- Mar 3, 2017
- 1,747
- 6,598
- 136
Yeah. But apparently not IncavatorSo DozerBull. Gotcha!
It’s not unreliable, you just have to know what the benchmark is doing to determine how valuable its results are for each use case.Geekbench 6 is so unreliable, its results seem to be random. You can't trust GB 6's results under any circumstances. I've already written about it on reddit. Just look at some of the highest results in the GB6 database.
one of many examples of nonsense. According to GB6, 14900 has the same MT performance as TR with 96 cores
View attachment 101878View attachment 101879
View attachment 101880
As you can see in these results what I gave ST also not, look at the results in the first picture. As for MT, 64 core TRs have a score higher than 96 cores. So I wonder if GB 6 can even handle 16 cores(32 threads ) processors like 7950X 9950X and not skew their results.
The future upgrade path is pretty bleak given the low-hanging fruit iteration - Zen 6 - targets 10% IPC.Zen 5 is about single thread performance and future upgrade path.
Where do you have that from?The future upgrade path is pretty bleak given the low-hanging fruit iteration - Zen 6 - targets 10% IPC.
AMD core roadmap slide showing 10-15+ IPC for Zen 5 and 10% for Zen 6. Published by MLID.Where do you have that from?
If you want responsiveness, just fix the software.I really want this sort of performance in the PC world. Can you imagine a M4 level IPC x86 chip at 5.7GHz!!!!
The responsiveness be like butter.
Dozlubber was not so hot (or perhaps too hot, some say), but I could see myself run on Vacatinor all my life, any day.Yeah. But apparently not Incavator
Gates’ law: The speed of software halves every 18 months.If you want responsiveness, just fix the software.
Hardware will never win the race against software bloat.
That's the edited version, original slide just says "double digit IPC increase"
What about frequency bump on top of that due to 2/3 nm for Zen6?The future upgrade path is pretty bleak given the low-hanging fruit iteration - Zen 6 - targets 10% IPC.
That's the edited version, original slide just says "double digit IPC increase"
Adored asked my opinion on the slide and whether the 10-15% IPC claim was accurate or intentionally misleading/"sandbagging". I said I heard larger IPC gains, so he decided not to do a video on it and gave the slides to MLID.That the one that you intentionally suppressed because you thought it didn't make Zen5 look good enough, or a different one?
Yeah, its common misconception that just because rendering is something suited to GPUs and there are GPU rendering solutions available, people dont use CPUs for that task anymore and therefore multi-core performance is not really even needed outside of servers.
Another reason to get Zen5 for users who need the AVX-512. Looks like AMD knows what is doing and companies like Pixar probably sent feedback to AMD and AMD delivered with Zen5.
Looks like it but it doesn’t hurt to have options.Yeah, its common misconception that just because rendering is something suited to GPUs and there are GPU rendering solutions available, people dont use CPUs for that task anymore and therefore multi-core performance is not really even needed outside of servers.
The quality of the GPU render is awful next to the CPU render, and it's barely touched on in that blog post, and only in passing. The time comparison was definitely not to achieve the same end result.Looks like it but it doesn’t hurt to have options.
That thread is a goldmine of information.
Definitely not, there is still a need for CPU based rendering.The quality of the GPU render is awful next to the CPU render, and it's barely touched on in that blog post, and only in passing. The time comparison was definitely not to achieve the same end result.
So what's the consensus here, is zen 5 good yet?
And on the desktop AVX-512 too, no?It seems top-notch if you can take advantage of SMT with your workload.
And on the desktop AVX-512 too, no?
But that is not the selling point for zen 5 thoughIt seems top-notch if you can take advantage of SMT with your workload. If not then maybe just kinda good.