News Intel GPUs - are the drivers getting better yet? (Maybe? Please?)

[KompuKare]

Performance varies wildly from game to game.

And that has been the trouble with ARC since launch (and was predictable even before then).
Still, for those who like, erm, interesting hardware is ideal

 

[The red spirit]

And that has been the trouble with ARC since launch (and was predictable even before then).
Still, for those who like, erm, interesting hardware is ideal

I wonder if it's due to hardware architecture or drivers. Seems like it could be either, but if it's due to hardware, then it will never be fixed.

 

[mikk]

I wonder if it's due to hardware architecture or drivers. Seems like it could be either, but if it's due to hardware, then it will never be fixed.

It's a combination but depends on the API. On dx9 and dx11 it's mainly driver issue considering how the DXVK translation layer is able to improve the performance a lot in many games. DX12 is certainly closer to the real hardware capability, even though even the DX12 driver can't be as good as Nvidias. There are some flaws beside the driver bottleneck, the memory controller doesn't look that good.

 

[igor_kavinski]

https://www.techradar.com/news/intel-arc-a770-somehow-beats-the-nvidia-rtx-4090-in-8k-video-playback-tests

Well, they did SOMETHING right, it seems.

Hopefully Radeon 7000 series will also be good in this regard.

 

[João Bortolace]

Xe HPG whitepaper https://www.intel.com/content/www/us/en/developer/articles/technical/introduction-to-the-xe-hpg-architecture.html

 

[bullzz]

New Intel Driver Delivers Up To 8 Percent Performance Uplift On Arc GPUs | Tom's Hardware (tomshardware.com)

 

[clemsyn]

Looks like Intel is busy improving their drivers Good sign for competition in this space IMO.

 

[igor_kavinski]

https://twitter.com/i/web/status/1594775671112622080

They have definitely got interns working for them. How else do you mess up basic WIN32 GUI stuff?

 

[GodisanAtheist]

This popped into my head with the release of NV's next gen and the impending release of AMD's next gen:

 

[gdansk]

Less to do with next generation cards and more to do with the RX 6600 (XT) and RX 6650 XT prices.

 

[NTMBK]

This popped into my head with the release of NV's next gen and the impending release of AMD's next gen:

View attachment 72382

Sorry what, Intel made graphics cards? Nope, no idea what you mean

 

[IntelUser2000]

December 5 Beta driver: https://downloadmirror.intel.com/761751/ReleaseNotes_101.3959_BETA.pdf

The latest driver has optimizations for DX9/11/12 games. Larger than normal update.

Update: https://www.tomshardware.com/news/intel-arc-gpus-get-18x-higher-directx-9-performance-from-new-gpu-driver

Intel claims up to 80% gain in DX9 games with the new driver.

 

[mikk]

It's a hybrid dx9 approach with this driver, some games run dx9 native and some still use the d3don12 layer. There are big improvement even including non native dx9 games. Biggest gains come from CPU bound situations because ot the much imrpoved overhead and also less stuttering.

 

[IntelUser2000]

Some(I am guessing) DX11 and 12 performance improves as well. One reddit member was reporting 5-10 fps improvement in MW2.

Yes, Starcraft 2 gains should be huge because there aren't many games that are as CPU bound. For Counterstrike, remember the A770 performing identically to the A380? That's how bad it was.

 

[Dribble]

This popped into my head with the release of NV's next gen and the impending release of AMD's next gen:

View attachment 72382

JPR Descrete Market share
Other than AMD's terrible performance, the big news in this is Intel got 4% market share, I guess that's ARC?

 

[IntelUser2000]

JPR Descrete Market share
Other than AMD's terrible performance, the big news in this is Intel got 4% market share, I guess that's ARC?

No doubt that's ARC. They have 1% with A380 in Q2, and 4% with proper A770/750 lineup in Q3, that's in desktop. For Notebooks it goes from 9% to 5%, which makes sense as well since A380 mobile came out before desktop but no high end ARC, which results in share drop. Overall it goes from 5% in Q2 to 4% in Q2.

Ouch! That explains AMD's client share drop.

They went from having a 20% of 84 million market in Q2 to 12% of 75 million. That's 16.8 million units to 9 million, or 46% drop. Meanwhile, Intel went from 62% of 84 million to 72% of 75 million. That's 52 million to 54 million, meaning they gained volume in Q3.

If you just look at iGPUs.

AMD: 22% of 65 million in Q2 which is 14.3 million, to 13% of 62 million which is 8.06 million, which is a 44% drop.
Intel: 78% of 65 million in Q2 50.7 million, to 87% of 62 million which is 53.94 million, again a gain in volume share.

Looks like both Intel and Nvidia took AMD's share of the dGPU pie.

