• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Intel DG2 GPU with 256 Execution Units Offers GTX 1050 Performance

Joined
Mar 31, 2020
Messages
1,519 (0.82/day)
We have been receiving several leaks for Intel's upcoming DG2 GPUs with a 256 Execution Unit model recently appearing on Geekbench paired with a 14-core Alder Lake mobile CPU. The Alder Lake mobile processor featured an integrated Xe GPU with 96 Execution Units which were also benchmarked. The 256 Execution Unit graphics card tested is likely derived from the mid-range DG2-384 GPU. The 96 EU DG2 iGPU featured a maximum frequency of 1.2 GHz while the 256 EU variant increased that to 1.4 GHz. The DG2-256 scored an OpenCL score of 18,450 points in the Geekbench 5 benchmark which places it at GTX 1050 performance level. The DG2-96 integrated GPU scored 6,500 points which is comparable to a GTX 460. While these performance numbers are low it is important to keep in mind that these are just early results from a mid-range mobile offering and Intel is planning to release cards with 512 Execution Units which should compete with the RTX 3070 Ti and 6700 XT.



View at TechPowerUp Main Site
 
I suppose its a good start, but they really need plans for faster than this
 
So much for RTX 3070Ti / 6800XT performance...

This has half the EUs and is only performing at GTX 1050 levels, I don't see how doubling the EUs will more than double the performance.

Edit: I just looked it up, doubling the EUs would have to quadruple the performance if they expect it to match the 3070Ti. As the 3070Ti is 4 times faster than the 1050. I don't see that happening. I'm guessing DG2-512 will be lucky to get RTX 3060 performance.
 
Last edited:
If this was released 3 years back, I think it makes sense. Otherwise, between this and a GTX 1050, I will gladly go with the GTX 1050, or step up to a GTX 1050 Ti/ 1650. One problem here is that they are looking at some benchmark numbers to determine how well it performs against an AMD or Nvidia card. In practice, both AMD and Nvidia is expected to have better driver support and may perform better in real games.

Intel needs to start somewhere, but nothing exciting at this point in time. Especially so when this card is also limited to some Intel configurations which may further affect adoption rate.
 
So much for RTX 3070Ti / 6800XT performance...

This has half the EUs and is only performing at GTX 1050 levels, I don't see how doubling the EUs will more than double the performance.

Edit: I just looked it up, doubling the EUs would have to quadruple the performance if they expect it to match the 3070Ti. As the 3070Ti is 4 times faster than the 1050. I don't see that happening. I'm guessing DG2-512 will be lucky to get RTX 3060 performance.

Intel could clock the snot out of the core to get it up to 3070 Ti performance. It's what they've been doing with their CPUs, after all. Of course, all efficiency would be completely thrown out the window if they do take this route. But hey, it's not like the competition is known for its power efficiency either, right? :)
 
So much for RTX 3070Ti / 6800XT performance...

This has half the EUs and is only performing at GTX 1050 levels, I don't see how doubling the EUs will more than double the performance.

Edit: I just looked it up, doubling the EUs would have to quadruple the performance if they expect it to match the 3070Ti. As the 3070Ti is 4 times faster than the 1050. I don't see that happening. I'm guessing DG2-512 will be lucky to get RTX 3060 performance.
Agree. Or there is something we don't know?
 
So much for RTX 3070Ti / 6800XT performance...

This has half the EUs and is only performing at GTX 1050 levels, I don't see how doubling the EUs will more than double the performance.

Edit: I just looked it up, doubling the EUs would have to quadruple the performance if they expect it to match the 3070Ti. As the 3070Ti is 4 times faster than the 1050. I don't see that happening. I'm guessing DG2-512 will be lucky to get RTX 3060 performance.
3070ti is not 4 times faster than 1050, even gtx 1080ti is 6-10 times faster in different titles, so 3070 ti is at least 10 times faster on average.
 
Last edited:
The 1050 is 5 years old at this point. The drivers are a diamond crushing through rocks.
 
and Intel is planning to release cards with 512 Execution Units which should compete with the RTX 3070 Ti and 6700 XT.
Just not believable. 3060 like performance? That's believeable, but if a 96EU part is getting 1050 like performance then there is no way that a 512EU part will get 3070ti levels unless the clocks are through the roof.

The drivers are a diamond crushing through rocks.
And that means what?
 
since larrabee failure intel keep repeating the same thing over the years 2 years ago and now still same thing but no products being made instead of slide, oh wait there are few in reviewer from german or youtube and they cant even play any game with crappy driver . the more i read about intel dgpu the more i believe in unicorn exist
 
Intel is pathetic. A R9 290 from 2013 beats on the 1050 ti. Intel is a false start.
 
AMD and Nvidia have spent years optimizing their drivers for DX11/DX12/Vulkan, synthetic benchmark is just pointless indicator of performance across a wide variety of games.
That means DG2-512 could be as fast as 3070/6700XT in synthetic while limping along 3060 in actual games, or worse.
 
