News 11th-Gen Intel Rocket Lake-S Processor Benchmarked

For the curious, I just tested integrated graphics performance on UHD Graphics 630 -- using Core i7-9900K with DDR4-3200 memory, so basically as fast as Gen9.5 Intel GPUs get.

Fire Strike: 1341
Time Spy: 532

Obviously these Rocket Lake results are early and we don't know the GPU configuration, but that would be 30% faster on Fire Strike and 14% faster on Time Spy. Which is about half the performance you'd get from AMD's Vega 11.
 

MasterMadBones

Distinguished
Clocks are pretty much bang on with what I expected, maybe 100-200 MHz more for the final product. I've been saying for a while that in Ice Lake, the architecture was mostly the limiting factor for clock speed, not the node. Sure, 14+++ allows higher clocks than 10+, but the difference isn't huge.
 
For the curious, I just tested integrated graphics performance on UHD Graphics 630 -- using Core i7-9900K with DDR4-3200 memory, so basically as fast as Gen9.5 Intel GPUs get.

Fire Strike: 1341
Time Spy: 532

Obviously these Rocket Lake results are early and we don't know the GPU configuration, but that would be 30% faster on Fire Strike and 14% faster on Time Spy. Which is about half the performance you'd get from AMD's Vega 11.
Isn't the physics score basically a CPU score? Can you clock your 9900k at the same clocks this was running and tell us the physics score?
They show it in the link.
EZpATfkU4AU-f96
 
  • Like
Reactions: TJ Hooker
Clocks are pretty much bang on with what I expected, maybe 100-200 MHz more for the final product. I've been saying for a while that in Ice Lake, the architecture was mostly the limiting factor for clock speed, not the node. Sure, 14+++ allows higher clocks than 10+, but the difference isn't huge.
While clocks might be production related,it's also a great way to sell a future refresh instead of doing a new arch.
 
Isn't the physics score basically a CPU score? Can you clock your 9900k at the same clocks this was running and tell us the physics score?
They show it in the link.
EZpATfkU4AU-f96
Yeah, sure ... It was actually a 9700K, I was mistaken. But the scores in Fire Strike and Time Spy are here:
62

63

Pretty obvious the performance isn't tuned yet, unless this wasn't an 8-core chip.
 
  • Like
Reactions: TJ Hooker

Deicidium369

Permanantly banned.
BANNED
Mar 4, 2020
390
61
290
https://www.guru3d.com/news-story/i...ega-at-7nm-from-ryzen-4000-for-notebooks.html

https://wccftech.com/intel-tiger-la...e-faster-than-amd-ryzen-4000-renior-7nm-vega/

BUT

I think the Rocket Lake S is not a full 96EU like Tiger Lake - more like 24 or 32 EU. - so barely on par with the G7 Ice Lake variants. (~64 Gen11 = ~32 Gen12). This is a desktop flagship - so can't imagine anyone using the iGPU for anything other than curiosity.
That’s the question, of course: what is the GPU configuration in Rocket Lake S? There are good reasons for it to have at least a 64 EU graphics subsystem, like the fact that a lot of CPUs for desktop also go into notebooks. Given the performance, though, it seems like a 32 EU version. That or the drivers or clocks are very poor right now.

Anyway, I’ve got an integrated graphics article going up in the morning that digs into this stuff more. Basically, faster integrated graphics only makes sense for mobile solutions. For desktops, a dedicated GPU is easier, upgradable, faster, and doesn’t have the cooling limitations of a socketed CPU.
 

PCWarrior

Distinguished
May 20, 2013
199
81
18,670
Well in this particular benchmark we don’t know what the sustained clock speeds had been. For all we know it could have run at just the base speed of 3.2GHz. Also since we are talking about an engineering sample it is not just clock speeds that will improve. Some critical pathways of the CPU will also likely receive some reworking/optimisation.

Willow cove is expected to offer an increase in IPC of around 30%. But such IPC figures (and that is also applicable to AMD) are average figures. So on things like 7-zip the willow cove microarchitecture can offer an impressive 45-50% increase and in tile-based rendering a 35-40% uplift. Also, things that can leverage AVX 512 might even see a 2-2.2x speedup and who knows what speedups some iGPU-accelerated workloads will see. However, for many tasks, the new architecture will probably offer a modest uplift of less than 20% and closer to 10-15% in most cases. So, Intel can’t afford to lose too much clock speed compared to Comet lake. Intel will need to have a turbo to at least 4.6-4.7GHz across at least 4 cores in order to surpass Comet lake’s flagship on absolutely everything.

I do expect that they will reach these speeds. For example I expect the i9 11900K to have clock speeds that look something like these (or thereabouts):

4.7GHz - 2 cores active
4.6GHz - 4 cores active
4.5GHz - 6 cores active
4.4GHz - 8 cores active
4.3GHz - 10 cores active

So for things like Cinema4D or Blender (CPU rendering) I expect an overall uplift of around 20% when comparing the i9 11900K against the i9 10900K. For gaming I expect either a minor regression or parity or a minor gain of 1-2%. And any claimed major gains will probably be against a completely stock 10900K [i.e. (i) stock frequency table, (ii) power limits enabled, and (ii)RAM running at only 2933MHz] as I expect Rocket lake to have 3200MHz official memory support.