Monday, January 24th 2022

Intel Arc Alchemist DG2 GPU Memory Configurations Leak

Intel's upcoming Arc Alchemist lineup of discrete graphics cards generates a lot of attention from consumers. Leaks of these cards' performance and detailed specifications appear more and more as we enter the countdown to the launch day, which is sometime in Q1 of this year. Today, we managed to see a slide from @9950pro on Twitter that shows the laptop memory configuration of Intel's DG2 GPU. As the picture suggests, we can see that the top-end SKU1 with 512 EUs supports a 16 GB capacity of GDDR6 memory that runs at 16 Gbps speeds. The memory runs on a 256-bit bus and generates 512 GB/s bandwidth while having eight VRAM modules present.

When it comes to SKU2, which is a variant with 384 EUs, this configuration supports six VRAM modules on a 192-bit bus, running at 16 Gbps speeds. They generate a total capacity of 12 GBs and a bandwidth of 384 GB/s. We have SKU3 DG2 GPU going down the stack, featuring 256 EUs, four VRAM modules on a 128-bit bus, 8 GB capacity, and a 256 GB/s bandwidth. And last but not least, the smallest DG2 variants come in the form of SKU4 and SKU5, feating 128 EUs and 96 EUs, respectively. Intel envisions these lower-end SKUs with two VRAM modules on a 64-bit bus, and this time slower GDDR6 memory running at 14 Gbps. They are paired with 4 GB of total capacity, and the total bandwidth comes down to 112 GB/s.
Source: @9550pro (Twitter)
Add your own comment

29 Comments on Intel Arc Alchemist DG2 GPU Memory Configurations Leak

#26
AlwaysHope
MusselsThese intel GPU's dont even need to be great, or price competitive
Why?


Because intel will force them into hundreds of thousands of prebuilt systems and force them upon users with their usual shady business practises

(I look forward to finding out they only work on intel systems with 11th and 12th gen CPU's or some other such nonsense)
Ahhh, that's the beauty of having a 2 horse race isn't it? I mean AMD or Intel, take your pick! :(:ohwell:
Posted on Reply
#27
watzupken
Vayra86Dude, stop lying, you know these guys are all going to get fired or sued for releasing something that's under NDA. "Oh! Woopsie!"

@Fouquin see the point now?



Not bad? That's 100 Gbps short of a 3070ti and about equal to a 3070.
So the bestest, most optimistic outlook if Intel has near-perfect drivers is that their most expensive chip with 16GB (!) will end up somewhere along 3070 but likely a good 10% under it because its just not quite as refined.

I hope Intel is not looking for any more than 600 bucks for that, because if they do, and they also postpone beyond Q1 2022, its a DOA and you can easily wait until 2023 for something better. And by then, 500Gbps is lower midrange territory. So... Intel gonna scale up to 384 ~ 512 bit then? Hey Raja... did you think of using HBM? Or still didn't make up your mind? :rolleyes::rockout::toast::oops:

I'm not even half kidding. DG2 looks like old news already, its like Raja still has no grasp of time to market and pre empting that delay.

I mean.. this is just in:

In my own experience trying out the RTX 3070 Ti, the higher bandwidth as compared to the 3070 did not make a material difference in game performance. I've reduced the clock speed of the VRAM down and did not see a drastic drop in performance. The bulk of the sub 10% improvement in average performance is somewhat contributed by the increase in CUDA cores and likely less of the memory bandwidth increase. Cards at this range are meant to be solid 1440p performer, and may suit some games in 4K. So there is little concern about the memory bandwidth in my opinion.
MusselsThese intel GPU's dont even need to be great, or price competitive
Why?


Because intel will force them into hundreds of thousands of prebuilt systems and force them upon users with their usual shady business practises

(I look forward to finding out they only work on intel systems with 11th and 12th gen CPU's or some other such nonsense)
This is possible, but I don't think Intel will be that silly by locking AMD and Nvidia GPUs out of the ecosystem. They know very well that most of the dedicated GPUs are from Nvidia, and by tying Intel CPUs to Intel GPUs, people will just deflect to AMD based systems. In the end, there is a potential that they may lose sales on both CPU and GPU.
Posted on Reply
#28
Mussels
Freshwater Moderator
watzupkenIn my own experience trying out the RTX 3070 Ti, the higher bandwidth as compared to the 3070 did not make a material difference in game performance. I've reduced the clock speed of the VRAM down and did not see a drastic drop in performance. The bulk of the sub 10% improvement in average performance is somewhat contributed by the increase in CUDA cores and likely less of the memory bandwidth increase. Cards at this range are meant to be solid 1440p performer, and may suit some games in 4K. So there is little concern about the memory bandwidth in my opinion.
Semi-relevant, but i found overclocking the VRAM on my 3090 helped massively with mining, but was a performance loss in gaming - because it ate into the power limit of the GPU.

With a raised or unlimited TDP it was different, but there genuinely is a point that higher VRAM clock speeds is a negative.
Posted on Reply
#29
Vayra86
watzupkenIn my own experience trying out the RTX 3070 Ti, the higher bandwidth as compared to the 3070 did not make a material difference in game performance. I've reduced the clock speed of the VRAM down and did not see a drastic drop in performance. The bulk of the sub 10% improvement in average performance is somewhat contributed by the increase in CUDA cores and likely less of the memory bandwidth increase. Cards at this range are meant to be solid 1440p performer, and may suit some games in 4K. So there is little concern about the memory bandwidth in my opinion.
I hear you, but an Nvidia GPU is not an Intel GPU. Nvidia has deployed several things over the years to reduce the pressure on VRAM (dynamically). Its why they think they can make do with less than the competition, too. Nvidia has always had tighter bus / capacity / etc. And works around it. We're seeing more and more trickery applied that is dynamic, so its becoming more and more of a black box. Your FPS might be similar, but are you seeing the same detail?

As much as a 'core' in one architecture isn't the same as in another, bandwidth or even capacity similarly isn't the whole story if you compare different architectures. And even on different situations/games. But there are no true magic bullets here anyway; if Intel thinks it can make do with 500Gbps, we know they're limiting themselves to that bandwidth, and until we hear of special technology to work around it, it is what it is. A low number given the supposed performance level. AMD has infinity cache, for example, to make up for the X they lack in their GDDR6 ;)
Posted on Reply
Add your own comment
Oct 18th, 2024 08:43 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts