News Intel GPUs - waiting for B770

[DrMrLordX]

Is Intel joining AMD in the open source movement?

Arguably they're already there (see: Intel's Linux distro, their SYCL efforts, and a lot of other stuff).

 

[beginner99]

Arguably they're already there (see: Intel's Linux distro, their SYCL efforts, and a lot of other stuff).

Like most libraries data science related now come with a "warning" along the lines of "this build is optimized with intel oneapi for AVX AVX2". So yeah they are putting their oneapi in things already where AMD after a decade is still absent from.

It's kind of ironic as intel in general is also pretty big software house and AMD traditional a hardware company. So I would have expected intel to be able to deliver good drivers. (but we don't know if the chips are so buggy that driver development gets to be a chore?Or they simply didn't get enough budget. time and heads-up? Intel should have poached at least one senior person from NV or AMD driver team...also for that guys connection to the game developers for optimizations.

 

[IntelUser2000]

It's kind of ironic as intel in general is also pretty big software house and AMD traditional a hardware company. So I would have expected intel to be able to deliver good drivers. (but we don't know if the chips are so buggy that driver development gets to be a chore? Or they simply didn't get enough budget. time and heads-up? Intel should have poached at least one senior person from NV or AMD driver team...also for that guys connection to the game developers for optimizations.

They simply didn't care or felt the need to since it was iGPUs. Now they will have to because the problem is much closer to them if people complain. Laptop iGPUs it's like a thermal issue or lack of updates from the manufacturer or firmware not cooperating with each other. None of that for dGPU. Bad performance for drivers is entirely Intel's fault period. Also the target audience is more critical such as us.

Look how Cyberpunk was one of the problematic games for Iris Xe but a featured one for ARC. Remember at one point they likely wanted graphics to be like HD audio. I don't think it's a concidence low end of their iGPUs are branded HD Graphics. A function secondary to CPUs and be absorbed in it. All those headcount and driver teams don't matter if the direction isn't to improve gaming.

Performance is related to it too. 30 fps in Cyberpunk at minimum settings. Really not a big deal to have small glitches compared to say, an A770 getting 80+ fps at 1440p Ultra.

That's why everyone of their previous efforts didn't matter. They called Iris Xe like it was super good at gaming. Yea, it's way better but still pretty much the lowest end you can get and run AAA titles at lowest settings. dGPUs will change that quickly if they are serious about staying in the market, nevermind compete!

 

[moinmoin]

They simply didn't care or felt the need to since it was iGPUs. Now they will have to because the problem is much closer to them if people complain.

What I find odd is that the driver team could have started getting their drivers up to speed back when Intel announced to develop discrete GPUs almost 5 years ago. The current troubles make it seem Intel started to care about drivers way later.

 

[mikk]

Apparently on Wedneseday we will see A380 reviews from Computerbase, Golem, Igorslab.

But more on that in a moment. First of all, in this context, I would like to announce our official “launch” of the Gunnir Arc A380 for Wednesday 07/20/2022 at 12:00.

One sample went to Computerbase, one to Golem, one to a forum member from the 3D Center and one thankfully to me. And since we all want to test collegially with each other and not against each other, we have agreed on a common time when we will all go online at the same time. And we also agreed on the content in order to complement each other and not to overlap in a big way. Everyone does what they do best. That’s why you should also read the other two articles on Wednesday to get a really full picture.

Intel’s Arc A380 in the lab, new premium fans, 1400 watts chased through a GeForce RTX 3090 Ti, and an honorable mention from the other hemisphere | igor´sLAB

For once, we’ll start the week with a news article as a description of the current state and a preview of what you can expect from tomorrow onwards, which has already robbed me of the whole weekend.

www.igorslab.de

Here is a russian review:

OC A380 is quite a bit faster than the default A380.

 

[NTMBK]

Apparently on Wedneseday we will see A380 reviews from Computerbase, Golem, Igorslab.

No Anandtech review?

 

[KompuKare]

No Anandtech review?

Ha, ha, ha!

Computerbase should be a good review. Igor I sometimes watch his video reviews as it's sort of fun to watch the way he rambles along - eventually he does get to the point; I know lots of video reviewers do the same to get longer video but he also writes like that.

 

[igor_kavinski]

What I find odd is that the driver team could have started getting their drivers up to speed back when Intel announced to develop discrete GPUs almost 5 years ago. The current troubles make it seem Intel started to care about drivers way later.

Or they seem to have focused on 3DMARK and synthetics way too much. Once they got those running fine, they must have expected everything to just start working. It did work but it was slow coz shaders in games are targeting AMD/Nvidia hardware. Hopefully, we will get a game developer's opinion on Intel's dGPU at some point. Should be very interesting to hear how it differs from AMD/Nvidia GPUs and how those differences impact performance.

 

[igor_kavinski]

OC A380 is quite a bit faster than the default A380.

What's weird is that a roughly 100MHz GPU OC is adding almost 10-30 fps to the average. Is the RAM also overclocked? Doom Eternal OC improvement is pretty good.

 

[biostud]

I am looking forward to seeing how it all plays out. I have systems for older games. So I would not hesitate to buy an Intel ARC for new ones, if they are the best bang for buck.

The main problem as I see it, that newer games require more performance, and one of the best thing about having a mid/low range videocard is the huge backlog of games that plays really well.

 

[NTMBK]

Or they seem to have focused on 3DMARK and synthetics way too much. Once they got those running fine, they must have expected everything to just start working. It did work but it was slow coz shaders in games are targeting AMD/Nvidia hardware. Hopefully, we will get a game developer's opinion on Intel's dGPU at some point. Should be very interesting to hear how it differs from AMD/Nvidia GPUs and how those differences impact performance.

