5.75 or 5.85 - doesn't make any difference.
Yes, you are clutching at straws by effectively saying that since Anandtech got 5.75 GHz on their sample, it must mean that everyone else will also get the same, contradicting what you've already said about fMax varying by sample.Yep, I’m clutching at straws by using actual measured data. Sure thing.
Yes, you are clutching at straws by effectively saying that since Anandtech got 5.75 GHz on their sample, it must mean that everyone else will also get the same, contradicting what you've already said about fMax varying by sample.
Does that change the conclusions?Are we not using Anandtech’s SPEC results to calculate IPC in SPECint? Why would you not use their measured clock speed together with their measured scores? Where would anyone else’s sample come into the discussion?
C&C has an article on Cinebench and what instructions are in it:FWIU, these Cinebench figures are just an individual's opinion, nothing more. So why getting worked up so much about them?
--------
On possible Cinebench performance of Zen 5 (although I am not one of those who buy computers in order to run mostly Cinebench all day):
- If there will be 512 bit wide FP datapaths in Zen 5, can Cinebench (release XY) actually make use of them?
- If yes, do all Zen 5 products have the same FP pipelines per core? An earlier rumor spoke of 512 bit FP option or something like that. And I keep wondering what that was supposed to mean.
Based on what I understand from reading the article, it doesn’t appear to leverage AVX512.
And the AVX instructions in that graph are mostly scalar additions and multiplications. Very little vector instructions.C&C has an article on Cinebench and what instructions are in it:
https://chipsandcheese.com/2023/10/22/cinebench-2024-reviewing-the-benchmark/
Based on what I understand from reading the article, it doesn’t appear to leverage AVX512.
View attachment 97323
Yep, I’m clutching at straws by using actual measured data. Sure thing.
For today's YouTube Influencers that is the difference between DOMINATING and getting PWNEDBut how much of a difference does 5.7 vs 5.85 GHz make in this case? It's only 2.6% difference.
But we're not YouTubers. And the discussion was not about what vendor should top some chart.For today's YouTube Influencers that is the difference between DOMINATING and getting PWNED
Expect 100+ Influencers to make dozens of videos on how a vendor is topping the charts