News Intel GPUs - Intel launches A580

[psolord]

Man what a job, writing drivers....


....boss walks in, catches you playing games.

What are you doing?

Just testing the new drivers boss. There..see that crooked pixel...(points his finger at a 4k screen)... I NEED TO FIX THAT!

 

[Grabo]

I find the passage a bit confusing. Intel have been saying Q2 lately and I haven't seen a date. From the Igor's article:

"Several different sources have told me of another Intel card postponement, at least as far as SKU 1 to 3, the three performance models, are concerned. There are currently consistent rumors that the first models will be released between May 2, 2022 and June 1, 2022. Furthermore, the tape-out of the QS (qualified sample) should not take place for 1-2 weeks."

Has he heard from his sources a postponement from May 2, 2022 and June 1, 2022 which the current rumours are saying, to an unknown later date
or
has he heard from his sources a rumour that there's been a release postponement from an unknown date to somewhere between May 2,2022 and June 1,2022?

if the latter then still Q2 and I never expected anything before June. (Well before December but we'll see )

 

[IntelUser2000]

igorslab has a review of Gunner Iris Xe Max DG1. Seems to do fairly well. RX 550 class. Uses 20-30W depending on the game and the author figures the memory and the power losses to be about 10W so maybe 10-20W for the GPU.

He says the motherboard UEFI firmware has to be loaded into the NOR flash chip on the board before it works which means it doesn't have any GPU BIOS like a normal dGPU. So it's as we expected - Iris Xe integrated on a dGPU format.

 

[Borealis7]

Man what a job, writing drivers....


....boss walks in, catches you playing games.

What are you doing?

Just testing the new drivers boss. There..see that crooked pixel...(points his finger at a 4k screen)... I NEED TO FIX THAT!

now imagine a QA position at a game studio

and now imagine a Machine-Learning trainer/classification job, at a company which develops AI that recognizes pr0n and filters it out

 

[NTMBK]

now imagine a QA position at a game studio

Working long hours, playing the same game over and over, trying methodically to break it instead of actually enjoy it... it's a good way to make yourself hate a game.

 

[Borealis7]

Working long hours, playing the same game over and over, trying methodically to break it instead of actually enjoy it... it's a good way to make yourself hate a game.

some people log thousands of hours on the same game (counted by steam for example), i don't think they hate the game they're playing.

 

[DrMrLordX]

Working long hours, playing the same game over and over, trying methodically to break it instead of actually enjoy it... it's a good way to make yourself hate a game.

Shoot, I've done that before for free (closed beta, alpha, and pre-alpha testing games).

 

[Shivansps]

now imagine a QA position at a game studio

Ive done that a few years ago, in fact i got my name in a few game credits as QA. Its a terrible job, trust me. Most of the time you are fighting a game that is in pre-alpha state, broken as hell, that crashes as hell, while you are trying to do X thing that is assigned to you over and over.

Its a good way to raise your blood pressure and die if you ask me.

 

[Shivansps]

igorslab has a review of Gunner Iris Xe Max DG1. Seems to do fairly well. RX 550 class. Uses 20-30W depending on the game and the author figures the memory and the power losses to be about 10W so maybe 10-20W for the GPU.

He says the motherboard UEFI firmware has to be loaded into the NOR flash chip on the board before it works which means it doesn't have any GPU BIOS like a normal dGPU. So it's as we expected - Iris Xe integrated on a dGPU format.

I just saw it, it is considerably faster than the DG1s i tested, i guess this one is not limited to 25W TDP like the asus one, and plus it has all 96UEs avalible.

EDIT: Says there it is 25W PL1 and 41W PL2... unless PL2 is infinite that petty much pointless. Thats what i saw with the Asus DG1.

BTW, i dont longer have any DG1, they sold out pretty fast, much to my surprise.

