Seems about on par for the rumored 3CU "so you don't need a dGPU" iGPU, no?

ValantarSeems about on par for the rumored 3CU "so you don't need a dGPU" iGPU, no?

It seems a bit slow, it ought to be about twice as fast in 3D, no?

512mb allocated and running Quake Wars version 2007 down Linux, wow benchmarking.

Next time test with Quake 3 Linux on a CRT monitor. Think that would be more fitting towards a news post, about an iGPU that's claimed to be as fast as a 3060 mobile.

Still the 5.21GHz clock rates are pretty impressive for Ryzen.

InVasManiStill the 5.21GHz clock rates are pretty impressive for Ryzen.

How will this 5.21 GHz frequency affect the performance? Will it make Zen 4 competitive against the Intel's Raptor Lake with 8 performance cores, 16 efficient cores and 32 logical processors in total?

One-third of compute units = one-third of performance. nothing new here.

The only surprising thing is that someone already has the product in hand to test.

TheLostSwedeIt seems a bit slow, it ought to be about twice as fast in 3D, no?

Would it? 6900HX is 12CU (@2.4GHz), this is rumored to be 3-4CU. Even if the lower CU count makes it less bandwith starved, this still delivers ~1/3rd performance in the benchmark you highlighted. For a single, very old, very odd benchmark that seems pretty close to linear scaling, especially if we also factor in lower core clocks.

X71200512mb allocated and running Quake Wars version 2007 down Linux, wow benchmarking.

Next time test with Quake 3 Linux on a CRT monitor. Think that would be more fitting towards a news post, about an iGPU that's claimed to be as fast as a 3060 mobile.

You're confusing two separate things here. That "3060 competitor" iGPU is rumored for 7000-series APUs. This 3-4CU entry iGPU is rumored for 7000-series desktop CPUs (the ones that today don't have an iGPU at all). The two will ... well, not be in the same league, regardless of how true or fales the 3060-like rumors are, as even current 6000U/H series iGPUs trounce this.

InVasManiStill the 5.21GHz clock rates are pretty impressive for Ryzen.

Yeah, that's high. I know Ryzen 7000 is rumored to clock quite high, but that's still more than I would expet for a QS/ES chip.

ValantarYou're confusing two separate things here. That "3060 competitor" iGPU is rumored for 7000-series APUs. This 3-4CU entry iGPU is rumored for 7000-series desktop CPUs (the ones that today don't have an iGPU at all). The two will ... well, not be in the same league, regardless of how true or fales the 3060-like rumors are, as even current 6000U/H series iGPUs trounce this.

Half of my point still stands. :D Even testing that such GPU down Linux on this title is laugh-worthy.

ValantarWould it? 6900HX is 12CU (@2.4GHz), this is rumored to be 3-4CU. Even if the lower CU count makes it less bandwith starved, this still delivers ~1/3rd performance in the benchmark you highlighted. For a single, very old, very odd benchmark that seems pretty close to linear scaling, especially if we also factor in lower core clocks.

Sorry, I meant twice as fast as the performance its showing in these benchmarks, not twice as fast as the 6900HX, or at least another 50% or so.

As long as they offer up a model that's faster than the 5700g on desktop, I can accept that lower-end processors may be more cuts

But otherwise, what's the point of transitioning to RDNA3 anyway?

X71200Half of my point still stands. :D Even testing that such GPU down Linux on this title is laugh-worthy.

Sure, but this most likely isn't meant as a published benchmarks, just someone with a system in for some kind of testing development who wanted to run some kind of test to see what it could do.

TheLostSwedeSorry, I meant twice as fast as the performance its showing in these benchmarks, not twice as fast as the 6900HX, or at least another 50% or so.

I understood that, but I don't see why it would be that way - the number you singled out is ~1/3rd of the 6900HX, which aligns with this iGPU being ~1/3rd-1/4th of the iGPU in the 6900HX (3/4 CUs vs. 12, comparable or lower clocks, same architecture). Or is there something I'm missing?

defaultluserAs long as they offer up a model that's faster than the 5700g on desktop, I can accept that lower-end processors may be more cuts'

Remember, this is rumored to be in the IOD for CPUs, not desktop APUs, i.e. the change in CUs is 0->3, not 4/6/8/12 ->3.

defaultluserBut otherwise, what's the point of transitioning to RDNA3 anyway?

