Intel Fixing Its Driver Mess Arc GPUs
video description
Date: 2022-08-22
Related videos
Comments and reviews: 14
Mohamed
100% in agreement although I might say right now just even solidify that Intel is even in 3rd place on merit not just by virtue of being the only other player in town is most important. I also would add that these 3 tier sku's options is a bit of a stretch and maybe should just focus on the low tier and middle tier until they sort out the board partners and software driver situation; the high tier should be out in the market when they have a product that can competitively push them towards the 2nd place ranked GPU supplier. Third thing is they should improve + market the productivity components of these GPU's ie. rendering, encoding/ streaming, AI/ ML, Heteregenous Computing for scientific modelling tools ... the fundamentals of productivity and what sells the HPC themed systems to large render farms and HPC clients.
Finally I fail to see the point of Intel going to TSMC for the Arc product. The die size and efficiency aren't that much better than a 12nm tech node NVIDIA GPU. I think Intel could have stuck with the 10nm+++++++ process and still had an Arc product of equal quality with atleast the crutch to say they were handicapped by the tech node as a fallback and an indicator of how they can catch up and improve in time. I feel like the TSMC fab choice has led to what delayed them the most in their launch and it will set them back another launch cycle as they go back in house eventually. Better to go back to in house fab now I suppose because it's only going to jeopardize the GPU business down the line as the messy transition begins.
reply
100% in agreement although I might say right now just even solidify that Intel is even in 3rd place on merit not just by virtue of being the only other player in town is most important. I also would add that these 3 tier sku's options is a bit of a stretch and maybe should just focus on the low tier and middle tier until they sort out the board partners and software driver situation; the high tier should be out in the market when they have a product that can competitively push them towards the 2nd place ranked GPU supplier. Third thing is they should improve + market the productivity components of these GPU's ie. rendering, encoding/ streaming, AI/ ML, Heteregenous Computing for scientific modelling tools ... the fundamentals of productivity and what sells the HPC themed systems to large render farms and HPC clients.
Finally I fail to see the point of Intel going to TSMC for the Arc product. The die size and efficiency aren't that much better than a 12nm tech node NVIDIA GPU. I think Intel could have stuck with the 10nm+++++++ process and still had an Arc product of equal quality with atleast the crutch to say they were handicapped by the tech node as a fallback and an indicator of how they can catch up and improve in time. I feel like the TSMC fab choice has led to what delayed them the most in their launch and it will set them back another launch cycle as they go back in house eventually. Better to go back to in house fab now I suppose because it's only going to jeopardize the GPU business down the line as the messy transition begins.
reply
Cancer
Honestly, if you want a master class in how to break into a well-developed market, look at how the Japanese auto makers broke into the US in the 70s and 80s. Don't focus on being at the top of the game, focus on delivering a quality, reliable product (not necessarily a feature-packed one) at a good price point, target the mid-tier and budget sectors, and plan for the long term. The two big players are so busy fighting each other to have the best halo model that they're neglecting the price/performance of the budget sector. A 70s Toyota wasn't luxurious or powerful, but it was reliable and economical, and that swayed a lot of buyers.
I don't think Intel needs to worry about being #2 in dGPUs. #3 is perfectly fine, because it isn't a three horse race . They're not just competing against AMD and Nvidia for sales. They're also competing against eBay. And against the customer deciding not to upgrade at all. Just like how used cars make up a significant amount of sales for auto dealers, used GPUs are everywhere right now and the useful lifespan of a card for a gamer is longer than it has been in decades.
If you're convincing someone who might have otherwise decided to buy used or to not buy at all to buy a new GPU, you're still making a sale even if it comes at a worse margin. The key is to have enough of a value proposition that a few bucks more for a new card today will look better than a used card or saving for a top shelf model in six months to a year.
reply
Honestly, if you want a master class in how to break into a well-developed market, look at how the Japanese auto makers broke into the US in the 70s and 80s. Don't focus on being at the top of the game, focus on delivering a quality, reliable product (not necessarily a feature-packed one) at a good price point, target the mid-tier and budget sectors, and plan for the long term. The two big players are so busy fighting each other to have the best halo model that they're neglecting the price/performance of the budget sector. A 70s Toyota wasn't luxurious or powerful, but it was reliable and economical, and that swayed a lot of buyers.
