I am surprised to see nobody has brought up AT closing down yet.
At any rate, I want to try this patch pretty soon with my 7950X. First I have to figure out why my board seemingly has PCIE issues. I am actually glad X3D Zen 5 chips haven’t dropped yet, because it looks like I may need to get a new motherboard and RMA/sell this one.
You mean like this dedicated whole thread in that subforum?I am surprised to see nobody has brought up AT closing down yet.
The title is not exactly explicit about the content tbh.You mean like this dedicated whole thread in that subforum?
I think a part of it is that reviewers use the same test bench, only swapping the CPU and AMD CPUS have become fiddly with either 1 or 2 CCXs, 3D cache or no 3D cache, leading to windows getting a bit "confused", while regular consumers might only have had 1 single CPU during their lifetime of windows installation.To be fair there is a lot of conflicting reports, some people do see this uplift, some others don.t Can't even decide who is wrong and why the results are so different (is it VBS/HVCI getting randomly disabled or enabled, some GPU driver stuff, difference between JEDEC/XMP and tuned setups etc). I personally don't believe GoW5 uplift, but everything else looks plausible (albeit this can just only elevate W11 to W10 levels of CPU performance, but who knows, I'm not sure I'm ready to install this abomination just for 5-10% more fps in (some) CPU-limited games)
lol so now the update went from affecting one game out of 40 for Intel to all of a sudden it gives Intel more gains than AMD in all games?
The Hardware Canucks tweet mentioned above.Where are you seeing this data?
The Hardware Canucks tweet mentioned above.
I'll update this chart later when the bigger SKUs launch, but thought I'd share my current GB6.2 1T "IPC" chart with available Zen5 SKUs. The relative 100% is QC's Oryon 80 SKU, but I'm a bit lazy to change it atm.
| CPU | GB6.2 1T Pts | Peak 1T Freq. (GHz) | Pts / GHz | Pts / GHz % |
| Apple M4 | 3715 | 4.400 | 844 | 118.7% |
| Apple M3 Pro (12C) | 3138 | 4.056 | 774 | 108.8% |
| Apple M2 Pro | 2663 | 3.504 | 760 | 106.9% |
| Apple M1 Pro | 2409 | 3.220 | 748 | 105.2% |
| Qualcomm X1E-80-100 | 2845 | 4.000 | 711 | 100.0% |
| Arm Cortex-X4 (8G3 Galaxy) | 2287 | 3.390 | 675 | 94.9% |
| Arm Cortex-X3 (8G2 Galaxy) | 2107 | 3.360 | 627 | 88.2% |
| AMD 9700X | 3373 | 5.525 | 610 | 85.8% |
| AMD HX 370 | 2884 | 5.100 | 565 | 79.5% |
| Arm Cortex-X2 (8+G1) | 1806 | 3.200 | 564 | 79.3% |
| Intel i9-14900K | 3294 | 6.000 | 549 | 77.2% |
| AMD 7950X | 3083 | 5.700 | 541 | 76.0% |
| Arm Cortex-X1 (G3X G1) | 1596 | 2.995 | 533 | 74.9% |
| Intel i9-12900HK | 2611 | 5.000 | 522 | 73.4% |
| Intel Ultra 185H | 2573 | 5.100 | 505 | 70.9% |
| Intel Ultra 125U | 2163 | 4.300 | 503 | 70.7% |
| AMD 7840U | 2562 | 5.100 | 502 | 70.6% |
| Intel i7-1365U | 2583 | 5.200 | 497 | 69.8% |
| Intel i5-1255U | 2313 | 4.700 | 492 | 69.2% |
| AMD 6800H | 2063 | 4.700 | 439 | 61.7% |
Our workstation or synthetic benchmarks such as Cinebench, Blender, V-Ray, Corona, Handbrake or even 7-Zip and Geekbench do not benefit from the optimizations.
They are distinct, but have the same value by default. The user may set differing values for them within SKU dependent bounds.I'm not even sure if the EPYC platform has a distinction between TDP and PPT
Sorry if I'm stating the obvious and if I missed the context, but in general compression is more sensitive to branch prediction, while decompression is more sensitive to memory bandwidth (for not too complex compression algorithms). That's a generic statement that might apply or not to 7-zip My point is that changes to uarch don't apply the same to compression and decompression.
They are distinct, but have the same value by default. The user may set differing values for them within SKU dependent bounds.
As a further detail, "performance determinism" mode requires TDP = PPT (a.k.a. PPL); this mode is in turn the default mode for some but not all SKUs. (Performance determinism equalizes the performance across all CPUs of a given SKU by pulling down the better specimen in the silicon lottery bell curve to lower power levels. The alternative mode of operation is "power determinism", in which all CPUs of a given SKU are allowed to pull up as close as possible to the socket power limit, causing somewhat unequal performance between different specimen from the silicon lottery.)
