News Intel GPUs - Intel launches A580

[eek2121]

I have become a little concerned because intel keeps stating 3070-ish performance.
Seems like nVidia is ready for a possible 3x performance increase. Basically nVidia can pull out just enough performance to easily double intels best which would suck however having more cards & a third player is still welcome in my opinion.

Do note that neither NVIDIA nor AMD are expected to launch new GPUs before October 2022. Intel’s GPUs will have been out for 10 months at that point.

 

[VirtualLarry]

Do note that neither NVIDIA nor AMD are expected to launch new GPUs before October 2022. Intel’s GPUs will have been out for 10 months at that point.

Navi24? Watch some recent MLID videos.

 

[stebler]

https://twitter.com/x/status/1471465888876736515

This driver contains a lot of information related to upcoming GPUs. For now I'm just posting the raw output of strings.

Device ID and SKU info not defined for ELG_X2 yet, using X3 Device ID
GenRastyAdapterInfo::<lambda_572fc71ba3bc47791fb46288d94b5934>::operator ()
D:\qb\workspace\21461\source\gfx-driver\Source\Rasty\FakeKMD\FakeKMDAdapterInfo.cpp
Device ID and SKU info not defined for ELG_X4 yet, using X3 Device ID
GenRastyAdapterInfo::<lambda_fe00de434ef88e60962a1c73ca3e3804>::operator ()
RASTY_HW_PLATFORM
gen11-lp
RASTY_HW_SKU_STEPPING
RASTY_HW_SKU
gen9
skl-gt2
skl-gt3e
Intel Rasty Renderer - SKL GT3e / Gen 9
skl-gt4e
Intel Rasty Renderer - SKL GT4e / Gen 9
kbl-gt2
Intel Rasty Renderer - KBL GT2 / Gen 9
cfl-gt2
Intel Rasty Renderer - CFL GT2 / Gen 9
Intel Rasty Renderer - SKL GT2 / Gen 9
Invalid %s SKU: %s
GenRastyAdapterInfo
gen11
gen11-hp
Rasty no long supports Gen11-HP
Intel Rasty Renderer - ICL GT2 / Gen 11-LP (%s)
ICL_1x4x8
ICL_1x6x8
ICL_1x8x8
gen12-hp
xe_hp
xe_hpg
xe_lpg
xe_lpgplus
dg2_8x4x16
Intel Rasty Renderer - DG2_8x4x16 / Xe HPG (%s)
dg2_2x4x16
Intel Rasty Renderer - DG2_2x4x16 / Xe HPG (%s)
Intel Rasty Renderer - ATS / Xe HP (%s)
ATS devices no longer supported by uSim launcher on Windows
mtl_2x4x16
MTL_2x4x16
mtl_3x4x16
MTL_3x4x16
mtl_1x4x16
MTL_1x4x16
Intel Rasty Renderer - %s / Xe LPG (%s)
arl_6x4x16
Intel Rasty Renderer - ARL_6x4x16 / Xe LPG Plus (%s)
gen12-lp
xe_lp
Intel Rasty Renderer - RKL_1x2x16 / Xe LP (%s)
rkl_1x2x16
Intel Rasty Renderer - DG1_1x6x16 / Xe LP (%s)
dg1_1x6x16
tgl-gt2
Intel Rasty Renderer - TGL_1x6x16 / Xe LP (%s)
tgl_1x6x16
Unsupported TGL LP GT.
gen12_9
xe2_hpg
elg_x1_2x4
ELG_x1_2x4
elg_x2_5x4
ELG_x2_5x4
elg_x4_10x4
ELG_x4_10x4
elg_x3_7x4
wave2_5
ELG_x3_7x4
Unsupported XE2 HPG SKU: %s
Intel Rasty Renderer - %s / Xe2 HPG (%s)
xe2_lpg
lnl_2x4
LNL_2x4
Unsupported XE2 LPG SKU: %s
Intel Rasty Renderer - %s / Xe2 LPG (%s)
gen13
Intel Rasty Renderer - Take the red pill / Xe3 HPG
morpheus
Invalid Gen Platform: %s
RASTY_DISABLE_IOMMU

 

[ultimatebob]

Device ID and SKU info not defined for ELG_X2 yet, using X3 Device ID

Intel Rasty Renderer - Take the red pill / Xe3 HPG
morpheus
Invalid Gen Platform: %s
RASTY_DISABLE_IOMMU

Funny... I figured that Intel would want us to take the blue pill

 

[Fanatical Meat]

Dammit, when are these things coming out *I know when it’s done or soon*
I want a monitor upgrade and that won’t happen without a new card.

 

[mikk]

MTL_3x4x16= 192
ARL_6x4x16= 384


Adore said Arrow Lake gets 320EUs but from the driver Arrow Lake is up to 384EUs, exactly a doubling from Meteor Lake.

