News Intel GPUs - Intel launches A580

[IntelUser2000]

The stability and game support has improved dramatically, but the performance is still very far off from what you'd expect given the die size (and thus cost). So it seems to be a flaw in the design of the hardware.

TAP said DX11 has to be a labor of love forever. And indeed improvements are coming but due to them laxing on iGPU drivers for the past 20 years, Battlemage is going to inherit the weakness.

You could say it's a "flaw in the hardware" that it's a way different architecture from the conventional dGPUs from AMD/Nvidia, but in all technicality it isn't a flaw but lack of work on the driver side. The press can claim whatever they want, but things always required a lot more hand-tuning, including hardware/software relation in GPUs. All the non-work on the iGPU driver side is coming back to bite them in the behind, HARD.

Intel has a chance to make themselves the Volkswagen of GPUs. That is, if they can avoid being the Trabant of GPUs.

Hopefully not the VW of few years ago.

 

[IntelUser2000]

What if, Intel opensources the drivers and takes a few years of hiatus from the dGPU market, keeping their technology alive in mobile iGPUs? I'm sure Pat doesn't feel good that ARC didn't register enough on Nvidia's threat radar to make them mock it. One good thing about opensource drivers is that Intel may be able to find better GPU driver writers more easily once they decide to get serious again.

Open Source is no magic. Linux, and Ethereum both suffers from the same issue. I always argued that Vitalik Buterin embodied the "decentralized" mindset so much so that it felt like an excuse to be not as responsible, and he take charge until the whole thing develops. If it was a company, then he would have had no choice but to do that. Humanity as a whole swing in such extremes that it's probably the #1 reason they fail. Where's the middle ground?

I have used Linux and it's freaking non-user friendly. You cannot expect the masses to use that. Windows is far, far easier to use. Android only works because it's locked down to the point that password-locked Standard account on Windows feels like it's the Wild Wild West!

And there's a huge disconnect between techies(like us) and the commoners. If they do that, then people will simply discard Intel. They have no choice but to continue.

Unlike software companies that can use the excuse of using post-release patches on the world to fix the problems that should have been fixed at launch, Intel is a hardware focused company and they have no excuse.

Battlemage needs to fix the following at the minimum, period!
-ReBAR performance issues(when you are aiming at the "affordable" market)
-Power use in idle and multi-monitor
-VR support
-Basic stability issues and UI problems with ARC Control

@Exist50 Will Gelsinger(and Team) be the person that most(including me) hope that he would be, or fail anyways, either because Intel is beyond the point they can recover or that he wasn't enough?

He WAS fantastic in his previous roles. 99% employee approval rate? That kind of thing doesn't happen with chance. But there's also the saying that when people get promoted, that they sometimes get work that's beyond their capability, and thus fail. Running Intel is way bigger than any of his previous roles. But that history is why I still believe he can succeed.

Regardless if Battlemage doesn't do the minimum(and way more!) then it would disappoint the market and no matter what they wish sales will tank again and they'll be FORCED to can the dGPU group.

Conversely, if the sales of future ARC generations are fantastic there's going to be no one that'll stop them. There are numerous stories of companies that overlooked what were initially side-projects but were wildly successful, even eventually shifting the entire focus of the company to strengthen that product.

It took 2 years before Brian Kraznich's bad management came into play, when initial rumblings of a possible struggle with Skylake were surfacing. And I can't believe people have fond memories of Skylake. It was mediocre in 2015, when it was released. And in the 3rd year, 14nm was delayed 6+ months.

2 years from Gelsinger is Now. Everything will increasingly be attributed to his doing. Meteorlake will be the first product that'll have some effect, so will Emerald Rapids. In 2024, we'll see the result of his team's work.

 

[beginner99]

I don't think that is really the case. Arc was never intended to be on an Intel process, as far as I'm aware. If anything, I think they isolated it from Intel's fabs to avoid cross-contamination.

I mean this is of course never official said but a mid-range dGPU is a very good fab filler. I therefroe fully agree with jpiniero that ARC was initially planned as such. And they can reuse the tech to some degree for the data center cards spreading out the research costs on that part as well.

I think that the point of Arc is to get all this sweet, sweet AI supercomputer money. GPU's are becoming increasingly useful outside of gaming.

I agree it's for the server market but I think it is also too late. GPUs are a waste of silicon and power for AI. Once the dedicated AI chips roll out, gpus will be obsolete for that. But that is a much bigger issue for NV than intel and there is still a large compute market outside of AI.

 

[Fanatical Meat]

People think it's because of Arc falling short of expectations, but the 2 year delay on PVC is more likely to be the reason he was fired.

Look at the guys who is replacing him, he is a “supply chain sales” expert. That guy isn’t going to let ship dates slip. That guy also sounds like he has a big accounting background which likely will suck. Big companies I have worked for when the accountant focused guys lead it becomes about reducing (business) costs and increasing profit. Those guys suck to work with.
Above comes with the full disclosure I’m a sales & service guy. Accountant goals typically don’t align with my goals or problem solving manner.

 

[Hans Gruber]

They said that Arc A770 on paper was supposed to be equal to or better than a 3070. That wasn't Intel but experts gauging the raw power based on silicon, die size and technology within the Arc architecture. After release it fell between 3060 and 3060ti in best case scenarios, sometimes worse. With driver optimization and important fixes. The Arc A750 has bumped itself up to 3060 performance levels. The A750 was said to be 10-20% below A770 performance.

Yes, ARC was delayed by a year or more so performance for ARC cards is trailing (AMD & Nvidia) the market by more than a year.

