Question RTX 5090 4K Gaming build - Core Ultra 265K tuned vs 9800X3D - Talk me out of or for returning my 265K parts and getting 9800X3D parts at microcenter

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

511

Diamond Member
Jul 12, 2024
3,971
3,680
106
If one can convince a single X3D CCD to run at 250 251W they might be closer than one expects.
I'd try it but I don't have sufficient cooling and I still don't think 5500MHz would be enough.
You can limit it to 9800X3D power and it will still be faster by a decent margin.
 

Schmide

Diamond Member
Mar 7, 2002
5,739
1,029
126
A few last points.

If you get a new early adopter chip. You should not expect a mature experience. Things are going to be found in the wild, kinks worked out. Cough Cough Assrock burning chips.

This goes for intel too. Their chips were cooking themselves. We will never know if they are truly fixed, but they could be better. Core Ultra had some serious growing pains. I think over the course of 6 months they fixed a ton of bugs and got like a 20% increase. (psst. you had to update your bios and drivers to get this)

The weird thing that bit me recently. Bitdefender freaks out with riot vanguard there is no way to know this and I spent months with random focus problems while playing riot games. Whitelist the riot game, it works fine. But there is no way to tell why it's doing this. Riot vanguard is not a great program and bitdefender probably has every right to freak out about the hooks it puts in the OS. One way I found out it wasn't the hardware. It happened on two different systems exactly the same issue.

I think the best advice these days, and it is a minefield, is to do updates on a regular schedule. Do backups too. If something gets out of wack, don't be afraid to uninstall the driver and reinstall it. If things really get out of wack. Get DDU from TechPowerUp, download chipset before, clean everything and reinstall. Don't be afraid to update bios. It rarely breaks things and often fixes small things like memory training.

Don't be a muddler.
 

Wolverine2349

Senior member
Oct 9, 2022
522
168
86
If one can convince a single X3D CCD to run at 250 251W they might be closer than one expects.
I'd try it but I don't have sufficient cooling and I still don't think 5500MHz would be enough.

I doubt it unless that got those cores like to 6.5GHz or faster.

I mean Lion Cove has IPC on par or slightky better than Zen 5. And Skymont e-cores have Zen 4 IPC. Core Ultra has fast cores.

Its memory subsystem because of interconnects is what holds back its gaming and makes it rgress a little or even modestly from Raptor Lake monolithic die.

9800X3D the 96MB of 3D vCache is what makes it so kick a*** for gaming combined with good IPC of Zen 5 on par or barley behind Lion Cove cores.
You can limit it to 9800X3D power and it will still be faster by a decent margin.

You mean for gaming not for productivity workloads right?
 

gdansk

Diamond Member
Feb 8, 2011
4,457
7,503
136
I mean Lion Cove has IPC on par or slightky better than Zen 5. And Skymont e-cores have Zen 4 IPC. Core Ultra has fast cores.
No, it's actually not even that close at all. At 5.5GHz LC is about 11.1 SPECint 2017 rate-1 and 5.5GHz Z5 3D is about 13.3.
Maybe you're not actually measuring and going by gut.
 
  • Like
Reactions: Tlh97

Wolverine2349

Senior member
Oct 9, 2022
522
168
86
No, it's actually not even that close at all. At 5.5GHz LC is about 11.1 SPECint 2017 rate-1 and 5.5GHz Z5 3D is about 13.3.

Oh wow. I thought it wa sjust the meory subsystem on Arrow Lake that was issues.

i mean SPECint 2017 does it also show same for 55GHz vanilla Zen 5 or does 3D vCache also help a lot for SpecInt 2017?
 

511

Diamond Member
Jul 12, 2024
3,971
3,680
106
No, it's actually not even that close at all. At 5.5GHz LC is about 11.1 SPECint 2017 rate-1 and 5.5GHz Z5 3D is about 13.3.
Maybe you're not actually measuring and going by gut.
SPEC is skewed by X3D look at Vanilla Z5.
 

gdansk

Diamond Member
Feb 8, 2011
4,457
7,503
136
i mean SPECint 2017 does it also show same for 55GHz vanilla Zen 5 or does 3D vCache also help a lot for SpecInt 2017?
Spec is sensitive to L3 why do you think X3D gains so much against vanilla Zen 5.
Coincidentally, it helps in that benchmark almost in proportion to how much it helps games...

ARL L3 is cooking it in Spec as well.
And in the real, memory-accessing world too. YHWH-willing Intel will fix that cache mess (in NVL).
 
  • Like
Reactions: Tlh97 and 511

Wolverine2349

Senior member
Oct 9, 2022
522
168
86
12.2 @ 5.5GHz. So still significantly better.

Is SPEC sensitive to AVX512.