 

[GodisanAtheist]

Do note that neither NVIDIA nor AMD are expected to launch new GPUs before October 2022. Intel’s GPUs will have been out for 10 months at that point.

-Wonder if we'll get a "drop in replacement" 6nm refresh line from AMD meaning AIB's can just drop the newer chip right into existing boards and take advantage of 6nm improvements with very little engineering rework required.

Would be a good hold over for the majority of 2022 for AMD at least, perhaps on their larger N21 dies just from a cost savings + margins perspective.

Old AMD? Probably not, but new AMD is getting refreshingly aggressive on all the fronts so I wouldn't put it past them.

 

[stebler]

By looking at the GenRastyAdapterInfo function of igd_rasty_kmd32.dll in Ghidra, we can see how the platforms are grouped.

gen9
- SKL GT2
- SKL GT3e
- SKL GT4e
- KBL GT2
- CFL GT2

gen11 / gen11-lp
- ICL_1x4x8
- ICL_1x6x8
- ICL_1x8x8

gen11 / gen11-hp
- Rasty no long [sic] supports Gen11-HP

gen12-lp / xe_lp
- TGL_1x6x16
- DG1_1x6x16
- RKL_1x2x16

gen12-hp / xe_hp
- ATS

gen12-hp / xe_hpg
- DG2_2x4x16
- DG2_8x4x16

gen12-hp / xe_lpg
- MTL_1x4x16
- MTL_2x4x16
- MTL_3x4x16

gen12-hp / xe_lpgplus
- ARL_6x4x16

Starting from here the EU configs seem to mean something else.

gen12_9 / xe2_hpg
- ELG_x1_2x4
- ELG_x2_5x4
- ELG_x3_7x4
- ELG_x4_10x4

gen12_9 / xe2_lpg
- LNL_2x4

gen13 / xe3 (_hpg)
- Take the red pill

 

[IntelUser2000]

Why nothing about Alderlake and Raptorlake? Are we supposed to pretend they don't exist? Are they like Intel's version of Vega iGPU where we have that for 2 years?

192EU sounds nice but Meteorlake is on Intel 4 process, and 2 generations ahead of Alderlake, which isn't even here yet in mobile form. Plus it uses that Foveros/tiles/chiplets whatever they want to call it.

Starting from here the EU configs seem to mean something else.

gen12_9 / xe2_hpg
- ELG_x1_2x4
- ELG_x2_5x4
- ELG_x3_7x4
- ELG_x4_10x4

gen12_9 / xe2_lpg
- LNL_2x4

Slices x Cores?

ELG x 4_10x4 = 4 Cores, 10 Slices per die, 4 connected with EMIB or Foveros.

1x 128
2x 320
3x 448
4x 640

The numbers seem reasonable for a Battlemage part that's supposed to be a year from Alchemist. 640 is a small gain from 512, but you double that to be at top, and double it again to be at top with next gen AMD/Nvidia parts.

LNL: 128

 

[mikk]

The numbers seem reasonable for a Battlemage part that's supposed to be a year from Alchemist. 640 is a small gain from 512, but you double that to be at top, and double it again to be at top with next gen AMD/Nvidia parts.

LNL: 128

You have to assume Gen12.9 has 16 EUs each subslice, this is what we don't know yet.

 

[Glo.]

MTL_3x4x16= 192
ARL_6x4x16= 384


Adore said Arrow Lake gets 320EUs but from the driver Arrow Lake is up to 384EUs, exactly a doubling from Meteor Lake.

3072 ALUs in an APU/SOC.

3072 ALUs - RTX 2080 Super/RTX 3070 Ti (actual ALU config).

Would intel get perf/ALU parity with Nvidia Ampere in one generation?

I really, really cannot wait for this product...

 

[jpiniero]

If you were wondering how Halo Infinite runs on Xe...

 

[DrMrLordX]

If you were wondering how Halo Infinite runs on Xe...

Will Arc be using the same driver stack?

 

[uzzi38]

If you were wondering how Halo Infinite runs on Xe...

Confirmed Intel driver moment.

My experience with Far Cry 6 in a nutshell too:

https://media.discordapp.net/attachments/844435792080142357/899397114151460864/IMG_20211017_214224.jpg

Last I checked Deathloop crashes once you load into an area, and various games have occasional texture glitches every now and then as well which take away from the experience.

As things stand Intel drivers are a bit too hit or miss for my liking.

 

[IntelUser2000]

Will Arc be using the same driver stack?

They will be based on it, otherwise it means completely rewriting drivers and negating any progress made in the past decade.

Putting a dGPU out means it'll have to put their driver development into overdrive or risk failures in the space.

 

[NTMBK]

They will be based on it, otherwise it means completely rewriting drivers and negating any progress made in the past decade.

Putting a dGPU out means it'll have to put their driver development into overdrive or risk failures in the space.

