News Tested: Ryzen 7950X3D's iGPU Isn't Faster Than 7950X's

[Admin]

We tested the Ryzen 9 7950X3D's integrated GPU to determine if it delivers the huge gains we've seen posted at various locations. (Spoiler: Nope)

Tested: Ryzen 7950X3D's iGPU Isn't Faster Than 7950X's : Read more

 

[usertests]

The problem I had with the story is that it portrayed the 2 CU Ryzen 7000 iGPU as much weaker than the 32 EU UHD 770 used in Alder Lake and Raptor Lake, when other tests I've seen show them as about equal.

Source:
https://www.techpowerup.com/review/amd-ryzen-7-7700x/22.html

Ryzen 7600X and 7950X review: Zen 4 starts off expensive but impressive

Thermal behavior is a bit odd, but there's lots to like about these new CPUs.

arstechnica.com

Simply put, Intel's IGPs are far more powerful. The irony of this situation is almost painful: AMD not only has far more experience making GPUs but it also makes far more powerful IGPs in other products.

Simply untrue, guy with 4 names.

 

[deesider]

Simply put, Intel's IGPs are far more powerful...
AMD ... also makes far more powerful IGPs in other products.

I'm not sure critical thinking is his forte.

 

[JamesJones44]

Honest question because I haven't had time to review the architecture. Does the 7xxxX3D iGPU even have an interconnect to the V-Cache? I didn't think AMD was structuring the V-Cache to act like a UMA layer for both the CPU and GPU.

 

[InvalidError]

I didn't think AMD was structuring the V-Cache to act like a UMA layer for both the CPU and GPU.

The CPU has to be cache-coherent, so it would make sense for the IGP to access stuff in-cache if it happens to be there (ex.: draw command queue) so it doesn't waste time fetching a stale copy from system memory. Doubt it has any mean of using the CPU cache as its own in any other way.

 

[AgentBirdnest]

Props to Paul for investigating and testing this. : ) You do good work.

PC Mag updated their article:

Editors' Note, March 2, 2023: Some scrutiny by Tom's Hardware(Opens in a new window) and online commenters pointed out that our integrated graphics (IGP) testing numbers for the earlier Ryzen 9 7950X chip were unnaturally low; this was likely due to early-driver issues that only became evident with this review of the 7950X3D, and made the 7950X3D's IGP look like a leap forward by comparison. We have retested the Ryzen 9 7950X and updated the numbers in the IGP testing table here and in that original review, and made minor tweaks to this article's intro and IGP Testing section to reflect the new numbers. We'll also be retesting the IGPs on the Ryzen 7 7700 and 7700X in the coming days. The original, fundamental conclusions we made about this chip, though, have not changed.

 

[Amdlova]

Long time ago, before the hummus sapiens... intel make a cpu with edram crystal well with true 128mb l4 cache. Maybe has helped with graphics... but that glue amd shinigamis have can't be too good

 

[rluker5]

It is good this came out. Generally speaking, desktop igpus are terrible at gaming. Pretending they are good is just setting people up for disappointment.

One can be better than the other, but if they are still worse than a 10 year old dgpu, sorry they are terrible.

But they are good for non gaming display tasks, have been for a long time and having one is a big usability improvement for Ryzen for those who don't game.

 

[InvalidError]

It is good this came out. Generally speaking, desktop igpus are terrible at gaming. Pretending they are good is just setting people up for disappointment.

One can be better than the other, but if they are still worse than a 10 year old dgpu, sorry they are terrible.

Depends on the IGP: while the basic IGP in Intel and AMD's normal desktop lineups may be only adequate for basic desktop use and simple games. Intel's mobile IGPs with ~3X the performance is getting into usable territory and AMD's latest 780M does make most current sub-$200 GPUs irrelevant performance-wise.

Once AMD or Intel decides to bring some DRAM on-package, we'll probably see "big IGPs" aiming somewhere in the neighborhood of RTX4060-4070 performance with a tile/chiplet whatever the size the DRAM chips stacked on top of it are.

 

[JamesJones44]

Depends on the IGP: while the basic IGP in Intel and AMD's normal desktop lineups may be only adequate for basic desktop use and simple games. Intel's mobile IGPs with ~3X the performance is getting into usable territory and AMD's latest 780M does make most current sub-$200 GPUs irrelevant performance-wise.

Once AMD or Intel decides to bring some DRAM on-package, we'll probably see "big IGPs" aiming somewhere in the neighborhood of RTX4060-4070 performance with a tile/chiplet whatever the size the DRAM chips stacked on top of it are.

Totally agree here. DRAM on package will likely allow iGPUs to compete for lower end dGPU share.