I would assume it's supposed to be faster? Either faster, more efficient, or takes up less die area - those are the three general reasons for such transitions, though the latter is unlikely in this scenario.

ValantarSure, but this most likely isn't meant as a published benchmarks, just someone with a system in for some kind of testing development who wanted to run some kind of test to see what it could do.

Check the second picture, literally everything they benched it up with is pure cringe in today's terms. He benched with Furmark.. An idiot knows better to install Windows and test on a more suiting ground than some poor PC Mark or something I've never heard about in my life. Garbage.

ValantarI understood that, but I don't see why it would be that way - the number you singled out is ~1/3rd of the 6900HX, which aligns with this iGPU being ~1/3rd-1/4th of the iGPU in the 6900HX (3/4 CUs vs. 12, comparable or lower clocks, same architecture). Or is there something I'm missing?

I thought Zen 4 was getting an improved GPU arch with better performance, but maybe I've missed something. Yes, it has fewer CUs, but I thought they were supposed to be more on par with what AMD put inside the Steamdeck.

X71200Check the second picture, literally everything they benched it up with is pure cringe in today's terms. He benched with Furmark.. An idiot knows better to install Windows and test on a more suiting ground than some poor PC Mark or something I've never heard about in my life. Garbage.

I mostly agree that these tests aren't worth much, but again, to me this just indicates that these tests are done by someone who isn't really well versed in benchmarking at all, but rather had a chip in hand and wanted some kind of demonstration of its features. Using the Phoronix Test Suite limits them to tests included in that, which are in turn limited to tests that work across the many operating systems that test suite works on. It is in no way a modern GPU benchmarking suite, nor is it meant to be one.

I mean ... you're assuming that this person has the time, resources and freedom to install whatever OS they want to and run whatever software they want to. That is quite the assumption given that this is early hardware in very limited distribution. Most likely they used what they had - a Linux install and a widely used Linux (+many other OSes) benchmarking suite. That's all. Nobody is claiming that this is a good benchmark.

TheLostSwedeI thought Zen 4 was getting an improved GPU arch with better performance, but maybe I've missed something. Yes, it has fewer CUs, but I thought they were supposed to be more on par with what AMD put inside the Steamdeck.

I guess it might be getting RDNA3, though if these chips are to lauch in early H2, that seems unlikely to me (unless AMD has already had the RDNA3 design ready for a while). I'd expect a 6+ month delay between the first RDNA3 GPU-only product launch and any product integrating it, given history and the difficulty of coordinating these things between design teams, the time frames for die layouts and tapeouts, etc.

They likely do, it's Phoronix so if anything, you can be ensured a Linux nerd is likely to have way more free time down his hands than your average individual. Not an excuse for not actually properly testing the thing, these numbers are all literally meaningless and mis-informative at best.

X71200They likely do, it's Phoronix so if anything, you can be ensured a Linux nerd is likely to have way more free time down his hands than your average individual. Not an excuse for not actually properly testing the thing, these numbers are all literally meaningless and mis-informative at best.

... yet that seems to be what the test suite entails. So, are you suggesting that person singlehandedly update the Phoronix test suite to something more relevant for gaming? 'Cause these results do seem to be a byproduct of someone simply running that test suite, likely to gauge overall performance in a simple, ready-made manner.

Also, why on earth are "Linux nerds" more likely to have more free time than the average person? Like ... what? How does that make sense? Do you have access to some kind of survey telling you this? And you're arguing as if they have access to this PC in their free time. Again: what?