I mean Zen 5 is the first AMD desktop to have true AVX 512 fully. Intel does not have ACX512 as they fused it off on later Alder Lake P cores. Zen 4 was half bakced and like AVX 2 X2 or something.

Its strange because in Spec Raptor Cove was like 6% faster than Zen 4. And Lion Cove supposedly 9% faster than Raptor Cove at same clock speed.

Per AMD Zen 5 only had a 15% IPC uplift over Zen 4 in non AVX512 workloads. But in AVX512 workloads the uplift was massive.

In such case non AVX512 and everything else Lion Cove and Zen 5 I think are close or maybe Lin Cove barely ahead. AVX512 changes that though.

Games do not care and never will care about AVX512. On the other hand they do and by a lot care about L3 cache which is why the 7800X3D and 9800X3D are so good per benchmarks.
 

DaaQ

Golden Member
Dec 8, 2018
1,951
1,407
136
I kind of prefer intel.

Though in reality is Intel's situation now as bad as AMD's was with Bulldozer in CPUs?

Some have mentioned that on other blogs. Some say not even close??

Of course AMD was way worse financially with Bulldozer than Intel is now. But how baout just raw top end CPU performance. While AMD wins is the gap as bad as Intel vs Bulldozer 10-13 years ago?

Intel is nowhere near AMD Bulldozer levels of bad. They have that going for them. Their future is still pretty bleak at the moment though IMHO but we have to see how how 18A and Nova Lake turn out. I am less than optimistic based on recent (and not so recent) history.
Can't read the 3rd page. It is just that my sig from OCN was with an intel person that was blasting the power consumption of FX8350/9590 lol.

Ignore original question is about ASRock BIOS 3.40

OCN sig.jpg
Oh Tsumi where did you go.

The IPC or clock speed......priceless.
 

Schmide

Diamond Member
Mar 7, 2002
5,739
1,029
126
Is SPEC sensitive to AVX512.

I mean Zen 5 is the first AMD desktop to have true AVX 512 fully. Intel does not have ACX512 as they fused it off on later Alder Lake P cores. Zen 4 was half bakced and like AVX 2 X2 or something.

Its strange because in Spec Raptor Cove was like 6% faster than Zen 4. And Lion Cove supposedly 9% faster than Raptor Cove at same clock speed.

Per AMD Zen 5 only had a 15% IPC uplift over Zen 4 in non AVX512 workloads. But in AVX512 workloads the uplift was massive.

In such case non AVX512 and everything else Lion Cove and Zen 5 I think are close or maybe Lin Cove barely ahead. AVX512 changes that though.

Games do not care and never will care about AVX512. On the other hand they do and by a lot care about L3 cache which is why the 7800X3D and 9800X3D are so good per benchmarks.
Enough with the FUD. Zen4 has full avx2 support. As did every chip past bulldozer. Which basically had avx2 except its FMA was FMA4. Zen4 also has full avx512 support. In fact it has more avx512 support than every other intel chip before Sapphire Rapids for which it is basically equal except for the FP16 operation which Sapphire Rapids is the only chip that supports it. No one is going to adopt the intel only 16bit float instruction when the whole industry has gone BF16. You'll say anything to put down amd without even doing the simplest websearch.
 
Last edited:

Wolverine2349

Senior member
Oct 9, 2022
522
168
86
No, only what an autovectorizing compiler can find. Which is none.


If you say so. But the instructions, not the execution width, have been found useful. It shouldn't be required for games for a decade or more but never say never.

Well AVX512 has been around since 2016 as Intel has it in earlier HEDT class CPUs before getting rid of it in mainstream CPUs and 0 games use it and still not all if not most productivity programs use it.
 
  • Like
Reactions: 511

gdansk

Diamond Member
Feb 8, 2011
4,457
7,503
136
Well AVX512 has been around for more than 10 years as Intel has it in earlier CPUs before getting rid of it in mainstream CPUs and 0 games use it and still not all productivity programs use it.
And the structural reasons for that are simple. Intel is still, through 2026, not shipping consumer parts capable of executing the instructions even at half rate.

But soon. PTL or NVL, I can't remember which right now, finally ends that madness.
 

511

Diamond Member
Jul 12, 2024
3,971
3,680
106
And the structural reasons for that are simple. Intel is still, through 2026, not shipping consumer parts capable of executing the instructions even at half rate.

But soon. PTL or NVL, I can't remember which right now, finally ends that madness.
NVL though they ship AVX-512 with ICL/TGL/RKL
 

gdansk

Diamond Member
Feb 8, 2011
4,457
7,503
136
NVL though they ship AVX-512 with ICL/TGL/RKL
Alas, the x86 vendors needed this:
The EAG includes Torvalds and Sweeney. So don't be surprised when Unreal 7 requires AVX-512/10 to give even 55 fps at 1080p with upscaling in 2035.
 
  • Like
Reactions: igor_kavinski