Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[Hitman928]

I'm rebenching 7000 series for hwbot atm while i wait for new delid frame for my delidded 9950X, so it have to wait.. But 7zip is first on my list when i get my Zen5 system put back together.

But i can share one other little interesting thing.
Pre agesa 1.2.0.2 it was recommneded to use the "X3D core parking" meta for Zen5 dual CCD cpus, for best gaming performance.
Now with fixed cross CCD latency fixed ob agesa 1.2.0.2 you actually get higher performance with not using the core parking (high performance powerplan)

Following screenshots belong to other guy from overclock.net forum, so this will only serve a preview:

CCD1 parked -> game is running on CCD0 only (balanced powerplan)
View attachment 107758

Core parking disabled -> both CCD enabled (high performance powerplan)
View attachment 107759

He will share more games when he is done with testing

Are there like for like settings comparison in a before and after scenario? I ask because HWUB didn’t show any real issue with the dual CCD models with both CCDs enabled and no core parking in their launch day review.

 

[MarkPost]

Are there like for like settings comparison in a before and after scenario? I ask because HWUB didn’t show any real issue with the dual CCD models with both CCDs enabled and no core parking in their launch day review.

View attachment 107778

There is a measurable difference perfomance in Cyberpunk with affinity set to CCD0 or without affinity. This is my 9950X.

Without affinity:

With affinity (CCD0):

This is with pre BIOS to the recently new AGESA version (not available for my board yet)

 

[tsamolotoff]

Wth is up with those SC2 framerates? It's 2024 and top end CPU can't run that game faster than 100 FPS?

You can try it yourself, here's a video-instruction how to do it :

Replay:

Яндекс

Найдётся всё

disk.yandex.ru

 

[Hitman928]

There is a measurable difference perfomance in Cyberpunk with affinity set to CCD0 or without affinity. This is my 9950X.

Without affinity:

With affinity (CCD0):

This is with pre BIOS to the recently new AGESA version (not available for my board yet)

Interesting, I wonder why HWUB isn't showing the same. Do you see the same thing if you disable one CCD entirely?

Edit: Also, which power plan are you running?

 

[Josh128]

I'm rebenching 7000 series for hwbot atm while i wait for new delid frame for my delidded 9950X, so it have to wait.. But 7zip is first on my list when i get my Zen5 system put back together.

But i can share one other little interesting thing.
Pre agesa 1.2.0.2 it was recommneded to use the "X3D core parking" meta for Zen5 dual CCD cpus, for best gaming performance.
Now with fixed cross CCD latency fixed ob agesa 1.2.0.2 you actually get higher performance with not using the core parking (high performance powerplan)

Following screenshots belong to other guy from overclock.net forum, so this will only serve a preview:

CCD1 parked -> game is running on CCD0 only (balanced powerplan)
View attachment 107758

Core parking disabled -> both CCD enabled (high performance powerplan)
View attachment 107759

He will share more games when he is done with testing

Ah Zen 5. The gift that keeps on giving. I wonder if it'll still be needing re-reviews and further testing️ by the launch week of Zen 6. 🤦‍♂️ 🤷‍♂️

 

[MarkPost]

Interesting, I wonder why HWUB isn't showing the same. Do you see the same thing if you disable one CCD entirely?

yes, this is with CCD1 disabled in BIOS:

 

[tsamolotoff]

Ah Zen 5. The gift that keeps on giving. I wonder if it'll still be needing re-reviews and further testing️ by the launch week of Zen 6

If you don't retest, you simply do it wrong for many reasons. There are always some new features and tweaks added after the product launches (sometimes these tweaks are bad for performance, as for me personally zen3 performed better with inital agesa than with the subsequent releases, although if your CPU can go above 1867 fclk it might be otherwise). For example, Zen4 didn't have either (or both) refi / gdm on release, 1:2 mode was non-functional, the CPUs rarely trained above 6200 in 1:1 mode and there were some other problems. Subsequently they've got fixed so you have to retest to see the changes, otherwise your tests would be highly irrelevant (like almost all TPU CPU tests, for example)

 

[Hitman928]

yes, this is with CCD1 disabled in BIOS:

Thanks for the confirmation. Still have no idea why HWUB is getting different results. Maybe just a difference between the built-in benchmark and actual game play but that doesn't seem like a likely reason.

 

[MarkPost]

Thanks for the confirmation. Still have no idea why HWUB is getting different results. Maybe just a difference between the built-in benchmark and actual game play but that doesn't seem like a likely reason.

well, honestly speaking, I'm not very confident with HWU results in general, I think their testing metodologhy isnt the more reliable. For example, I fear they use the same OS installation for the same CPU brand, regardless the model they test. An indication about it:

 

[MS_AT]

Hope this will look fine but this doesn't want to dispaly properly on my mobile... anyway I dont mean to downplay improved performance. I am just interested if the gain is related to what they did to improve latency benchmark or something else. That's it. So far it seemed that the core to core latency test was an outlier that did not have real influence on anything else. Now they released the latency fix with other improvements and I am interested what is contributing to the gain you have measured.