To be clear:
- This is early test hardware, either QS or ES, and definitely not available to anyone outside of labs with some degree of access control
- It might be that engineers working there are allowed to play with this hardware on their off time, in the lab, but that is quite unlikely given that such hardware is rare and difficult to get ahold of.
- There is absolutely no reason to expect a hardware engineer, linux programmer, user, sysadmin, whatever, to be a gamer, or to be particularly versed in benchmarking (or interested in it). If my impressions from these forums are anything to go by, there seems to be somewhat of an inverse correlation between people working professionally with hardware and Linux and the likelyhood that their enthusiasm for PC hardware is related to gaming. A lot of people just like to run servers, do server tasks, program, make stuff, render stuff, whatever. And the vast majority of people doing this do not know how to benchmark a PC for those tasks -that is a separate skill that is not necessary for performing those tasks.
- Not everyone working with stuff like this has it as a hobby. Some people just have a job, and do other stuff on their free time. Not every engineer is an enthusiast.

It is absolutely possible that the person running the tests is an enthusiast, and that they for some reason thought these were relevant gaming tests, but ... that seems pretty unlikely, all things considered. Incompetence is not the most likely explanation for the selection of these particular benchmarks.

Wow, half a wall text for somebody who had the time to run something as useless as Furmark for benchmarking (and a load of other useless tests), can't do ONE test that actually means something? How freaking informative!

X71200Wow, half a wall text for somebody who had the time to run something as useless as Furmark for benchmarking (and a load of other useless tests), can't do ONE test that actually means something? How freaking informative!

Wow, a response devoid of any arguments that still fundamentally misses the point. This is not a benchmark produced in order to demonstrate gaming performance. Your (or our) desires for a modern gaming benchmark were most likely not part of the reasoning behind choosing the Phoronix test suite. Arguing as if it was is unreasonable. You keep responding as if it's somehow unreasonable for this one, leaked benchmark to not conform to your desires. You're approaching this from a fundamentally misunderstood POV. It's one thing to want it to be more representative - I do too! - but another thing completely to complain on this level that one single leaked non-professional, non-gaming-oriented benchmark isn't what you're hoping for. That happens. It's expected. That's the nature of early leaks - they're always unrepresentative and weird. It wouldn't be any more helpful if this was, say, GeekBench, after all. We're going to have to wait for more leaks (or till after launch) to see more broadly representative stuff. And that's fine.

ValantarI guess it might be getting RDNA3, though if these chips are to lauch in early H2, that seems unlikely to me (unless AMD has already had the RDNA3 design ready for a while). I'd expect a 6+ month delay between the first RDNA3 GPU-only product launch and any product integrating it, given history and the difficulty of coordinating these things between design teams, the time frames for die layouts and tapeouts, etc.

I can see a case to prioritize RDNA3 for APU first with Zen4 since AMD is transitioning to DDR5. They may as well double down on both DDR5 and a newer architecture and have a greater impact. Just from the mobile market I see a huge case to prioritize getting new APU's with a new architecture out quickly.

InVasManiI can see a case to prioritize RDNA3 for APU first with Zen4 since AMD is transitioning to DDR5. They may as well double down on both DDR5 and a newer architecture and have a greater impact. Just from the mobile market I see a huge case to prioritize getting new APU's with a new architecture out quickly.

Yeah, that's a valid point, and (especially if RDNA3 delivers a marked performance increase per CU/clock) I would expect (or at least hope) for that to happen with 7000-series APUs when they launch at CES2023. I wouldn't expect the same for these IOD GPUs for the desktop CPUs though, as they just aren't performance sensitive, and thus lost likely wouldn't be worth the extra effort (and possibly die area) of pushing to get a brand-new design implemented. They way I see these, they are a "you don't need a dGPU to use the PC" solution, not something aimed at any real level of 3D acceleration.

defaultluserAs long as they offer up a model that's faster than the 5700g on desktop, I can accept that lower-end processors may be more cuts

But otherwise, what's the point of transitioning to RDNA3 anyway?

Remember that AMD is going to offer iGPUs in all of its Desktop Ryzens, it's like having iGPU even on the 5900X & 5950X but for the next gen. (eg. 7900X & 7950X).