 

[igor_kavinski]

Ive done that a few years ago, in fact i got my name in a few game credits as QA. Its a terrible job, trust me. Most of the time you are fighting a game that is in pre-alpha state, broken as hell, that crashes as hell, while you are trying to do X thing that is assigned to you over and over.

Its a good way to raise your blood pressure and die if you ask me.

Aren't the developers supposed to test if what they wrote is working as it should before handing it off to testers to check for other related issues? That would waste a lot of time if they change one line of code then ask the tester to test then again change and so on. Both would need to be sitting together or side by side for this process to work efficiently.

 

[igor_kavinski]

BTW, i dont longer have any DG1, they sold out pretty fast, much to my surprise.

WTH??? Who bought them and for what purpose? I didn't know there was market for complete useless crap. Must have been someone who had no idea what they were getting and just trusted the Intel name.

 

[Shivansps]

Aren't the developers supposed to test if what they wrote is working as it should before handing it off to testers to check for other related issues? That would waste a lot of time if they change one line of code then ask the tester to test then again change and so on. Both would need to be sitting together or side by side for this process to work efficiently.

Devs always have their own in-site testers, so the builds that gets to QAs is already pre-tested, but at some stages of development there are issues that you have to deal with, all my QA jobs were remote, at first some for free just to gain experience, then some paid ones. But i stopped doing those years ago.
For example, if im asssigned to test missions/quests in X area of the game, and you need to do it because the team that does that job is not the same team that does, graphics for example. You are forced to deal with the game stability and any other bug/balance issue that is not related to your task, but, it gets in the way. Things DO get better as the development goes on, but so does the pressure on the devs.

Except for one case, i never played any of the games i tested, i was sick of it at the end.

WTH??? Who bought them and for what purpose? I didn't know there was market for complete useless crap. Must have been someone who had no idea what they were getting and just trusted the Intel name.

All pre-build pcs, we were not allowed to sell the gpus alone. They key was price, they were really cheap GT730 cheap, so it was a really cheap combo with the 10100F and 11400F.
Im petty sure Asus just wanted to get rid of them and that the real price should be more than that.

 

[Insert_Nickname]

All pre-build pcs, we were not allowed to sell the gpus alone. They key was price, they were really cheap GT730 cheap, so it was a really cheap combo with the 10100F and 11400F.
Im petty sure Asus just wanted to get rid of them and that the real price should be more than that.

Undoubtedly. Especially since they'll only ever work in LGA-1200 boards with specific chipsets. Alternative was likely a landfill.

That said, that kind of combo does make a lot of sense.

 

[kschendel]

Aren't the developers supposed to test if what they wrote is working as it should before handing it off to testers to check for other related issues?

It's a balance. I'm a dev (not gaming) and I'm expected to deliver code that compiles and works when you take the obvious paths through execution. That might be basic exercising of a feature, or showing that a bug testcase is fixed. There's also a canned test suite to run.

The problem is that software isn't linear. Particularly in a game, there are a huge number of paths through the code, as well as multi-thread timing issues. You might find a bug that only occurs if you visit area X, then Y, then attack Z with weapon W shooting at just the right time. Canned suites won't find these, you need a human to sit and mess with it looking for breakage. Devs can't do that because then they aren't devs any more, they are QA testers.

 

[IntelUser2000]

Igorslab review also says you can buy them at Taobao for $110 USD and the prices are "always increasing".

Guess Intel GPUs are subject to the GPU market pricing too.

@Shivansps Sounds to me like your DG1 was suffering from immature firmware issues? Good that this DG1 is not in a worse condition game compatibility wise than integrated Xe.

 

[Shivansps]

Igorslab review also says you can buy them at Taobao for $110 USD and the prices are "always increasing".

Guess Intel GPUs are subject to the GPU market pricing too.

@Shivansps Sounds to me like your DG1 was suffering from immature firmware issues? Good that this DG1 is not in a worse condition game compatibility wise than integrated Xe.