Desktop: Nvidia 7% gain in Q3 over Q2, Intel 3%.
Notebook: Nvidia 9% gain in Q3 over Q2, Intel -4%*
Overall: Nvidia 8% gain in Q3 over Q2, Intel -1%

*I assume the Intel loss in Q3 over Q2 is because no real A770M/A750M devices out there. Either OEMs are not using or Intel is not ready. Since there's no reviews the latter could be the case.

But remember prior to Q2 there was no Intel ARC. So Y/Y shows....

Desktop: Nvidia 7% gain in Q3 2022 over Q2 2021, Intel gained 4%.
Notebook: Nvidia 3% gain in Q3 2022 over Q2 2021, Intel 5%
Overall: Nvidia 5% gain in Q3 2022 over Q2 2021, Intel 4%

That all comes at the loss of AMD. 21% to 10% for Desktops Q3 2022 to Q2 2021, 13% to 5% in Notebooks, and Overall 17% to 8%.

AMD's marketing must be real crap. I was expecting Intel to take Nvidia's share, not AMD. Also I think the "drop" in Nvidia's share between Q2 2022 and Q2 2021 is due to Nvidia having massive crypto mining share, but in Q2/Q3 the slump kicked in which resulted in Nvidia gaining it among gamers.

JPR also made a mistake of copying Desktop iGPU result as a Notebook iGPU result. What idiots!

 

[guidryp]

December 5 Beta driver: https://downloadmirror.intel.com/761751/ReleaseNotes_101.3959_BETA.pdf

The latest driver has optimizations for DX9/11/12 games. Larger than normal update.

Update: https://www.tomshardware.com/news/intel-arc-gpus-get-18x-higher-directx-9-performance-from-new-gpu-driver

Intel claims up to 80% gain in DX9 games with the new driver.

Previously I posted that you could get those kinds of gains switching to DXVK, and it looks like that Intel did that themselves:

https://downloadmirror.intel.com/761751/readme.txt

------------------------------------------------
SOFTWARE: dxvk
The zlib/libpng License

Copyright (c) 2017-2021 Philip Rebohle
Copyright (c) 2019-2021 Joshua Ashton
This software is provided 'as-is', without any express or implied warranty. In no event will the authors be held liable for any damages arising from the use of this software.

I wondered why they were sticking with the crappy Microsoft implementation, when DXVK was performing so much better.

 

[IntelUser2000]

I wondered why they were sticking with the crappy Microsoft implementation, when DXVK was performing so much better.

They said they are using a hybrid method.

I assume it has to do with compatibility issues in some cases? It's not better in everything. Like someone said on the ARC reddit Skyrim/Oblivion(I can't bother to remember) that it runs bad with DXVK and the DX11 version runs much better.

The new software organization seems to be having a great effect. The Intel of old never got to properly fixing graphics related bugs until the new gen came out with it's own bugs.

Anyway yes AMD's graphics revenue did crater in Q3 over Q2, because I forgot "Gaming" includes console revenue. So Intel has equal volume share to AMD in mobile, about 40% in desktop, and in overall half of AMD for GPU.

Although I know analyst numbers can be quite off. They are more of a ballpark number than anything.

 

[Dayman1225]

So I was able to retest my A770 in CSGO after the new driver.
CPU is i9 9900K, ran at 1440p on these settings

https://twitter.com/i/web/status/1600311570076438529

 

[mikk]

Can you try 1080p noAA in CSGO? I think you should get at least 2x average fps in a CPU bound scenario.

 

[Dayman1225]

Can you try 1080p noAA in CSGO? I think you should get at least 2x average fps in a CPU bound scenario.

Yeah ill give it a try tomorrow once I have some time

 

[IntelUser2000]

I read that some games treat DXVK as malware or antimalware flags it. So that could be the reason they had to wait and get things fixed rather than simply use DXVK. And also the gains are not universal.

I find it interesting how in S|A's article it said most of the early driver development was done in Russia and the game it performs well in is Metro Exodus.

 

[GunsMadeAmericaFree]

>>Intel claims up to 80% gain in DX9 games with the new driver.

Dang, that's quite an improvement. The really poor performance on older games was one of the main things that gave me pause about these things. Now I just need to wait for a sale/rebate on one of the lower end ones.

 

[Tup3x]

I wonder if Portal RTX works and if it runs better than on AMD...

 

[mikk]

A770 basically on par with AMD/Nvidia in a CPU bound setting in CSGO with the new driver. In the higher WQHD resolution there is a slightly bigger dop off but nothing major.

Intel Arc 31.0.101.3959: Neuer Treiber soll die Leistung unter DirectX 9 verbessern

Intel bessert bei den Treibern für die Arc-Grafikkarten stetig nach. Nun soll insbesondere die Leistung unter DirectX 9 steigen.

www.computerbase.de