- Mar 3, 2017
- 1,747
- 6,598
- 136
Why there is Jensen? Asus fappening in him?I downloaded the source video from bilibili (
Not bad considering the benchmark doesn't favor Ryzens.13% faster than 7945hs in cpu-Z
MT clocks don't really matter, ST is far more important.I downloaded the source video from bilibili (because the quality in the player is horrible for me) and reuploaded it to youtube:
I don't know how reliable task manager clocks are, but multi seems to run at 3.7ghz?
View attachment 100794
View attachment 100793
1.21x faster in cinebench 2024 nt than 14900K
I made a run without and with native gcc flag on a CPU supporting AVX-512. Quickly looking at the disassembly it looks like only AVX2 was used (but I only looked at x264, so I might be wrong). The result is a mixed bag but it proves that some integer tests were auto vectorized.You'd almost always have to change the source code to take advantage of ISA extensions. Just because a compiler supports AVX2 or SME or whatever doesn't mean it will actually get used. It will almost never get used with generic source code - you have to fiddle with the source code to figure out how the compiler expects it to be written to have a shot at it getting used - basically have to look at the assembler output until it does what you want.
That, of course, is something you can't do with SPEC since the source code is fixed.
You do realize what Zen stands for don't you? Zen implies balance. Going with a huge core would be the antithesis to what the whole philosophy is about.My headcannon is that AMD did their usual penny pinching thing and instead of coming out with a large core that would smash the competition, they went for something cheaper that aimed at great power efficiency.
RDNA3 already did it.One person told to me that the >40% zen5 uplift chips had a ton of bugs and could not be launched.
Oh I missed it I should have put that 9000 thing with IPC... Fail for me... Saved Millennial card for you!
It is not easy to increase IPC gen to gen. That is, there is no magic bullet. AMD considers ~16% as generation improvement. For next couple generations, AMD will try to find bottleneck of Zen5 and will try to optimize.Because if you completely ignore the leakers, Zen 5 is still a breaking point in Zen's history.
Zen 1 was a beta, Zen 2 was the "real" Zen, Zen 3 was a highly efficient rework, Zen 4 mostly a shrink and tweak.
I/O, core, and general design, Zen has more or less been a highly stable thing. Zen 2->3->4 is more about refinement than real change.
Zen 5 was the moment where it made sense to really revamp the design for the next 2-3 generations. Zen 6 will be a tick, and there's no news yet on Zen 7.
It's not as if Zen 5 NEEDED to be massive, but it was the occasion to get a huge leap ahead.
Instead, it is apparently just a small change, the core gives 16% better according to AMD, and there's not much surprise.
It's really not about the leakers, but about what Zen's doing. Skymont is breathing down its neck now, just like Zen 2 did back in the day.
And again, it's WEIRD, just plainly weird, that it was shipped with 105W marked on the manifest but now ends up being 65W. I'm all for power efficiency and whatnot, but it feels like the target changed. Like AMD took in the backlash from the Zen 4 95°C limit and power draw getting much higher (and still didn't beat Intel fully), and decided to play different cards. Except that manifest was 40 days ago and the announced power was like 5 days ago.
So:
- did AMD bork the arch and it has some RDNA 3 type of problem where it would use way too much power to get its max IPC?
- did they make a Zen 3 Redux where the changes are small AGAIN and Zen 5 just is another case of AMD dodging risk and taking baby steps?
- is there a large discrepancy between the announced IPC number and actual Zen 5 (but why would they sandbag now?)
- other?
There's so many questions...but I don't think we'll know until the delid happens, the core gets shot, and we get 3rd party reviews.
Anything before that is just more noise from hypers and doomers.
View attachment 100793
1.21x faster in cinebench 2024 nt than 14900K
View attachment 100796
9950X = 1.21x 2233 = 2700
7950X = 1965
37.5% faster than 7950X and it doesn't even use AVX
Why not showing such a huge gain with other applications in IPC slide?
Also blender and handbrake also seems to be around 35-50% faster than 7950X
| Median | |
Intel Core i7-1165G7 @ 2.80GHz (AVX512) | 55.44 | |
| 59.41 | |
Intel Core i9-11900KF @ 3.50GHz (AVX512) | 183.49 | |
Intel Core i9-9900KF CPU @ 3.60GHz | 163.09 |
Something doesn't add up.
I checked several blender database sheet and confirmed blender doesn't utillized AVX512,
but still 9950x has as huge as ~56% advantage in blender against 14900k, and IPC chart still showed blender IPC ~23%.
Also handbrake doesn't utilize AVX512 as well, and ~55% faster than 14900k, although showing as miserable as IPC ~11% in IPC chart.
14900k is with Intel default setting from May which had not been revised.
comparing 9950x to 7700x with one ccd enabled and bench under nT might lead to borked result but still isn't enough to explain the weird.
Something doesn't add up.
I checked several blender database sheet and confirmed blender doesn't utillized AVX512,
but still 9950x has as huge as ~56% advantage in blender against 14900k, and IPC chart still showed blender IPC ~23%.
Also handbrake doesn't utilize AVX512 as well, and ~55% faster than 14900k, although showing as miserable as IPC ~11% in IPC chart.
14900k is with Intel default setting from May which had not been revised.
comparing 9950x to 7700x with one ccd enabled and bench under nT might lead to borked result but still isn't enough to explain the weird
Blender database Median Intel Core i7-1165G7 @ 2.80GHz (AVX512) 55.44
Intel Core i7-7700HQ CPU @ 2.80GHz 59.41 Intel Core i9-11900KF @ 3.50GHz (AVX512) 183.49 Intel Core i9-9900KF CPU @ 3.60GHz 163.09
View attachment 100802
EDIT: I could say, all results whatever from official slides or Geekbench database could be borked. How could an STX ES having stronger ST than retail version?
I PMed you with a larger explanation that I wont post here as its not exactly Zen5 related.what is the benefit of shrinking the cores/CCD altogether?
why dont they fill a big die area as much as possible? (i guess they will do that with strix halo)
Is there any reason to believe that the reviews will show higher IPC increase than the 16% that AMD announced?
Why would AMD be sandbagging Zen5 intentionally? Do they have a track record of doing that in the past? 🤔
Didn't they sandbag on Zen 3 and 4 ?Is there any reason to believe that the reviews will show higher IPC increase than the 16% that AMD announced?
Why would AMD be sandbagging Zen5 intentionally? Do they have a track record of doing that in the past? 🤔
Not afair, the performance of Zen 3 and 4 matched the performance they announced.Didn't they sandbag on Zen 3 and 4 ?
Sandbagged Zen 4 prior to release:Not afair, the performance of Zen 3 and 4 matched the performance they announced.