Correct me if I'm wrong, but isn't AVX-512 pretty tiny in silicon area? I thought I saw a die shot with it called out and it wasn't much of the chip, but I could be wrong there. You are correct that AVX-512 is better as a separate accelerator. That is Intel's long term plan to mix-and-match chiplets with foveros on a CPU. But they have to start somewhere to get software started using it.
Nah, not really. Intel has essentially zero presence right now in HPC and AVX-512 won't change that.
My education was in computer simulations of chemical reactions. I don't do that any more. But, I do somewhat pay attention to the field. AVX-512 helps. https://www.ansys.com/blog/maximize-mechanical-hpcNah, not really. Intel has essentially zero presence right now in HPC and AVX-512 won't change that.
The most recent big deals have been going to AMD but Intel is still the CPU in the vast majority in the Top 500 list. AVX-512 must be popular enough with them that AMD was asked to add it.
It couldn't possibly be used in things like photoshop, video editing, compression, science, etc.
How old are those machines? The CPU is mostly a commodity part to coordinate the compute units. Those HPC builders are certainly not happy about having to cool those Intel CPU space heaters on top of the GPU... no wonder Fujitsu is going their own way.
If you're referring to the A64FX, Fujitsu is throwing out the entire GPGPU paradigm in favor of SVE.
Not entirely... last I heard there is still some collaboration with GPGPU vendors.
Part of the problem was that AVX-512 was in server chips, as-of-now out-of-date workstation chips, and recently laptop chips. Neither of those is really the target market for software like Photoshop that could really utilize AVX-512. Until desktops and newer workstations have AVX-512, there really wasn't much incentive to optimize consumer software for it. As time goes on, there will be more and more software that uses it.It *can* be, but for whatever reason the instruction set scored very few decisive victories for Rocket Lake over Comet Lake in productivity apps that do (or at least should) support AVX512.
It think it would be interesting if Rocket Lake 11900k was benchmarked against an 10nm Ice Lake-X, assuming Intel's still going to release it for their HEDT platform and there's a cut down version with only 8 or 10 cores and compare their power usage.
Part of the problem was that AVX-512 was in server chips, as-of-now out-of-date workstation chips, and recently laptop chips. Neither of those is really the target market for software like Photoshop that could really utilize AVX-512. Until desktops and newer workstations have AVX-512, there really wasn't much incentive to optimize consumer software for it. As time goes on, there will be more and more software that uses it.
There are a lot of people who do Photoshop work on their laptops, so having those kind of optimizations makes a big difference. Adobe just released a native version for Apple's M1 Macs that's up to 50% faster than the old x86 notebooks from Apple. Don't underestimate the importance of having that kind of performance in the laptop market.
The mobo could have a max vcore of 2v, is there any way of us knowing? and it limits the clocks to the all core turbo, so massive O/C volts and power draw for normal clocks, all core turbo only...the massive cooler doesn't help with the clocks since those are locked but helps with increasing the power draw even more since it hits the cut off point for temps later or not at all.Anandtech's review did use massive cooler and infinite power budget for those test - that's actually the best possible situation for Rocketlake.
The mobo could have a max vcore of 2v, is there any way of us knowing? and it limits the clocks to the all core turbo, so massive O/C volts and power draw for normal clocks, all core turbo only...the massive cooler doesn't help with the clocks since those are locked but helps with increasing the power draw even more since it hits the cut off point for temps later or not at all.
How is that actually the best possible situation for Rocketlake?
The mobo could have a max vcore of 2v, is there any way of us knowing? and it limits the clocks to the all core turbo, so massive O/C volts and power draw for normal clocks, all core turbo only...the massive cooler doesn't help with the clocks since those are locked but helps with increasing the power draw even more since it hits the cut off point for temps later or not at all.
How is that actually the best possible situation for Rocketlake?
I would rather see a professional review using settings that are not brain dead.You would rather see the results with the processor hitting tjmax and thermal throttling?
| Comet Lake | All Core SSE Frequency | All Core AVX2 Frequency | Per Core Frequency | All Core Die Sense Vcore | Power Limit | % Capable |
| 10700K | 4.90GHz | 4.80GHz | 4C+100MHz | 1.190V | 210W | 100% |
| 10700K | 5.00GHz | 4.90GHz | 4C+100MHz | 1.210V | 220W | Top 63% |
| 10700K | 5.10GHz | 5.00GHz | 4C+100MHz | 1.230V | 240W | Top 22% |
| 10700K | 5.20GHz | 5.10GHz | 4C+100MHz | 1.250V | 270W | Top 2% |
WTAF did I just read?The mobo could have a max vcore of 2v, is there any way of us knowing? and it limits the clocks to the all core turbo, so massive O/C volts and power draw for normal clocks, all core turbo only...the massive cooler doesn't help with the clocks since those are locked but helps with increasing the power draw even more since it hits the cut off point for temps later or not at all.
How is that actually the best possible situation for Rocketlake?
Power limits don't work the way you think they work mate.I would rather see a professional review using settings that are not brain dead.
If you use the power limit for a 5.2 Ghz overclock then also show the performance of the 5.2Ghz overclock.
If you show the performance of the base all clock turbo show it with the according power limit.
But showing the lowest performance possible with the highest power limit possible is just ridiculous but that's what every review is doing.
And yes these are 10700k numbers because the 11700k isn't out yet.
Comet Lake All Core SSE Frequency All Core AVX2 Frequency Per Core Frequency All Core Die Sense Vcore Power Limit % Capable
10700K 4.90GHz 4.80GHz 4C+100MHz 1.190V 210W 100% 10700K 5.00GHz 4.90GHz 4C+100MHz 1.210V 220W Top 63% 10700K 5.10GHz 5.00GHz 4C+100MHz 1.230V 240W Top 22% 10700K 5.20GHz 5.10GHz 4C+100MHz 1.250V 270W Top 2%
L3$ latency while still regressing over the previous gen is now significantly better.AT received updated microcode and also confirmed they were using 1:1
L3$ latency while still regressing over the previous gen is now significantly better.
Makes little sense to have 45 cycle L3 latency and 150 cycle inter core latency.
You might want to brush up on the basics of cache coherency and data movement before spewing more FUD.
You have no idea why microcode patches are deployed. Engineers do not test something different from what they intend to release, as that would defeat the entire point of testing. Patches are applied as bugs are found after general sale, as @scineram pointed out. They almost always result in performance degradation. I have never seen a bug patch improve performance in my career.