Fake news and Xitter, who'da thought?HWC follow up tweet.
View attachment 106538
The slide they posted on Twitter is junk data meant to be click bait for their video.
TL;DR: the HX 370 "IPC" is ever so slightly higher, so it now beats Arm's 2021 Cortex-X2 uArch and clocks a healthy +1.9 GHz higher.
//
Notebookcheck finally reviewed their 9700X, with its GB6.2 results added to the database (but not the review, though that's common of NBC). It actually had the highest score of any outlet by one point (not enough to change any calculations). A new high-scoring HX 370 has been added and that did shift its ranking +1.
CPU GB6.2 1T Pts Peak 1T Freq. (GHz) Pts / GHz Pts / GHz % Apple M4 Apple M3 Pro (12C) Apple M2 Pro Apple M1 Pro Qualcomm X1E-80-100 Arm Cortex-X4 (8G3 Galaxy) Arm Cortex-X3 (8G2 Galaxy) AMD 9700X AMD HX 370 Arm Cortex-X2 (8+G1) Intel i9-14900K AMD 7950X Arm Cortex-X1 (G3X G1) Intel i9-12900HK Intel Ultra 185H Intel Ultra 125U AMD 7840U Intel i7-1365U Intel i5-1255U AMD 6800H
FWIW, NBC's system used Windows 11 23H2 (Build 22621.3958; which is actually 22H2, maybe a typo on the numbering; I assume the build is correct); thus, this is pre-"branch prediction" fixes, but according to HardwareLuxx, it did not change much (but I'm willing to be disproven with benchmarks), emphasis
| HWUB | HWC |
Starfield | 0.00% | 4.73% |
Cyberpunk | 7.20% | 10.50% |
Counter Strike 2 | 2.00% | 7.94% |
Baulder’s Gate 3 | 1.70% | -1.20% |
Forza Horizon 5 | 5.60% | 0.39% |
Horizon Forbidden West | 0.60% | -1.19% |
Total War: Warhammer 3 | 3.90% | 0.28% |
| Gears 5 | 35.10% | |
| Fortnite | 30.60% | |
| Remnant 2 | 22.30% | |
| Star Wars: Jedi Survivor | 21.90% | |
| World War Z | 21.50% | |
| Homeworld 3 | 20.00% | |
| The Calisto Protocal | 19.50% | |
| The Riftbreaker | 19.20% | |
| Party Animals | 19.10% | |
| Forza Motorsport | 18.50% | |
| Call of Duty: Modern Warfare 3 | 17.70% | |
| F1 24 | 17.20% | |
| Hogwarts Legacy | 17.20% | 14.10% |
| Rocket League | 16.10% | |
| Far Cry 6 | 15.40% | |
| Microsoft Flight Simulator 2024 | 15.20% |
It is for a future version of GB. The total ST score is simply the total ST points divided by the frequency detected. Basically an extension of their way of calculating MT score applied to ST.What's points per GHz got to do with it? Different architectures have different ways to get the same answer. Points per watt I'd understand. Otherwise just rank by performance.
GeekRNG?It is for a future version of GB. The total ST score is simply the total ST points divided by the frequency detected. Basically an extension of their way of calculating MT score applied to ST.
Riiiggght. But it's pointless. Who cares what frequency it's at?It is for a future version of GB. The total ST score is simply the total ST points divided by the frequency detected. Basically an extension of their way of calculating MT score applied to ST.
Best to do something like that with an automated testsuite like Phoronix's. Testing all that manually is a real pain. Or just use PCMARK10 and call it a day.Maybe I'm just weird, but I would totally watch/read a benchmark of the difference between VBS/Core Isolation/HVCI on/off on the last ~5 generations of CPU's from both AMD and Intel.
It’s a proxy for clock normalized single-core performance. That’s all. I don’t think it’s completely useless. It’s gives a general ranking of the performance of each p-core. Particularly within microarchitectures.Riiiggght. But it's pointless. Who cares what frequency it's at?
Why use a proxy when you have the actual scores? If Apple cores are twice as fast per clock but clocked half as fast then they have the same performance. The follow on is then which used the most power, so iyou use performance per watt. Ranking by performance per clock is..meaningless. Just means it targeted a different frequency in the design.It’s a proxy for clock normalized single-core performance. That’s all. I don’t think it’s completely useless. It’s gives a general ranking of the performance of each p-core. Particularly within microarchitectures.