News Intel GPUs - Battlemage officially announced, evidently not cancelled

[NTMBK]

It's also really hard to hire good quality software engineers right now- especially in a niche field like GPU drivers, when most people with that experience already work at Nvidia or AMD. The team has probably been chronically understaffed.

 

[igor_kavinski]

@HurleyBird and @Stuka87 Your mission, should you choose to accept it, is to infilitrate the GPU divison of Intel, stay there as long as you can bear and report from the trenches on what the hell exactly is going on there? We salute you for your service!

 

[HurleyBird]

I assume that the schedule was extremely harsh and QA and devs were raising all of the alarms, but management just told the to ship whatever even if it's alpha or beta quality at best.

That may be the case for some problems -- display output compatibility comes to mind, but is emphatically not the case for the UI issues. That sort of bug takes mere minutes to fix if you're lucky, and maybe a couple of hours if you are not. I can't remember anything that took me more than a couple hours at least. It's inconceivable that such issues were known yet not resolved.

As an aside, I wonder if driver development and QA is being done outside of North America. At first blush, the voltage slider bug looks particularly egregious, but it would make a lot of sense if it were a localization issue that did not manifest itself in the team's locale. If that's the case, the fault lies much more on the shoulders of QA than dev, since localization problems are pretty much inevitable in a developer's career and any competent QA team should test localization. For context, see the Turkey test.

 

[Dribble]

I can't believe that I'm saying this - but I don't think that it's correct blaming QA or the devs. I would have to assume that management pushed the dev and QA teams to release a full list of features no matter what, and that's what they did. I assume that the schedule was extremely harsh and QA and devs were raising all of the alarms, but management just told the to ship whatever even if it's alpha or beta quality at best.

There's no way the devs/QA don't know the state of what they were releasing - it has to be the management that's telling them to ship whatever no matter what.

I would suspect you need to go further back to the hardware - I bet it's not right and makes it impossibly complex to develop efficient drivers for. It may well be that it's going to take an iteration or two of hardware development until they have something that you can really develop a good driver for. It may also be that management have been told that this gen of hardware is broken, but they can't really bin the whole lot due to pressure from even higher management so they release something as quietly as possible in china in an attempt to keep the higher ups and the share holders happy.

 

[Heartbreaker]

I would suspect you need to go further back to the hardware - I bet it's not right and makes it impossibly complex to develop efficient drivers for. It may well be that it's going to take an iteration or two of hardware development until they have something that you can really develop a good driver for. It may also be that management have been told that this gen of hardware is broken, but they can't really bin the whole lot due to pressure from even higher management so they release something as quietly as possible in china in an attempt to keep the higher ups and the share holders happy.

It's not really about efficiency at this point. It's a buggy mess, even in the UI. The hardware really isn't what's driving UI bugs.

 

[Dribble]

It's not really about efficiency at this point. It's a buggy mess, even in the UI. The hardware really isn't what's driving UI bugs.

If the hardware is tough to code drivers for then it will be take longer to write the drivers and make it tricker to make the card reliably fast. Being as they had to release the card, and it has to have good performance (as that more then anything is what reviews look at) then reliability is going to suffer.

 

[Heartbreaker]

If the hardware is tough to code drivers for then it will be take longer to write the drivers and make it tricker to make the card reliably fast. Being as they had to release the card, and it has to have good performance (as that more then anything is what reviews look at) then reliability is going to suffer.

This has nothing to do with UI bugs. I specifically pointed out UI bugs...

 

[igor_kavinski]

This has nothing to do with UI bugs. I specifically pointed out UI bugs...

Give the lone poor guy at Intel a break who is doing the UI for ALL their software products. Priorities. ARC comes last.

 

[Stuka87]

@HurleyBird and @Stuka87 Your mission, should you choose to accept it, is to infilitrate the GPU divison of Intel, stay there as long as you can bear and report from the trenches on what the hell exactly is going on there? We salute you for your service!

I would actually love to go in there and replace whoever is currently managing it. I enjoy challenges like that.

I am really quite grateful that here at my current place of employment, I have been given the authority as to when something can or cannot be sent to production. Sales/Marketing can say how much they want something all they want. But if the features arent ready, they arent going out. Customers care WAY more about a working product than they do a new feature.

I get the feeling that in this case, its a failure of the product manager. Its up to them to say what features to have ready for deployment first. And instead of selecting the most important things only, they basically spread their recourse out to have everything done. So instead of a base feature set that worked well, they now have a full feature set that is entirely unusable.

