News AMD's new Zen 5 chip up to 55% faster than Intel's Core i9-13900K in leaked benchmark — AMD's Ryzen 9 9950X purportedly shines in AVX workloads

[usertests]

IMO, not all gamers might be interested in buying these slightly expensive X3D processors, because not everyone wants the fastest and the greatest.

People who don't want the latest and greatest shouldn't be buying a new CPU on AM5, they should be buying discounted CPUs on AM4. Or whatever refurbished office PCs they can find and put a GPU in (even a quad-core is probably fine).

AMD could release a 9600X3D if they wanted to, which would be a better value for gaming than the 8-16 cores. Although if we're talking about future launches (Zen 6/7), maybe the 6-core will become a relic of the past.

Wasn't there a recent leak or comment that the X3D parts would be coming quicker than usual for Z5?
I could've sworn I saw that just before/after travelling.

Late Q3 would still be faster I guess. There was a 5 month gap between 7950X and 7950X3D, and an additional few weeks for 7800X3D. We are already almost in Q3 (July to September).

 

[Thunder64]

I have no interest in an X3D part. I guess that makes me a troglodyte. Maybe in the future I will.

 

[KnightShadey]

Late Q3 would still be faster I guess. There was a 5 month gap between 7950X and 7950X3D, and an additional few weeks for 7800X3D. We are already in Q3 (July to September).

Yep, thanks, 🤠🤙 found the articles, and yeah my dyslexic + info overloaded brain was still thinking "Sooner Than Sept" vs the original Sooner than CES of the original Computex quote/leak.

Still hoping the rumours of Strix Halo in a similar speeded-up timeframe with fall announce at least hold true, vs again 2025. 🤞

 

[helper800]

That's No EXAGGERATION ! 😜 🤣 🤠🤙

Too many "tion"s there, Sir !

There can never be enough alliteraTION!

 

[abufrejoval]

Imho a 32 threads CPU was never a gaming CPU (same consideration for 13900, 14900 and so on).

I'd modify that to "a 32 threads CPU was never going to be JUST a gaming CPU".

I used to hate the fact that my older 18-22 core Xeon E5 v3 and v4 workstations, which obviously had cores capable of running at 4GHz or beyond, wouldn't allow one, two or four of them actually getting anywhere close to what their (in terms of core design) nearly identical desktop brothers could do.

The were so server designed (and clock limited) by Intel that their max clocks were kept much slower as "obviously" no-one would run a $5000 CPU with only 1-4 cores returning investment. Instead everyone would try to keep them loaded and that meant 2-2.5 GHz all-core to avoid meltdown and perhaps 3.6 GHz single core, not 4.2 the quad core i7s with the same core design could do.

They simply didn't anticipate (or want to support) that someone might be tempted to turn a workstation into a gaming rig after hours!

And they might have counted themselves lucky when a lot of these former cloud server CPUs found their way into very cheap x99 boards in and from China, because such recycling can hurt demand for new hardware.

And consequently what I love about these dual CCD Ryzens is the fact that you can have both in a single system, the workstation that delivers quite significant oomp without blowing out your ears nor your budget, while it can do double duty as a gaming rig, not compromising on top clocks for the most critical gaming threads.

Buying that new hardware as work tools from tax exempt income and later being able to pass them on as gaming rigs to kids and family lessens the pain in the purse at least a little bit.

It's the flexibility which made PCs so great, not being perfect at only one single job.

 

[abufrejoval]

I guess that makes me a troglodyte.

I had to look that one up... is a good one (as Greek usually is), might weave that non-chalantly into a future sentence here or there

 

[Thunder64]

I had to look that one up... is a good one (as Greek usually is), might weave that non-chalantly into a future sentence here or there

Yes, I find it is a good one to use on the rare occasion. Glad you enjoyed it.

 

[hotaru251]

Like what's the point in comparing AVX512 when it's disabled on the intel chip.

intel's cpu's will have thei "new" version of av512 as they are doing avx10 https://www.tomshardware.com/news/intels-new-avx10-brings-avx-512-capabilities-to-e-cores

and for the chips that dont have it?
thats a reason to possibly pick 1 or other if you can benefit from it (so valid thing)

if someones really into emulating and might wanna give ps3 a shot they'd ebenfit from knowing this stuff.

 

[TerryLaze]

if someones really into emulating and might wanna give ps3 a shot they'd ebenfit from knowing this stuff.

While having avx512 would be good, the CPUs that lack avx512 have so many more cores, in the form of e-cores, that the performance is still the same, and with the amount of power that avx draws I would bet that the power draw will be pretty much the same as well.