You'd have hoped that in advance of the launch, they would have massively increased the resources devoted to the driver team, gone out and talked to the developers on key upcoming games like Halo Infinite, and got on top of these bugs. The fact that they're still this broken so late in the day is troubling.

 

[DrMrLordX]

They will be based on it, otherwise it means completely rewriting drivers and negating any progress made in the past decade.

Putting a dGPU out means it'll have to put their driver development into overdrive or risk failures in the space.

That's what I was afraid of . . .

The fact that they're still this broken so late in the day is troubling.

This x100.

 

[GodisanAtheist]

If there is something that Raja really got right during his time at AMD it was the driver program, which unlike hardware cycles is something I can reasonably see him actually having a direct influence on.

I'd be a little shocked if this is something that he drops the ball on at Intel of all places.

That being said, the above points are certainly not a good omen of things to come. I wonder if Intel has some sort of NDA with devs as to not talk about the sorts of issues they're working on and overcoming.

 

[mikk]

Lack of game game developer support is a problem, because they don't care about Intel just yet. If there is a graphics glitch like this they report to AMD/Nvidia well in advance before the game comes out to the public. Intel can't talk to every game dev, what a rubbish. Intel needs dGPUs in the market widely available and game devs cannot ignore them anymore at some point. There are issues right now but it's not like every second game has glitches like above. One game has a glitch and the internet is like they're still this broken. They may have alpha or beta branches of unreleased drivers in the lab which are currently prepared for the dGPUs in a few months.

 

[NTMBK]

Lack of game game developer support is a problem, because they don't care about Intel just yet. If there is a graphics glitch like this they report to AMD/Nvidia well in advance before the game comes out to the public. Intel can't talk to every game dev, what a rubbish. Intel needs dGPUs in the market widely available and game devs cannot ignore them anymore at some point. There are issues right now but it's not like every second game has glitches like above. One game has a glitch and the internet is like they're still this broken. They may have alpha or beta branches of unreleased drivers in the lab which are currently prepared for the dGPUs in a few months.

It's the biggest expected game launch of the year. Intel should 100% have had developer relations teams working with 343, providing testing assistance to find these issues, and having developers assigned to fix them. I'm not saying that they should (or could) talk to every single developer out there, but you could see from orbit that Halo Infinite was going to be a big deal.

 

[jpiniero]

Same guy has a video of FF7 Remake, and that seems to work okay albeit (below?) 720p. Note that the game forces resolution scaling so who knows what the actual resolution is.

 

[IntelUser2000]

If there is something that Raja really got right during his time at AMD it was the driver program, which unlike hardware cycles is something I can reasonably see him actually having a direct influence on.

We can't expect AMD Raja to be same as Intel Raja. First, there was that HardOCP article about him hating being at AMD and wanting to work at Intel.

Second, Intel is much, much bigger so he'll have a lot less influence.

I don't expect huge changes but if they stick to the space eventually they'll figure it out. The biggest issue with Intel is they drop it at the slightest hint of failure.

Trying now and trying when they actually have a gaming GPU is very different. Right now even with Iris Xe when they have gaming issues I bet most will be like "iGPUs are shit". But if a customer actually goes out and buy a dGPU then they'll no longer have that excuse and will go directly complain to vendors which will go to Intel.

 

[LightningZ71]

This is a LONG standing Intel problem. From their partnership with PowerVR on Atom processors about half a decade ago where the released a grand total of ONE bug fixes driver after the OG one, to their abandonment of the GPU in KadyLake G, to their most recent abandonment of one of their HPC versins of Xe. Intel doesn't have the stomach to put effort into continual driver development on ANYTHING that isn't critical to their cash cow CPUs, which means chipset drivers (which change very little) and the security processor platform critical for their business customers. Making Xe drivers better doesn't make them IMMEDIATE cash.

Frankly, I was borderline flabbergasted when they announced XeSS being platform agnostic. DLSS is likely the only reason for them to consider it, but I doubt seriously that it ever goes anywhere when they likely believe they can just compute their way past NVidia without it.

 

[Tup3x]

You'd have hoped that in advance of the launch, they would have massively increased the resources devoted to the driver team, gone out and talked to the developers on key upcoming games like Halo Infinite, and got on top of these bugs. The fact that they're still this broken so late in the day is troubling.

To be honest, I'm not that worried about that. I'd be worried if these issues are still these with launch drivers. Reviewers would totally flame them if something like Halo Infinite would work properly.

 

[Dayman1225]

Intel Alchemist/DG2 は AV1エンコードをサポートか | Coelacanth's Dream

Intel GPU のメディア関連機能 (動画エンコード/デコード、ポストプロセッシング) をサポートする VA-API (Video Acceleration API) 向けドライバー、Intel(R) Media Driver で Alchemist/DG2 、Ponte Vecchio 、Xe-HP

www-coelacanth--dream-com.translate.goog

Looks like DG2 (supposedly) supports AV1 encode upto 8192x8192 res.