I'm sure the Intel engineers knew exactly what to expect- they can read tech blogs too, and they do in fact write GPU drivers for a living

There's not much point working on game-specific optimizations when the fundamentals of the driver are broken. You need to get the driver correct for all functionality, and performing well on baseline non-optimized code. Then you can invest the time in game-specific optimizations, without building those optimizations on top of shifting sands.

 

[Tup3x]

What's weird is that a roughly 100MHz GPU OC is adding almost 10-30 fps to the average. Is the RAM also overclocked? Doom Eternal OC improvement is pretty good.

I find this very weird. Power consumption OC vs. no OC.

Left is stock A380, center is A380 OC and right is GTX 1650.
And Cyberpunk...

Yeah.... it looks like A380 might not be too bad after all. Their default power limits are extremely weird by the looks of it.

 

[Hitman928]

I find this very weird. Power consumption OC vs. no OC.
View attachment 64693
Left is stock A380, center is A380 OC and right is GTX 1650.
And Cyberpunk...
View attachment 64694
Yeah.... it looks like A380 might not be too bad after all. Their default power limits are extremely weird by the looks of it.

How are they getting a 23% - 40% increase in performance with a 6% increase in clock frequency? Power increase for the shown clock increase also seems. . . weird.

 

[igor_kavinski]

And Cyberpunk...
View attachment 64694

That shows the non-OC ARC GPU at 100MHz. Something is definitely amiss.

Also, the OC GPU is also placing more demands on the CPU.

 

[maddie]

Trust the tools too much?

 

[Tup3x]

How are they getting a 23% - 40% increase in performance with a 6% increase in clock frequency? Power increase for the shown clock increase also seems. . . weird.

Didn't the DG1 have some weird power issues? There's clearly something weird power related issues going on.

Of course, it could be that the wattage reported is totally off and increasing the power limit actually does wonders....

 

[eek2121]

What I find odd is that the driver team could have started getting their drivers up to speed back when Intel announced to develop discrete GPUs almost 5 years ago. The current troubles make it seem Intel started to care about drivers way later.

How would they? At minimum, they couldn’t start until the initial design was complete. Then maybe they could start, however they would still run into roadblocks as the hardware goes through revisions. Remember all their previous stuff was completely different.

Also, windows drivers, especially graphics drivers, are challenging to write. Last I checked, the DDK only included a basic VGA driver as an example.

Do we know when the 750 launches in the US? I might pick one up for fun to put in a Threadripper machine I have. Also curious if anyone tested RT performance…

 

[IntelUser2000]

That shows the non-OC ARC GPU at 100MHz. Something is definitely amiss.

Also, the OC GPU is also placing more demands on the CPU.

That's just because it's running the game much faster so the bottleneck gets shifted more to the CPU.

The results are very interesting. The slight overclock resulting that much improvement suggests there are immature power management that gets disabled under overclock. Similar to DTT.

Reminded me of Celeron 300A. Not much at base but people got it for OC performance.

 

[SPBHM]

the Resizable BAR reliance from the a380 to perform is pretty disappointing from an old system upgrade point of view, performance overall is not that terrible.

 

[IntelUser2000]

I looked at it again and the non-OC one goes from max clocks to 100MHz. So obviously it's power management being wonky again just like the laptop DTT review. OC a slight bit would turn them off, dramatically improving the performance.

Also it's showing that in all except RSS. The utilities can't show fast enough but still long enough to show the 100MHz frequency.

Correction: Not a slight power difference. So it is indeed a power management problem. Still quite low though.

 

[IntelUser2000]

Ok, so Tomshardware explains the video.

What he did change were the power targets and voltage offsets in Intel's own graphics utility, with a setting of 55% on the "GPU performance Boost" slider and a +0.255mv voltage offset.

This reminds me of what Tom Peterson said. So at default the power is too low. 55% power increase results in 40% performance improvement, ranging 25% to 60%. That's weird because it's under 40W lot of the time. Why did Intel make that decision?

 

[VirtualLarry]

So at default the power is too low. That's weird because it's under 40W lot of the time. Why did Intel make that decision?

Doesn't Intel normally design their desktop platforms, to be laptop-first, desktop-second, most of the time?

 

[IntelUser2000]

Doesn't Intel normally design their desktop platforms, to be laptop-first, desktop-second, most of the time?

Maybe this is between them and Gunnir. Screw that. When are we going to see Asus? Gigabyte? MSI? Heck even Zotac and Galax none of this "Gunnir" nonsense.

the Resizable BAR reliance from the a380 to perform is pretty disappointing from an old system upgrade point of view, performance overall is not that terrible.

Rebar "issue" also looks like a power related issue. It's not ramping as it should without Rebar off. So logically the solution is to increase the power limits to be in line with Rebar on.

Watch Dogs
Rebar off: 30-35W
Rebar on: 35-40W
Rebar on + OC: 50W

 

[Stuka87]

Rebar "issue" also looks like a power related issue. It's not ramping as it should without Rebar off. So logically the solution is to increase the power limits to be in line with Rebar on.

Watch Dogs
Rebar off: 30-35W
Rebar on: 35-40W
Rebar on + OC: 50W

This can easily be explained by the GPU having issues loading data with Rebar off. If no-rebar induces a bottleneck, the wattage will be lower as the GPU will not be able to push as many frames.

 

[moinmoin]

How would they? At minimum, they couldn’t start until the initial design was complete.

Both the hardware and the drivers base on the iGPUs which exist for well over a decade already. Improving the drivers is something Intel could have done starting from the announcement. At first the opposite happened instead.