Also rpcs3 has done a lot of development lately and had a lot of performance improvements especially for low end CPUs, up to 100% faster in some titles.
It's a miss conception that you still need a super high end system to play games on rpcs3

Latest RPCS3 Version Offers Greatly Improved Performance on Low-End CPUs

The latest version of the popular PlayStation 3 emulator, RPCS3, allows for greatly improved performance on low-end CPUs.

wccftech.com

View: https://www.youtube.com/watch?v=19ae5Mq2lJE&t=3s

 

[abufrejoval]

It was Intel's poor decision making that led them to disable AVX-512 in Alder Lake and beyond, even though they were trying to get it to work at the last minute, and later cook up AVX10. AVX-512 support clearly helps in some programs that people use, like RPCS3.

AVX-512 was working just fine on Tiger Lake already, at least with what little code I could find to test it.

Somehow it dawned just way too late on Intel engineers that lack of AVX-512 support in the E-cores was going to produce a boatload of headaches for operating systems and applications.

And in the end I guess it was mostly fear of bad press, that had them turn AVX-512 into dark silicon: plenty of engineers would have liked to live with having a bit of management overhead for a big boost in choice workloads.

And I don't think it entirely an x86 issue, because from what I can tell there are also some (slighter) discrepancies between the exact ISA extensions BIG and little cores support on ARM64, which aren't always updated in lock-step.

Too bad there is no equivalent to 'cpuid' on ARM64 so I ran into a wall just trying to figure out how the OS was dealing with that on my Rockchip RK3588 based Orange PI 5+, which has A53 and A76 cores with generational differences.

 

[abufrejoval]

While having avx512 would be good, the CPUs that lack avx512 have so many more cores, in the form of e-cores, that the performance is still the same, and with the amount of power that avx draws I would bet that the power draw will be pretty much the same as well.

That the type of generalization that won't survive the onslaught of opposing data.

I'd recommend against it.

 

[oofdragon]

While I too am excited about 9800X3D bringing another 10% to 20% boost in performance, I dont see a single game where I would personally notice it tbh. I usually game at 1440p and there's no game that the 7800X3D isn't able to hit the target fps I expect. Heck even the 5800X3D hit the targets for me. These processor are already so fast that imo the new ones benefit only 1080p competitive players who are missing those last fps to hit their 500mhz monitor limit at the lowest resolution. What I'm "missing" right now is the GPU performance, around 30% higher what the 4090 can do, now that is going to he my next upgrade. I like to play it at best settings 1440p, and even games like Fortnite you can already get to 240fps with the 7800X3D at epic settings, so for me Zen5 will be just reading about. Depending on what the 5090 brings to the table I feel like I would only really be looking into CPU upgrade around Zen8.... but.. since I support AMD yeah I would buy a 9900X3D when a 9900XTX hit the market just to call it the ultimate over 9000 gaming PC lol

 

[TerryLaze]

That the type of generalization that won't survive the onslaught of opposing data.

I'd recommend against it.

I was talking only about ps3 emulation because that's the only thing the person I answered to was talking about.
If you have access to any "onslaught" then please do share with the rest of us.

 

[abufrejoval]

I was talking only about ps3 emulation because that's the only thing the person I answered to was talking about.
If you have access to any "onslaught" then please do share with the rest of us.

My mistake then, assuming it was meant more generally: sorry!

 

[TechyIT223]

Need proper gaming benchmarks though. These floating point numbers are of little value for everyday user.

Just a clickbait news for now.

 

[Amdlova]

X3D 3DO 3DFX I really don't care my game fps is locked on 60fps max 75 fps... I care more about Watts and heat. And this fancy cache don't do anything to improve on these conditions.
I like super core cpus 12+ cores. Want peak performance per time basis. Before that back to two watts idle condition.

The "Gaming" on computer industry's is a cancer...

This is a gaming cpu... sound so stupid and childish...
My 13600T office cpu can't run any game because it is a 35w cpu from a one litter chassis.

 

[saunupe1911]

IMO, not all gamers might be interested in buying these slightly expensive X3D processors, because not everyone wants the fastest and the greatest.

I think AMD is trying to shoot themselves at their foot this time by releasing theX3D parts so close to the launch of the non-X3D SKUs.

I mean doing this might also cannibalize the sales of the non-X3D parts as well. Previous X3D parts launched after 7-8 months of time period/gap. There would still be gamers who might be more interested in the non-X3D parts though.

Nah I think AMD sees that there's two distinct markets now for those CPUs. Let the market decide what's best for their usage needs and overall system. This would only backfire if the CPUs are extremely close in performance across the board.