Im not sure from were the firmware comes from anymore. They told me that it was needed to use the bios they provided, then i started to realise it was not needed. For example the B460M-A R.20 0206 bios (the first bios, dated as 2021/04/12) did booted up the DG1. I did noticed issues with that specific board unless i was using the bios they provided. So it is not very clear how the firmware is distributed. I never had a chance to test with other vendors, they didnt allowed that anyway.
My best guess is that for Asus, DG1 support comes with 11th gen support.

So this company GUNNIR comes up and start selling these DG1s and im really not sure how that is supposed to work. Im guessing the firmware that Asus have on their boards is for the 80EU version only.

 

[IntelUser2000]

@Shivansps Still trying to wonder why there's a big difference between the DG1 you got and this one. Actually they also tested the other DG1 and it had horrible compatibility. And I don't mean horrible as some might say which is one out of 10 games don't work properly. No, they could barely get any game to work.

Mature firmware could be a culprit. Not sure why drivers can make that much of a difference since it's essentially the same chip as the integrated Iris Xe. At 95mm2 die it sounds like it's Tigerlake with all the CPU cores stripped out.

 

[Shivansps]

@Shivansps Still trying to wonder why there's a big difference between the DG1 you got and this one. Actually they also tested the other DG1 and it had horrible compatibility. And I don't mean horrible as some might say which is one out of 10 games don't work properly. No, they could barely get any game to work.

Mature firmware could be a culprit. Not sure why drivers can make that much of a difference since it's essentially the same chip as the integrated Iris Xe. At 95mm2 die it sounds like it's Tigerlake with all the CPU cores stripped out.

Did they say exactly what driver version they used? Because Intel website is a f* mess with that. Depending on how you search you may find at least 3 diferent drivers that looks to be the newerest one.

example:
Intel® Iris® Xe Graphics and Intel® UHD Driver for 12th Gen Intel® Core™ Processors

Intel | Data Center Solutions, IoT, and PC Innovation

Intel's innovation in cloud computing, data center, Internet of Things, and PC solutions is powering the smart and connected digital world we live in.

www.intel.com

1/31/2022

Intel® Graphics – Windows* DCH Drivers <-This is the one to use

Intel® Graphics – Windows* DCH Drivers

This download installs Intel® Graphics Driver 31.0.101.3790/31.0.101.2114 for Xe Dedicated, 6th-13th Gen Intel® Core™ Processor Graphics, and related Intel Atom®, Pentium®, and Celeron® processors. Driver version varies depending on the Intel Graphics in the system.

www.intel.com

3/8/2022

Intel® Graphics - BETA Windows* DCH Drivers

Intel® Graphics - BETA Windows* DCH Drivers

This download installs the Intel® Graphics Beta Driver for Xe Dedicated, 6th-12th Gen Intel® Core™ Processor Graphics, and related Intel Atom®, Pentium®, and Celeron® processors.

www.intel.com

11/15/2021

Asus even have a older one

ASUS DG1-4G - Support

www.asus.com

 

[Glo.]

Intel might be preparing three DG2 GPUs after all, SOC3 with 256EUs appears - VideoCardz.com

Intel DG2 GPU with 256 Execution Cores? Intel might have just confirmed that there is a third variant of the DG2 GPU, codenamed SOC3. In the recent patches to Intel’s Graphics System Controller Firmware Update Library (IGSCU FU) for Linux, there is an entry with a third DG2 GPU, supposedly...

videocardz.com

Possible SOC with 256 EUs and 128 bit bus(?).

Could it be similar scenario to 128 EU? In laptops 128 EU has 64 bit bus, on desktop 96 bit. Is it possible that in laptops 256 EU has 128 bit bus and on desktop - 192 bit?

 

[mikk]

@Shivansps Still trying to wonder why there's a big difference between the DG1 you got and this one. Actually they also tested the other DG1 and it had horrible compatibility. And I don't mean horrible as some might say which is one out of 10 games don't work properly. No, they could barely get any game to work.