Intel is pathetic. A R9 290 from 2013 beats on the 1050 ti. Intel is a false start.
I'm in a "wait & see" frame of mind. The DG1, for the market it's aimed at, hit the mark spot-on. Saying that the DG2 will be a 3070ti competitor is more ambitious then I am willing to accept. However, credit where it's due, with the DG1, Intel succeeds where it wanted to. Willing to give the benefit of the doubt, but only in a reasonable way. I think we can expect something good from DG2.
 
The fact is that it takes them absurdly long to perform a simple copy paste of said execution units, well, not as simple, but how hard can it be, they have the basic structure. in this case could be downgaded to DDR4 for all we know, a 192 bit card using GDDR6 has to be better than 3060 at least.
 
The fact is that it takes them absurdly long to perform a simple copy paste of said execution units, well, not as simple, but how hard can it be, they have the basic structure. in this case could be downgaded to DDR4 for all we know, a 192 bit card using GDDR6 has to be better than 3060 at least.
they're in panic mode and about to lose more loyal customers + fan bois if the upcoming alderlake fail it would be a fatal blow
 
So much for RTX 3070Ti / 6800XT performance...

This has half the EUs and is only performing at GTX 1050 levels, I don't see how doubling the EUs will more than double the performance.

Edit: I just looked it up, doubling the EUs would have to quadruple the performance if they expect it to match the 3070Ti. As the 3070Ti is 4 times faster than the 1050. I don't see that happening. I'm guessing DG2-512 will be lucky to get RTX 3060 performance.

It is possible.
Just double the EUs to 512
Then juice up to 3.0GHz
3070Ti performance right there.

Please don't mind the under the table chiller, thx
 
It is possible.
Just double the EUs to 512
Then juice up to 3.0GHz
3070Ti performance right there.

Please don't mind the under the table compressor unit, thx
You do realise that would make it a 400w card that competes with a 220w one & I don`t think it would win again the 3070Ti. But we shall see when independent testing will be done.
 
You do realise that would make it a 400w card that competes with a 220w one & I don`t think it would win again the 3070Ti. But we shall see when independent testing will be done.
Yes.

We have seen this in their W-3175X "Demo" already.
Use a 1000W CPU with a chiller to fight competitor's 250W offerings
 
I dont really understand all this negativity, this is basically the same nonsense AMD tends to be the victim off.

Rediculous claims that get people hyped: "6ghz out of the box, 10 cores for just 270 dollar and 50% ipc gains!!" and then when its "only" 4ghz for 8 cores at 350 dollars with 20% ipc gains, even though it would be historically a fantastic evolution, people will be disappointed because it does not match the rediculous hype.
And anti AMD fanboys will use this borderline self fufilling prophesy to talk crap about it "hehe so much for that 6ghz claim right? cant trust AMD guys, stick with Intel omegalulz".

Pretty sure Intel never said it would match the 6700xt or 3070(ti), that is just speculation from (probably amongst others) Morse's Law is Dead, so take that with a mountain of salt.

Sure a gpu that matches a 1050 is hardly something interesting for most of us, but if the big version can do RTX3060/RX5700xt performance for 260 bucks, that would be great.
 
I dont really understand all this negativity, this is basically the same nonsense AMD tends to be the victim off.

Rediculous claims that get people hyped: "6ghz out of the box, 10 cores for just 270 dollar and 50% ipc gains!!" and then when its "only" 4ghz for 8 cores at 350 dollars with 20% ipc gains, even though it would be historically a fantastic evolution, people will be disappointed because it does not match the rediculous hype.
And anti AMD fanboys will use this borderline self fufilling prophesy to talk crap about it "hehe so much for that 6ghz claim right? cant trust AMD guys, stick with Intel omegalulz".

Pretty sure Intel never said it would match the 6700xt or 3070(ti), that is just speculation from (probably amongst others) Morse's Law is Dead, so take that with a mountain of salt.

Sure a gpu that matches a 1050 is hardly something interesting for most of us, but if the big version can do RTX3060/RX5700xt performance for 260 bucks, that would be great.
You know, I`m sure most of us are well informed about the semiconductor industry, so as usual hype is a factor like it`s always been for the last 10 years or so. My expectations have been tempered after so many disappointments over the years from all 3 camps.(AMD, Intel & Nvidia)
 
I just took a look at this OpenCL Benchmark site and chart. Intel Iris Xe MAX, with 96 EU I think, scores 22236 points [https://browser.geekbench.com/opencl-benchmarks], so for the final 256 EU GPU we can expect a much higher performance than this early sample (or at least most of us should hope so).
 
is it 256 EU or 2 56 alu, if 256 EU its a PoS
 
If those benchmarks are real, the 2X of this die would perform up to RTX2060/RX5700 lvl of performance even when running @2,3GHz. Let's hope for the best for the GPU market.
 
Back
Top