 

[CakeMonster]

I'd love if these improvements did away with the need for the additional core parking and thread priority driver in the 9950X that the 7950X doesn't need.

 

[Hitman928]

Anyone with a 7950x and Cyber Punk that is willing to run the benchmark with and without a CCD disabled?

 

[whoshere]

Spent some quite time digging for the best 9950X and 14900K scores in GeekBench 5 which I still trust and love more than any modern benchmark.

AMD just couldn't. I've decided to postpone my Zen 5 purchase and I'll instead wait for ARL reviews.

Power efficiency is commendable but I expected Zen 5 to beat RPL/RPL-Refresh. Hasn't happened.

Flame bait in a vendor thread is not permitted. Please read and follow the CPU forum rules stickied at the top.

Mod DAPUNISHER

 

[MarkPost]

Spent some quite time digging for the best 9950X and 14900K scores in GeekBench 5 which I still trust and love more than any modern benchmark.

AMD just couldn't. I've decided to postpone my Zen 5 purchase and I'll instead wait for ARL reviews.

Power efficiency is commendable but I expected Zen 5 to beat RPL/RPL-Refresh. Hasn't happened.

what crap of comparison is that?

 

[Det0x]

Spent some quite time digging for the best 9950X and 14900K scores in GeekBench 5 which I still trust and love more than any modern benchmark.

AMD just couldn't. I've decided to postpone my Zen 5 purchase and I'll instead wait for ARL reviews.

Power efficiency is commendable but I expected Zen 5 to beat RPL/RPL-Refresh. Hasn't happened.

lol your comparing a watercooled 9950X score with a LN2-cooled 14900K @ 7.4ghz
Seems like a fair comparison haha 🤣

 

[Hail The Brain Slug]

lol your comparing a watercooled 9950X score with a LN2-cooled 14900K @ 7.4ghz
Seems like a fair comparison haha 🤣

Also win 10 vs. win 11, probably a fair bit of score difference just from the OS alone.

 

[Markfw]

lol your comparing a watercooled 9950X score with a LN2-cooled 14900K @ 7.4ghz
Seems like a fair comparison haha 🤣

I don't know why we allow such a biased comparison at all.

 

[Josh128]

Looks like another price cut for 9950X over at Amazon US. Slowly but surely.

 

[CouncilorIrissa]

lol your comparing a watercooled 9950X score with a LN2-cooled 14900K @ 7.4ghz
Seems like a fair comparison haha 🤣

Because this is a bait, rather than a comparison.

 

[Hitman928]

Spent some quite time digging for the best 9950X and 14900K scores in GeekBench 5 which I still trust and love more than any modern benchmark.

AMD just couldn't. I've decided to postpone my Zen 5 purchase and I'll instead wait for ARL reviews.

Power efficiency is commendable but I expected Zen 5 to beat RPL/RPL-Refresh. Hasn't happened.

I fully expect to see you make the same comparison between a water cooled ARL and that RPL score after ARL launches. I’m curious what your conclusion will be.

 

[Hail The Brain Slug]

View attachment 107824

Looks like another price cut for 9950X over at Amazon US. Slowly but surely.

I just sold my 7950X on craigslist and I am trying to wait for X3D but I'll be damned if I can't resist the price if they keep lowering it like this.

 

[PJVol]

This is my 9950X

Well... it seems my old horse isn't that far behind

 

[jpiniero]

Looks like another price cut for 9950X over at Amazon US. Slowly but surely.

Think AMD will get the memo that they should just stick with VCache models for DIY?

 

[Markfw]

Think AMD will get the memo that they should just stick with VCache models for DIY?

Not everybody needs a gaming cpu. Until they put vcache on both ccds, I just need 16 fast cores. the 9950x is perfect for what I do. (but I wish I had 32 fast cores ! and 4 memory channels)

 

[branch_suggestion]

Wth is up with those SC2 framerates? It's 2024 and top end CPU can't run that game faster than 100 FPS?

Do they have some crazy custom map benchmark or something? Or did Blizzard just bloat it to the moon? Still running DirectX 9 too?

I remember playing SC2 beta in 2010 on an i5-750 and it ran good enough to get me to Diamond rank

SC2 is infamous for being a CPU killer to this day. DX9/10 with a single thread with a lot of CPU heavy calcs.
WoW had a big engine overhaul some years ago to DX11/12 with native multithreading.
SC2 therefore remains relevant and loves big L3s, Zen3 was a colossal uplift vs Zen2 in SC2 and X3D pushed it a lot further than that. Nice to see the evolution continue.