The other DG1 was an engineering sample which was broken.

Did they say exactly what driver version they used? Because Intel website is a f* mess with that. Depending on how you search you may find at least 3 diferent drivers that looks to be the newerest one.

They did use v1340, you can see this in the GPUz screenshot. It's not really a mess btw, the higher the build number the newer it is. For example the beta driver is old, it's obsolete. You can see this on the build number the last 4 digits.

 

[IntelUser2000]

Possible SOC with 256 EUs and 128 bit bus(?).

Could it be similar scenario to 128 EU? In laptops 128 EU has 64 bit bus, on desktop 96 bit. Is it possible that in laptops 256 EU has 128 bit bus and on desktop - 192 bit?

Intel is very competent making numerous dies to maximize yield and profit margins. Two is a surprise. They've been focusing on modularity with Gen graphics for over a decade now.

If they wanted to, they could have 512, 384, 256, 192, 128 die versions and cut them down further.

The other DG1 was an engineering sample which was broken.

I'm not just talking about that. The Gunner Iris is the first functional DG1. All others have significant issues, including the one @Shivansps was playing with.

Intel needs WAY more work to be on par with AMD/Nvidia. They are kinda dysfunctional in that department. Software, hardware, developer relations, all have to be a huge notch up from where they are at right now.

I wouldn't give them a flying chance on dGPUs if they weren't ramping up their efforts slowly using their massive iGPU marketshare. For 15 years now. They aren't entering a completely new market, as if they were a company out of nowhere.

 

[mikk]

Intel needs WAY more work to be on par with AMD/Nvidia. They are kinda dysfunctional in that department. Software, hardware, developer relations, all have to be a huge notch up from where they are at right now.

On the hardware side I'm confident they are competitive with Xe HPG against RDNA2 and Ampere. Intel claimed 1.5x higher clock speeds or 1.5 improved perf/w for Xe HPG on TSMC 6nm.

Biggest issue with Xe LP right now is their low clock speed+perf/w and possibly bandwidth efficiency as well because with LPDDR4-4266 or DDR5-4800 Xe LP seems to perform better in comparison to Vega 8.

Clock speed issue RDNA2 vs Xe LP

Same amount of shader units: constant 2200Mhz against max 1350 Mhz (no GPU clock speed there but this SKU can't boost higher than 1350 Mhz)
1.5x clock speed and they boost it to 2000+ Mhz. Xe HPG should also have a better IPC, although we have to wait by how much. Frontend/backend on HPG needs to feed only 64 EUs instead of 96EUs, 14% more threads per EU, L1 double the size etc. The architecture should be fine.

 

[Glo.]

Intel is very competent making numerous dies to maximize yield and profit margins. Two is a surprise. They've been focusing on modularity with Gen graphics for over a decade now.

If they wanted to, they could have 512, 384, 256, 192, 128 die versions and cut them down further.

The more I dig into this, the more I looked into Locuza's schematics of the die layouts, the more I did armchair calculations, the more I saw possibility that indeed 256 EU die would have 192 bit bus.

 

[Shivansps]

They did use v1340, you can see this in the GPUz screenshot. It's not really a mess btw, the higher the build number the newer it is. For example the beta driver is old, it's obsolete. You can see this on the build number the last 4 digits.

I know, and it is more clear now because only one of them seems to be getting updates, but back in dec/jan the version and the dates were all too close to each other.

 

[aigomorla]

Man what a job, writing drivers....


....boss walks in, catches you playing games.

What are you doing?

Just testing the new drivers boss. There..see that crooked pixel...(points his finger at a 4k screen)... I NEED TO FIX THAT!

Nvidia response:

Break last years (insert game X) optimizations since no one plays last years titles to fix it, and let forum staff know of a possible fallout when steam goes on sale on that title.