 

[igor_kavinski]

As an aside, I wonder if driver development and QA is being done outside of North America.

https://twitter.com/x/status/1506620198589112320

Don't know about drivers but GPU post-silicon validation team is in Bangalore.

https://twitter.com/x/status/1399966271182045184

If we suppose that final silicon was handed over to Lisa on the date of this tweet, it would be a miracle for her team to deliver something as robust as the AMD/Nvidia drivers who've been at it for years.

 

[igor_kavinski]

So instead of a base feature set that worked well, they now have a full feature set that is entirely unusable.

Yeah everything in there but not polished. Giving the team mandates that "everything MUST be ready" which leads to everything half-baked due to time constraints.

 

[KompuKare]

If we suppose that final silicon was handed over to Lisa on the date of this tweet, it would be a miracle for her team to deliver something as robust as the AMD/Nvidia drivers who've been at it for years.

And imagine that some promised feature still didn't work with that final tape-out and the driver team had spend ages writing for that feature while it was talked about and had poured resources into trying to simulate that feature(s). While a moving target is sort of to be expected, we have no how idea at what rate the target was moving and whether it could take evasive actions by also become a backward moving target!
(Metaphor almost stretched to breaking!)

 

[Leeea]

Lot of blame being sent Lisa's way in some posts here.

But with what looks like the timelines involved, they gave Lisa mission impossible. She never had even a ghost of a chance.

 

[igor_kavinski]

If they had developed the GPU in secrecy and launched A380 today, there would be almost no expectations from them because we wouldn't know how long it took them to develop it. But announcing it 5 years ago, hyping it up from time to time and the end result being not stellar is inexcusable. It points to there being a much more widespread problem at Intel. They set lofty goals but the execution is sorely lacking.

 

[GodisanAtheist]

@HurleyBird and @Stuka87 Your mission, should you choose to accept it, is to infilitrate the GPU divison of Intel, stay there as long as you can bear and report from the trenches on what the hell exactly is going on there? We salute you for your service!

- From the sounds of it, you could offer any bum off the street $20 to infiltrate Intel's GPU driver division and they'd not only get the job, they'd quickly be promoted to senior management.

 

[LightningZ71]

Intel has had a decade, a DECADE+, to establish a solid framework for a graphics driver User Interface (from their work on iGPU drivers and ui). They even have something marginally workable for their latest iGPU drivers and the Xe iGPU already that they could have easily built off of and it appears that they are further messing that up.

This was everyone's fear for Intel's effort and it apoears to be well founded.

 

[Insert_Nickname]

Give the lone poor guy at Intel a break who is doing the UI for ALL their software products. Priorities. ARC comes last.

It probably the guys intern doing the actual work. Wouldn't be at all surprised.

Intel has had a decade, a DECADE+, to establish a solid framework for a graphics driver User Interface (from their work on iGPU drivers and ui). They even have something marginally workable for their latest iGPU drivers and the Xe iGPU already that they could have easily built off of and it appears that they are further messing that up.

The "problem" with Intel IGP/GPUs is that they've always been something of an afterthought. It was something added after everything else to fill out the die area. So long as they send something useful to the display, they're good enough.

The original "Extreme" Graphics (2) couldn't even manage that. So that is a very low bar to clear.

 

[shady28]

Well, after watching GN talk about these drivers and how they basically had to re-install on three different test engines, monitors that didn't work with the card, artifacts appearing with smoothsync, and so on - I'm thinking Koduri needs to get the axe.

There's no way I'm putting one of these in my PC. One of the reasons I've stuck to Nvidia for the past decade, even during times they had an inferior price/performance product, is drivers. There's too much important stuff on my PC and time is too valuable to spend a couple of weekends re-installing everything due to a driver issue.

I also agree with GN on this, they should have stuck to simple geometry / rasterizing. Adding in all the features like SmoothSync, Xess, ray tracing and so on was dumb. I'd guess that 95% of people don't know what those are and don't care about them, nor even use AMD/Nvidia variations. I don't - I don't even bother to try to keep up all the new acronyms and what they mean, and haven't for the past decade. I just use the damn thing and it works.

 

[Aikouka]

I would actually love to go in there and replace whoever is currently managing it. I enjoy challenges like that.

I could imagine it being more frustrating than fun. The biggest problem that I could imagine running into is that you know how to fix the problem, but due to some red tape (company processes, upper management meddling, etc.), you simply cannot fix it or it takes an inordinate amount of time to do so.