I don't think Intel needs to worry about being #2 in dGPUs. #3 is perfectly fine, because it isn't a three horse race . They're not just competing against AMD and Nvidia for sales. They're also competing against eBay. And against the customer deciding not to upgrade at all. Just like how used cars make up a significant amount of sales for auto dealers, used GPUs are everywhere right now and the useful lifespan of a card for a gamer is longer than it has been in decades.
If you're convincing someone who might have otherwise decided to buy used or to not buy at all to buy a new GPU, you're still making a sale even if it comes at a worse margin. The key is to have enough of a value proposition that a few bucks more for a new card today will look better than a used card or saving for a top shelf model in six months to a year.
reply
Chris
It's good to see Intel respond, but it's bitter-sweet for any early adopters on Xe-powered laptops and handheld gaming devices that have been reporting issues for well over a year now and are usually met with this may be fixed in 3-6 months . That's only after a long back-and-forth that makes it very evident that Intel have massively under-invested in QA and testing. Issues like Game crashes on startup or Massive visual corruption upon entering game should not require significant effort on the part of the reporter before Intel looks into the issue. Similarly, many driver revisions have regressed previously-working features (either performance issues or outright crashing) and that seems to indicate that they have wholly inadequate regression testing and automated testing.
One might be more forgiving if there wasn't such a stark contrast with the reliability and performance of the equivalent Linux graphics driver. Often their response is this is an issue in the game and so we can't do anything , which I sympathise with, but that ignores the reality that all competing vendor drivers ship with a huge database of application-specific work-arounds. That's part of modern graphics driver development. I really hope they continue to improve, and at a much faster rate than they are now. I'm glad GN are shining a light on them in such a public way.
reply
It's good to see Intel respond, but it's bitter-sweet for any early adopters on Xe-powered laptops and handheld gaming devices that have been reporting issues for well over a year now and are usually met with this may be fixed in 3-6 months . That's only after a long back-and-forth that makes it very evident that Intel have massively under-invested in QA and testing. Issues like Game crashes on startup or Massive visual corruption upon entering game should not require significant effort on the part of the reporter before Intel looks into the issue. Similarly, many driver revisions have regressed previously-working features (either performance issues or outright crashing) and that seems to indicate that they have wholly inadequate regression testing and automated testing.
One might be more forgiving if there wasn't such a stark contrast with the reliability and performance of the equivalent Linux graphics driver. Often their response is this is an issue in the game and so we can't do anything , which I sympathise with, but that ignores the reality that all competing vendor drivers ship with a huge database of application-specific work-arounds. That's part of modern graphics driver development. I really hope they continue to improve, and at a much faster rate than they are now. I'm glad GN are shining a light on them in such a public way.
reply
Entenkommando
You should really take a look at the most recent video from Moore's law is dead. Because they're absolutely doing what you are afraid of.
Like how they claim Arc was developed specifically with Rebar in mind even though the cards were developed way before AMD announced the feature and also way before Intel started rolling out the feature on their own platform.
18:06 this quote can be translated as we are incapable of optimizing our game drivers in a timely manner. And in DX12 games it's not our job which is why we can claim better performance there, even though we aren't actively doing anything for it.
Meanwhile what they're claiming is that Arc works better on DX12 because it's a more modern architecture.
Or how they went out and proclaimed to urgently work on driver development for DX9 and DX11...3 weeks before cutting native DX9 support.
And they're absolutely not downplaying performance, but watch the video yourself if you want a more in-depth explanation.
I don't have anything against these guys personally and I'm sure they were a blast to have around. But they are PR agents first and foremost. Not an engineer talking from the inside. And they're doing what PR people do. Talk talk talk.
reply
You should really take a look at the most recent video from Moore's law is dead. Because they're absolutely doing what you are afraid of.
Like how they claim Arc was developed specifically with Rebar in mind even though the cards were developed way before AMD announced the feature and also way before Intel started rolling out the feature on their own platform.
18:06 this quote can be translated as we are incapable of optimizing our game drivers in a timely manner. And in DX12 games it's not our job which is why we can claim better performance there, even though we aren't actively doing anything for it.
Meanwhile what they're claiming is that Arc works better on DX12 because it's a more modern architecture.