Because these will be based on an MCM (chiplets) design and made mainly for desktop use which will eventually use a dGPU, they just added a small iGPU just to make things work, just like how intel does have iGPU on all of it's CPUs (minus HEDT).

AMD will still make monolithic die APUs, starting with laptops, and might make some for the desktops as well as the actual successor to the 5600G/5700G, lets call them the G-series. So 7600G & 7700G will come eventually as a Graphics focused APUs, these will max out at 8C as usual but will bring faster iGPU, and because these will have higher TDP than laptop versions (65W compared to 35-45W), these will definitely have more power than 5700G, as even the mobile focused versions (TDP <35-45W) is rumoured to compete against 1660/3050.

I'm interested more in the VCN Engine. As we will have a G-series for APU gaming. But seeing this to focus more in the CPU side with just basic iGPU, I wonder if they gimmicked the VCN Engine as well, especially the newer video formats like AV1 & VP9.

It could be the ideal CPU for a Plex media server, the 5600G/5700G is also good, but it doesn't support AV1, so it will depend on software decoding, I don't know if the CPU side can handle it (never used/tried AV1/VP9 before). My 5800X CPU sometimes becomes sluggish when handling high bitrate x265 decoding/encoding with Plex (the GPU is old & doesn't have HW x265), so I doubt it will be able to handle AV1/VP9.

XajelRemember that AMD is going to offer iGPUs in all of its Desktop Ryzens, it's like having iGPU even on the 5900X & 5950X but for the next gen. (eg. 7900X & 7950X).

Because these will be based on an MCM (chiplets) design and made mainly for desktop use which will eventually use a dGPU, they just added a small iGPU just to make things work, just like how intel does have iGPU on all of it's CPUs (minus HEDT).

AMD will still make monolithic die APUs, starting with laptops, and might make some for the desktops as well as the actual successor to the 5600G/5700G, lets call them the G-series. So 7600G & 7700G will come eventually as a Graphics focused APUs, these will max out at 8C as usual but will bring faster iGPU, and because these will have higher TDP than laptop versions (65W compared to 35-45W), these will definitely have more power than 5700G, as even the mobile focused versions (TDP :love:5-45W) is rumoured to compete against 1660/3050.

There are rumors of 55W laptop chips next generation (which are in turn rumored to compete with an RTX 3060 in performance, but I'll believe that when I see it), so I guess we'll see if these are still monolithic APUs or some form of MCM package. IMO there's generally an open question of how long AMD will stick with monolithic APUs, as we're nearing a point where they could start applying either LSI (TSMC's EMIB alternative) or 3D stacking CPU+GPU dice on top of an IOD in a Z-height and package size that's suitable for mobile. At some point, the monolithic APU will go away, once these interconnect technologies become sufficiently affordable and available to allow for savings through splitting up and optimizing the currently quite large APU die.

I mostly agree with you, and don't think we'll see a wholesale changeover in Ryzen 7000, but it will be coming sooner rather than later. Intel has already shown off their MCM mobile CPUs after all, and AMD can't afford to be left behind - and we know that they've been working closely with TSMC on interconnect tech for several years. What I'm more "realistically" hopeful for is that we might see an experimental MCM desktop APU in the 7000 series, where you have an IOD, CCD, and a small GPU die on the same package. Navi 24 already seems like a good candidate for this (even if it's somewhat redundant, as it wouldn't need its memory controllers in such an implementation), but if current RDNA3 rumors are correct about them moving PCIe and memory controllers to separate dice from the GPU die, a small RDNA3 die would be a shoo-in for an MCM APU with a decent CU count. Still, none of this is likely to work well or be economically feasible unless those interconnects are available in high volume and at low costs.

But again, considering what Intel is doing, and how TSMC has been promoting their progress in several different exotic chip packaging technologies for several years, I would be rather surprised if AMD didn't start adopting these in some form with Ryzen 7000. The tech should be ready by now. And given that through-package Infinity Fabric is the main barrier against mobile MCM usage (both in its high power draw and package thickness requirements), this could easily cascade into a rather dramatic reconfiguration of AMD's APU stack.