I also agree with GN on this, they should have stuck to simple geometry / rasterizing. Adding in all the features like SmoothSync, Xess, ray tracing and so on was dumb. I'd guess that 95% of people don't know what those are and don't care about them, nor even use AMD/Nvidia variations. I don't - I don't even bother to try to keep up all the new acronyms and what they mean, and haven't for the past decade. I just use the damn thing and it works.

Given that the ARC cards are not heavy performers, I would imagine that having xeSS might be a worthwhile endeavor. It's why I like GSync and DLSS on laptops as they are generally weaker than their desktop counterparts and not easy to upgrade. Features that can effectively boost the performance can allow these machines to be usable for longer than you would've normally seen.

 

[Aapje]

Lot of blame being sent Lisa's way in some posts here.

But with what looks like the timelines involved, they gave Lisa mission impossible. She never had even a ghost of a chance.

The evidence points rather strongly to Lisa telling the rest of Intel that her team could do it, though.

Also, as others have noted, the iGPU's drivers aren't very good for gaming either. The Xe architecture was already released as part of the 11th gen Core CPU's, so they probably already had access to Xe silicon to build drivers against since early 2020.

 

[Aapje]

Given that the ARC cards are not heavy performers, I would imagine that having xeSS might be a worthwhile endeavor. It's why I like GSync and DLSS on laptops as they are generally weaker than their desktop counterparts and not easy to upgrade. Features that can effectively boost the performance can allow these machines to be usable for longer than you would've normally seen.

The biggest value of the technology is if you run a 4k screen, with relatively weak hardware. At lower resolutions, the gap between temporal upscaling and lowering the graphics settings is smaller.

 

[Stuka87]

I could imagine it being more frustrating than fun. The biggest problem that I could imagine running into is that you know how to fix the problem, but due to some red tape (company processes, upper management meddling, etc.), you simply cannot fix it or it takes an inordinate amount of time to do so.

Well, its clear their current development process is flawed. At a minimum the product manager is at fault. Sure, upper brass could be saying "get this done by this date period". But its up the PM to push back on what can be done in that time period. And it clearly looks like they over promised.

But there appears to clearly be some issues in their SDLC. Its possible the team just isn't managed well, and is spread around in different countries. Or maybe its just a very inexperienced team. Which ever it is, it really needs some clean up.

 

[shady28]

I can't believe that I'm saying this - but I don't think that it's correct blaming QA or the devs. I would have to assume that management pushed the dev and QA teams to release a full list of features no matter what, and that's what they did. I assume that the schedule was extremely harsh and QA and devs were raising all of the alarms, but management just told the to ship whatever even if it's alpha or beta quality at best.

There's no way the devs/QA don't know the state of what they were releasing - it has to be the management that's telling them to ship whatever no matter what.

This, almost certainly, and it probably belongs in Raja's lap.

If they had just not turned on those features, and told marketing to keep quiet about them, they could quietly keep working on them after the card release until they are ready. Then users would get a pleasant surprise in the future, something good for building brand loyalty.
Focus on the basics first, get that working 100%, then do the bells and whistles for the vocal 5% who care.

Honestly from that video, the smart thing to do right now would be to have it automatically disable those features entirely for anything running that is using less than DX 12. Problem solved.

What's unforgivable though are the issues with inability to install patched drivers, inability to uninstall the drivers, and essentially breaking the box requiring a complete OS re-install (3X according to GN) to get it fixed.

This is *exactly* the thing with video drivers that had me switch from AMD GPUs in the 2000s to Nvidia about a decade ago. I am in a mood to give AMD another chance, but it's been 10+ years, those kind of screw-ups have long term costs.

 

[igor_kavinski]

I also agree with GN on this, they should have stuck to simple geometry / rasterizing. Adding in all the features like SmoothSync, Xess, ray tracing and so on was dumb.

BUT BUT BUT Raja's powerpoint slides pitching this GPU to the management included all of these. Not doing them all would make him look untruthful. He's a very honorable person. He delivered. He didn't say in the slides that he would also take care of the drivers.

 

[igor_kavinski]

But there appears to clearly be some issues in their SDLC. Its possible the team just isn't managed well, and is spread around in different countries. Or maybe its just a very inexperienced team. Which ever it is, it really needs some clean up.

Could be the same problem that Microsoft had (and maybe still has?): Testers aren't given enough respect by developers coz if they were that talented, they would be developers. Similarly, maybe inside Intel, a software engineer ranks a lot lower than a hardware or silicon engineer, thus the management isn't as focused on hiring rockstar software engineers/developers.