Or how they went out and proclaimed to urgently work on driver development for DX9 and DX11...3 weeks before cutting native DX9 support.
And they're absolutely not downplaying performance, but watch the video yourself if you want a more in-depth explanation.
I don't have anything against these guys personally and I'm sure they were a blast to have around. But they are PR agents first and foremost. Not an engineer talking from the inside. And they're doing what PR people do. Talk talk talk.
reply
Danil
Media person that do communication thru their video only and pretend they doing something - is just disrespect to people who actually reporting bugs and not scream on their socials about every step they do .
AMD did alot of job connecting with developers and fixing their bugs beginning from 2015(when they start driver rework from scratch). And yes this is better to ignore those loud media people who just scream, and keep working with people who actually do contribution to bug reporting and fixing (since AMD drivers is opensource).
Same with developers from Intel - they respond in Mesa driver git, they do almost instant communication with developers who actually need fix for very critical bugs that make their(developers) software not work on Intel hardware.
As someone who reported hundreds of bugs to Nvidia and AMD (half of them were there for years and were very critical, most of them very low level) for last 3 years - it just sad to see video where another media person scream and think they do something ...
reply
Media person that do communication thru their video only and pretend they doing something - is just disrespect to people who actually reporting bugs and not scream on their socials about every step they do .
AMD did alot of job connecting with developers and fixing their bugs beginning from 2015(when they start driver rework from scratch). And yes this is better to ignore those loud media people who just scream, and keep working with people who actually do contribution to bug reporting and fixing (since AMD drivers is opensource).
Same with developers from Intel - they respond in Mesa driver git, they do almost instant communication with developers who actually need fix for very critical bugs that make their(developers) software not work on Intel hardware.
As someone who reported hundreds of bugs to Nvidia and AMD (half of them were there for years and were very critical, most of them very low level) for last 3 years - it just sad to see video where another media person scream and think they do something ...
reply
J.J.
Charitably, I wonder if the following additional sentences were missing relative to a reality where Intel had thought just a little more with the marketing / PR filter on: Some of those [43] issues [that we filed with our engineering team, after watching the GN video] were already on their radar in full or in part. But indeed, a good many were not. And regardless, we are grateful that the community itself is providing such interested feedback. Or something to that effect.
Only bring it up because if true, it probably does smart a bit to get scolded and feel a bit like, damned if we do, damned if we don't. Ideally they are mature enough to understand that the scope of it is still very much optically in their favor and not be overly sensitive, and let time to the talking. But I'm sympathetic to the fact that in my experience, even otherwise pretty chill, mature adults don't do this very well if they haven't really worked on it on purpose.
reply
Charitably, I wonder if the following additional sentences were missing relative to a reality where Intel had thought just a little more with the marketing / PR filter on: Some of those [43] issues [that we filed with our engineering team, after watching the GN video] were already on their radar in full or in part. But indeed, a good many were not. And regardless, we are grateful that the community itself is providing such interested feedback. Or something to that effect.
Only bring it up because if true, it probably does smart a bit to get scolded and feel a bit like, damned if we do, damned if we don't. Ideally they are mature enough to understand that the scope of it is still very much optically in their favor and not be overly sensitive, and let time to the talking. But I'm sympathetic to the fact that in my experience, even otherwise pretty chill, mature adults don't do this very well if they haven't really worked on it on purpose.
reply
Lietu
Yeah it's definitely sad that a company the size of Intel does not understand the concept of vertical slices - just make the base package first, make it perfect, then add new features. Instead they've tried to bundle together 500 features and none of them work. It's like they've forgotten how version control works at Intel.
They can develop these new features in branches that are not going to be released until that branch is actually ready for release. Well, they could, if they knew what they're doing. They could also just use an Enable experimental features -setting or similar, to make it clear to users which things are still not quite stable.
Only one thing is clear right now: Intel really is not trying particularly hard with Arc. The hardware is barely available, and when it is, it's not great, and then the software supporting the hardware is atrocious.
reply
Yeah it's definitely sad that a company the size of Intel does not understand the concept of vertical slices - just make the base package first, make it perfect, then add new features. Instead they've tried to bundle together 500 features and none of them work. It's like they've forgotten how version control works at Intel.
