Where is this GPU innovation?

[Gat Pelsinger]

I've been always whished for if Intel and AMD just put more juice in their integrated graphics, especially that Nvidia MX series died and everybody wants a thin and light. I am writing this post especially because I just discovered that in Intel Kaby lake (8th gen), there were a series of processors that had an Intel processor with an AMD manufactured GPU on the same package. I actually knew this earlier but I had ignored it, until just now that I realized that there was a whopping HBM 2 die literally next to the GPU serving as the video memory.  I was so happy and astonished seeing this, just because I am tired of seeing I-GPUs suffer performance because of slow memory access, but this one was enjoying all the memory bandwidth, which was probably even more than desktop level graphics cards (not totally, because there was only 1 chip, but that was still a lot of bandwidth, plus all that reduction in latency).

 

This made me eager to see the benchmarks and I am absolutely bind blown. https://www.youtube.com/watch?v=XvTBghEDXHc. The Hades Canyon NUC which had the Vega graphics of that time (24 CUs, 100W TDP package, 1GHz+) was literally a quarter slower than a freaking RX 580! You might not understand this but that is a massive performance boost for integrated graphics (okay, sure I will correct my terms. It was not supposed to be "integrated graphics", a separate integrated graphics was actually still on the CPU package, instead it was viewed just like a dGPU, but that's still on the same CPU package and hence I will consider it as integrated). For context, the Skull Canyon NUC which had a 6700HQ literally didn't even belong in the charts, giving an unplayable fps in most games. 

 

I then also referred this video - https://www.youtube.com/watch?v=iE6-WshBh5g where they had the lower tier model (20 CUs) but also way lower TDP of 65W package and actually being inside a laptop rather than a NUC. That thing competed and even won against a GTX 1050! You will never see a modern I-GPU surpass the current gen least powerful discrete GPU.

 

In my understanding, a major role of this performance has to be the HBM2 memory. I get that it is very expensive, but doesn't it still belong to premium thin and light laptops? Also I think there was more juice packed into it. 24 CUs is a lot.

 

 

[da na]

In the Core 2 Duo days you'd get integrated graphics from Nvidia (and ATI, god forbid) - although sharing system memory they often had a good deal more horsepower than the Intel GMA iGPUs. Definitely would be nice to see a return of something like this...

Nvidia also tried GPUs with VRAM on the same package in like 2002 (GeForce MX4 420 Go). Not HBM mind you, but it's clearly very doable for budget cards.

[Electronics Wizardy]

Is there enough of a market to justify these parts? I feel like a 2x or 4x better iGPU would still be much slower than a low end dDGPU would be. And the current iGPUs are still enough for the vast majority of users that mostly use a browser, ms office and other low GPU usage programs. I don't see adding this making a ton of finnical sense. ESP if they have to bump up the memory bandwidth so its not a drop in replacement.

 

But this can be done, apple has done it very well with the M chips they have made. Arguably the fastest iGPUs out there.

[igormp]

37 minutes ago, Gat Pelsinger said:

but doesn't it still belong to premium thin and light laptops? Also I think there was more juice packed into it. 24 CUs is a lot.

I don't think people who are looking for such laptops care for the GPU performance. Personally I'd never buy a laptop with a power hungry GPU, I'd rather go with something that allows me to play media and attach a 4k screen for browsing while consuming the least amount of power as possible.

 

Anyhow, going back to your original question, AMD is going to release new APUs with more channels of memory in a custom design (so don't expect it in your regular ATX form factor), which should have a pretty hefty iGPU.