 

[rluker5]

Depends on the IGP: while the basic IGP in Intel and AMD's normal desktop lineups may be only adequate for basic desktop use and simple games. Intel's mobile IGPs with ~3X the performance is getting into usable territory and AMD's latest 780M does make most current sub-$200 GPUs irrelevant performance-wise.

Once AMD or Intel decides to bring some DRAM on-package, we'll probably see "big IGPs" aiming somewhere in the neighborhood of RTX4060-4070 performance with a tile/chiplet whatever the size the DRAM chips stacked on top of it are.

That is a relative measure. You can call phone graphics fast by a relative measure, but compared to the standard of desktop graphics they are not.
The 680m is slower than a 2013 290x. And an A380. At least according to Techpowerup.
The 680m, while excellent as a low power/low performance mobile solution, can't be called good in terms of the normal accepted reference of desktop gaming.

The igpu in a 7700x or 12700k is hardly a 680m when it comes to gaming.

People hoping otherwise will be let down.

But they do display the desktop and play video quite well. At least the Intel ones. I'm just assuming the AMD ones do as well since their dgpus are up to normal standards.

 

[InvalidError]

That is a relative measure. You can call phone graphics fast by a relative measure, but compared to the standard of desktop graphics they are not.

While it may not be fast relative to current-day $1500+ GPUs, early numbers on AMD's 780M (Ryzen 7840HS/7940HS) land in the realm of the GTX1060M, which is better than or on par with what 40+% of people on the Steam survey have. That is relatively fast compared to the installed base based on the only substantial public data source we have.

Since the 780M has 2.5X as much raw compute as the 680M (3.4 vs 8.9 TFP32) but is only ~25% faster, I'll hazard a guess that it is heavily hamstrung by LPDDR5. If it performed in line with its FP32 spec, it should be almost on par with the RTX3050. It may not please the 4K Ultra RT crowd but tons of more casual gamers would be perfectly fine with something that is already 3-10X faster than the antiques they currently have.

 

[waltc3]

Yes, there's a lot of strange stuff on the Internet these days...I think we all know that! Yes, the AMD iGPU is a very rudimentary GPU, simply designed to put up a solid video signal in the event that you need one. It's not meant for running 3d games or anything else, but simply providing a video signal sans a discrete GPU--or for other reasons unrelated to performance. AMD's APU's are the iGPUs that are meant to compete with Intel's. For years, AMD APUs trounced Intel's iGPUs running 3d games all the way back to ATi, IIRC--but if you want to run 3d games, especially if you want decent resolutions (2k or 4k) with eye-candy, then you want a discrete GPU in a desktop--not an iGPU, whether it's an APU, an iGPU, or an Intel iGPU. From what I read, the discrete laptop GPUs (not sure if that's an oxymoron) run pretty good these days.

I never expected to see people write up these new AMD iGPUs for performance purposes, as AMD has made no bones about the fact that they don't exist for even APU levels of performance. For me, the equation is 4x0 =0...

 

[rluker5]

While it may not be fast relative to current-day $1500+ GPUs, early numbers on AMD's 780M (Ryzen 7840HS/7940HS) land in the realm of the GTX1060M, which is better than or on par with what 40+% of people on the Steam survey have. That is relatively fast compared to the installed base based on the only substantial public data source we have.

Since the 780M has 2.5X as much raw compute as the 680M (3.4 vs 8.9 TFP32) but is only ~25% faster, I'll hazard a guess that it is heavily hamstrung by LPDDR5. If it performed in line with its FP32 spec, it should be almost on par with the RTX3050. It may not please the 4K Ultra RT crowd but tons of more casual gamers would be perfectly fine with something that is already 3-10X faster than the antiques they currently have.

The 7950X3D's igpu isn't a 780m. It shouldn't be hyped for it's gaming prowess.

Even something that is on par with the 780M, the A380, the slowest available first gen Intel dgpu is largely regarded as a bad card for gaming and better intended for other things like AV1 encoding, enhanced video calls, video super resolution etc. It is regarded as barely passable in gaming by most when it comes to new cards. This is the DGPU measure. The 780m is king of the low power/low performance mobile igpu category which is a different one and should be segregated to avoid misleading people.

 

[InvalidError]

The 7950X3D's igpu isn't a 780m. It shouldn't be hyped for it's gaming prowess.

I went out of my way to explicitly state the same in the post you replied to, differentiating between CPUs with basic IGPs and CPUs with "big IGP", no point in pointing that out.

Even something that is on par with the 780M, the A380, the slowest available first gen Intel dgpu is largely regarded as a bad card for gaming

On paper, the 780M should be almost twice as fast as the A380. Can't do miracles with only 130GB/s of memory bandwidth shared with the CPU.