They can develop these new features in branches that are not going to be released until that branch is actually ready for release. Well, they could, if they knew what they're doing. They could also just use an Enable experimental features -setting or similar, to make it clear to users which things are still not quite stable.
Only one thing is clear right now: Intel really is not trying particularly hard with Arc. The hardware is barely available, and when it is, it's not great, and then the software supporting the hardware is atrocious.
reply
Maya
Downloading components like (DX, MSVC, etc.) runtimes rather than bundling them does make a massive difference in the installer size, but depending on what component it is, it can get ugly.
For my NymphCast project I opted to download the MSVC runtime during installation from the Microsoft server rather than bundling it, because it'd double the installer size on Windows. The same is true for optional components, like wallpapers, that are downloaded as needed. None of that is really problematic, and it works quite well when testing it across a range of systems.
I'd be interested in learning exactly what mystery components they're downloading here. It has to be something that has a rather finicky setup for it to have such a destructive effect. But I'm sure it worked fine on the single developer's PC before it got pushed to production :)
reply
Downloading components like (DX, MSVC, etc.) runtimes rather than bundling them does make a massive difference in the installer size, but depending on what component it is, it can get ugly.
For my NymphCast project I opted to download the MSVC runtime during installation from the Microsoft server rather than bundling it, because it'd double the installer size on Windows. The same is true for optional components, like wallpapers, that are downloaded as needed. None of that is really problematic, and it works quite well when testing it across a range of systems.
I'd be interested in learning exactly what mystery components they're downloading here. It has to be something that has a rather finicky setup for it to have such a destructive effect. But I'm sure it worked fine on the single developer's PC before it got pushed to production :)
reply
Josua
I could foresee that enabling rebar from the factory could come with some small issue. This probably won't affect you all that did a fresh install of windows only recently, but may affect those who want to bring their old 5+ years old OS drives into their new system. Back then, installing your windows with MBR is fine. But if rebar is on, the system simply refuse those old MBR drives to act as the boot drive. In order to get rebar working, you need to convert that drive into GPT first. Although the process is very simple once you find out what the issue is, finding the issue itself could be hard since the BIOS don't tell you about anything. So i think that if manufacturers are going to force rebar on by default, there should be some mention to this whole MBR and GPT issue in their manual somewhere, at the bare minimum.
reply
I could foresee that enabling rebar from the factory could come with some small issue. This probably won't affect you all that did a fresh install of windows only recently, but may affect those who want to bring their old 5+ years old OS drives into their new system. Back then, installing your windows with MBR is fine. But if rebar is on, the system simply refuse those old MBR drives to act as the boot drive. In order to get rebar working, you need to convert that drive into GPT first. Although the process is very simple once you find out what the issue is, finding the issue itself could be hard since the BIOS don't tell you about anything. So i think that if manufacturers are going to force rebar on by default, there should be some mention to this whole MBR and GPT issue in their manual somewhere, at the bare minimum.
reply
Nicolas
It was important to release the Intel GPU before AMD and Nvidia release their new GPUs and they couldn t be to close. So I think this was a pure management decision, which is even understandable. You need time to prepare the release. You can even work on the drivers while everything else is done, including shipping etc. Because of the online driver download, you buy 1-2 months time, where a lot can be done. I guess the management just thought that the bugs would be minor after that time, perhaps even in exchange with the developers.
That deadline was critical and it didn t work out, which can happen, but shouldn t and can seriously damage a new product. So I wouldn t blame QA etc on Intels side and I hope GN don t judge them to hard for that, as long as they fix the problems quickly
reply
It was important to release the Intel GPU before AMD and Nvidia release their new GPUs and they couldn t be to close. So I think this was a pure management decision, which is even understandable. You need time to prepare the release. You can even work on the drivers while everything else is done, including shipping etc. Because of the online driver download, you buy 1-2 months time, where a lot can be done. I guess the management just thought that the bugs would be minor after that time, perhaps even in exchange with the developers.
That deadline was critical and it didn t work out, which can happen, but shouldn t and can seriously damage a new product. So I wouldn t blame QA etc on Intels side and I hope GN don t judge them to hard for that, as long as they fix the problems quickly
reply
gamersnexus
After having a very bad experience with a 5700XT (and other ATI/AMD cards in the past, beginning with the Rage series and seeing what problems Intel's ARC has, i don't think that nvidia's monopoly will ever change. Nvidia has no competition, when it comes to software stability and support.