 

[KnightShadey]

Nah I think AMD sees that there's two distinct markets now for those CPUs. Let the market decide what's best for their usage needs and overall system. This would only backfire if the CPUs are extremely close in performance across the board.

Agreed.

But even with close parts, the non-X3D parts should usually be much cheaper to make without the added $ cache $, also usually have higher yields (further reducing unit cost), and be more plentiful.
So, I would think even if performance was close, the regular parts would have the cost-pricing advantage to keep them attractive, and any of the cannibalized sales lost to the pricier X3D parts would already have an added profit margin built-in that AMD would welcome those additional X3D that would otherwise be standard ones.

The concern for AMD would be the inverse really, where the added cost of X3D showed no benefit.

Edit: just noticed in your reply the word 'cache' didn't get added from typing on my wonky cottage tablet, added it for context, although I know y'all go the gist.

 

[saunupe1911]

Agreed.

But even with close parts, the non-X3D parts should usually be much cheaper to make without the added, also usually have higher yields (further reducing unit cost), and be more plentiful.
So, I would think even if performance was close, the regular parts would have the cost-pricing advantage to keep them attractive, and any of the cannibalized sales lost to the pricier X3D parts would already have an added profit margin built-in that AMD would welcome those additional X3D that would otherwise be standard ones.

The concern for AMD would be the inverse really, where the added cost of X3D showed no benefit,

Well said! Honestly I think I may just upgrade my 5900X this year as it's paired with a 4090. We'll see if these Zen 5 chips really give a 4090 a push. Basically Christmas to myself lol.

 

[Supercomputer]

I don't know how these results were calculated in the table in the news, but these are highly underestimated/incorrect percentages.
Why weren't the results just posted as other websites did, screenshots from AIDA64, so that everyone could see the difference so everyone here is misled, and this distorts the entire discussion on the forum.

and so to the point

AES - 9950X is 2.53x faster than 13900k or 13900k 2.53x slower than 9950x, i.e. 9950x is 153% faster, not 55%!

FP32 - 9950x 3.15x faster than 13900k! so not 60% faster but 216% faster! (small difference !)

FP64 - 9950x is 3.12x faster than 13900k! i.e. approx. 214% faster!

so how did they count it in the article?

the differences regarding the 7950x are also very confusing.

there is also a big mistake in the title

 

[Supercomputer]

Agreed we have to wait for a full benchmark suite covering different apps and games.

So everyone seeing these results need to contain their excitement as this is mostly AVX 512 workloads which we were expecting big gains.

Why do I think that these AIDA64 tests are heavily based on AVX512 (bridges)?

AVX512, often mentioned above, states that these lower results in the table are due (only) to AMD adding AVX512 heh no, because e.g. AES does not use AVX at all (as stated by the manufacturer itself AIDA64) and 13900k is 2.5x slower.The correct results I gave are due to all the improvements listed below, not just AVX512

 

[Makaveli]

I don't know how these results were calculated in the table in the news, but these are highly underestimated/incorrect percentages.
Why weren't the results just posted as other websites did, screenshots from AIDA64, so that everyone could see the difference and yet everyone here is misled. and this distorts the entire discussion on the forum.

and so to the point

AES 9950X is 2.53x faster than 13900k or 13900k 2.53x slower than 9950x, i.e. 9950x is 153% faster, not 55%!

FP32 9950x 3.15x faster than 13900k! so not 60% faster but 216% faster! (small difference !)

FP64 9950x is 3.12x faster than 13900k! i.e. approx. 214% faster!
so how did they count it in the news?

the differences regarding the 7950x are also very confusing.

there is also a big mistake in the title

The percentages were corrected in the forum the results were posted in which was anandtech. The author didn't read the rest of the thread before posting this story.

 

[Supercomputer]

The percentages will corrected in the forum the results were posted in which was anandtech. The author didn't read the rest of the thread before posting this story.

on the anandtech forum? My answer was about incorrect values in the article on Tom's Hardware did you see how big the difference in percentages is in the table on Tom's Hardware?
And what author do you mean? Christopher Harper?

 

[Makaveli]

on the anandtech forum? My answer was about incorrect values in the article on Tom's Hardware did you see how big the difference in percentages is in the table on Tom's Hardware?
And what author do you mean? Christopher Harper?

The source of this data is from the anandtech forum.

 

[Supercomputer]

The source of this data is from the anandtech forum.

yes, because the results from AIDA64 appeared there because one of the users has friends. And the article on Tom's H. was based on these results (like all well-known computer portals), but completely wrong values were given in the table. So the article on Tom's needs to be corrected. and not on anandtech

results from AIDA64..

Table from Tom's