Like many other people, i don't like how nvidia prices their products, but i don't see myself buying anything else, because i want a good experience and i also need a reliable GPU for work.
Except the upcoming 4000 space heater series, which i will skip. I live in central Europe, where electricity is expensive and summers are hot. I don't see myself wasting 600W of electric heating in my house, even if gas and oil prices are high, because i already have a woodgas heating system.
reply
After having a very bad experience with a 5700XT (and other ATI/AMD cards in the past, beginning with the Rage series and seeing what problems Intel's ARC has, i don't think that nvidia's monopoly will ever change. Nvidia has no competition, when it comes to software stability and support.
Like many other people, i don't like how nvidia prices their products, but i don't see myself buying anything else, because i want a good experience and i also need a reliable GPU for work.
Except the upcoming 4000 space heater series, which i will skip. I live in central Europe, where electricity is expensive and summers are hot. I don't see myself wasting 600W of electric heating in my house, even if gas and oil prices are high, because i already have a woodgas heating system.
reply
Brianna
I think you're spot on with this commentary. The way I'm looking at these early until devices is for use as AV1 encoders and/or for lower end gaming rigs where you might not need the horsepower of a higher end card, but you also don't want to spend a lot. For example, a kid's rig where they're mostly running Roblox and Minecraft, maybe the occasional Stray or Sims. These are things where you need more than an IGP generally, but you really don't need a lot.
In order for me to consider that, I need stable drivers. I don't really care about the fancy features like VRR, ray tracing, etc. I care about the drivers working properly and being stable overall. If Intel can get that working properly, I will be happy to consider them as a reasonable player in the budget market.
reply
I think you're spot on with this commentary. The way I'm looking at these early until devices is for use as AV1 encoders and/or for lower end gaming rigs where you might not need the horsepower of a higher end card, but you also don't want to spend a lot. For example, a kid's rig where they're mostly running Roblox and Minecraft, maybe the occasional Stray or Sims. These are things where you need more than an IGP generally, but you really don't need a lot.
In order for me to consider that, I need stable drivers. I don't really care about the fancy features like VRR, ray tracing, etc. I care about the drivers working properly and being stable overall. If Intel can get that working properly, I will be happy to consider them as a reasonable player in the budget market.
reply
Rainman
One day, marketing departments will wake up and realise that rather than working on spin and feeding us BS, that actually they need to engage with engineering teams internally and help drive QA, improve reliability and performance of their products. One day they will might just understand that actually a reputation for quality and performance actually shifts units and drives sales, long-term. The fact that this product is being thrust into the market in it's current condition tends to support the view that this launch is leaning on the fact that just a few months ago the market would have bought almost anything at any price. Intel are too late to that game.
reply
One day, marketing departments will wake up and realise that rather than working on spin and feeding us BS, that actually they need to engage with engineering teams internally and help drive QA, improve reliability and performance of their products. One day they will might just understand that actually a reputation for quality and performance actually shifts units and drives sales, long-term. The fact that this product is being thrust into the market in it's current condition tends to support the view that this launch is leaning on the fact that just a few months ago the market would have bought almost anything at any price. Intel are too late to that game.
reply
KVahlman
One big thing to skew the commentary is also the fact that the only barely(?) available and investigated product in the discreet Arc line is from the very bottom of the lineup, which obviously will make people look at the absolute performance and go this is poop and totally miss the interesting architecture points. In HUBs recent comparison, the A380 lost very little performance relatively when resolution or settings changed which is super interesting from architecture pov. But since the absolute values were under 30fps, people are uninterested in anything but it being at the bottom of the charts (where it of course belongs too).
reply
One big thing to skew the commentary is also the fact that the only barely(?) available and investigated product in the discreet Arc line is from the very bottom of the lineup, which obviously will make people look at the absolute performance and go this is poop and totally miss the interesting architecture points. In HUBs recent comparison, the A380 lost very little performance relatively when resolution or settings changed which is super interesting from architecture pov. But since the absolute values were under 30fps, people are uninterested in anything but it being at the bottom of the charts (where it of course belongs too).
reply
Add a review, comment
Other channel videos