Battlemage is said to be the first high end GPU from Intel. Targeting 4080 performance. Just pointing out a few things. I think people thought that ARC was going after the flagship AMD and Nvidia cards. A lot of that was BS hype from this Raja clown for the last 2 or 3 years.

I am watching the driver updates for the ARC series. I am looking to see if the A770 can reach 3070 performance that originally retailed for $550 before the crypto BS.

 

[KompuKare]

I doubt they're making any money on these, but the PC market is pretty crazy right now. From their financial reports, sounds like they have a ton of excess inventory that they need to get through one way or another. They can either do a fire sale, or write it all off, and it seems they chose the former.

Also depends on the defition of making money:
full costs inc R&D and design,
full cost of making a card vs the sale price
cost of the die only
etc.
Anyway a 406mm² die on TSMC's 6nm using the often quoted TSMC defect rate of 0.07, get me 98 good dies per wafer (and A750 might be able to use some of the ~32 defect dies).

At $10,000 that's $102
At $15,000 that's $153
At $20,000 that's $204.

What TSMC charge them is probably unknown and unknowable but back when Intel went seeking wafers their quantity was pretty risky for TSMC no doubt they charged them high.

Obviously that is the raw die without packing, PCB, VRAM etc. Still, I feel at the lower figure they are just about able to sell them at $250, but if TSMC got a good price of desperate Intel €10k might be low and then they might indeed be making a loss.

Since their competition is Navi23 (~about 190 good dies per wafer) and AD104 (about 150 good dies per wafer but on the far cheaper Samsung 8nm), they cannot be doing well.

 

[IntelUser2000]

Look at the guys who is replacing him, he is a “supply chain sales” expert. That guy isn’t going to let ship dates slip.

The IFS guy also has an EE degree by the way. Kraznich had a FAB background, but he was the one that caused Intel to lose the leadership.

I am watching the driver updates for the ARC series. I am looking to see if the A770 can reach 3070 performance that originally retailed for $550 before the crypto BS.

Based on the last video I cannot say even Battlemage will reach 95% of the theoretical target. And it'll only happen due to software, not hardware work. They are literally fixing DX11 performance one game at a time.

DXVK is not a magic button as it has serious compatibility issues, but the worst is with multiplayer and DRM as lots of games consider it almost as cheating.

Believe it or not, I think they are aiming 2x perf/watt with Battlemage. That's 4070 Ti performance. It might sound untenable but remember it likely is on a more advanced process and it's only a second generation, meaning a LOT of learnings. That kind of a "bang" is what's needed so the project doesn't die.

Obviously that is the raw die without packing, PCB, VRAM etc. Still, I feel at the lower figure they are just about able to sell them at $250, but if TSMC got a good price of desperate Intel €10k might be low and then they might indeed be making a loss.

I have a product I want to sell on Amazon for $29.99, and I will be below 50% margins. But the product cost is $8 on the high end and likely even lower.

$100 just for the GPU die is basically losing money. Because $250 is to Newegg, and they have margins too. I think they need to sell to Newegg at max, $200, or even slightly lower at $180-190. GDDR6 spot pricing is $3-4. So we're at $125 now.

-Cost of a 4-6 layer board
-Multiple large polymer capacitors
-High current MOSFETs
-Many, many discrete components
-Power regulator ICs
-Brackets, fans, and sockets
-Marketing, promotions

$100 for GPU may mean they are losing $50-100 per card. At best they are even, and that's just the parts cost!

There's a saying low end GPUs are there to recoup the cost, and mid to high end GPUs are where they make the money. We're talking a mid-high end GPU being priced as a low-mid end card. I can also tell you inflation has seriously affected the electronics market. I had to pay 30-50% higher than last year. $250 is not mid-end anymore. This isn't 2005.

 

[sandorski]

I don't get why he made so much money being such a failure.

He might be on the cusp of being brilliant, but so far not quite. Maybe never.

 

[igor_kavinski]

He might be on the cusp of being brilliant, but so far not quite. Maybe never.

He should start a Youtube channel and make short movies where he shrinks himself down to nm size to explore the innards of semiconductors.

 

[sandorski]

He should start a Youtube channel and make short movies where he shrinks himself down to nm size to explore the innards of semiconductors.

Might be Infotainment.

 

[Aapje]

Battlemage is said to be the first high end GPU from Intel. Targeting 4080 performance.

I'll believe it when I see it. Intel overpromises both in quality and roadmap, so I don't trust them.

 

[Dayman1225]

I'll believe it when I see it. Intel overpromises both in quality and roadmap, so I don't trust them.

Luckily Intel has never made those claims for BMG, only rumours and leaks

 

[igor_kavinski]

No one can blame Raja if Battlemage sucks, right? They still have time to go over its design with a fine toothed comb and nuke some of the trademark Raja design tweaks.

 

[Fanatical Meat]

Personally I have been really wanting intel to be successful. I am completely tired of the nvidia excellent choice with now unrealistic price vs AMD that’s usually good enough for the price.
Sadly I need to think about if intel will continue to support a card when I finally enter the market again 2024 or 2025.

 

[Aapje]

Luckily Intel has never made those claims for BMG, only rumours and leaks

They presented the roadmap shown below to investors and I guess that it depends on what tier you assign to each category. Is Enthusiast the x080 category? If that is not the case, then you could get the weird situation that they hit x080 performance with Celestial and then claim to be done, even though they are obviously then still far off from the x090 tier.

 

[Dribble]

No one can blame Raja if Battlemage sucks, right? They still have time to go over its design with a fine toothed comb and nuke some of the trademark Raja design tweaks.

You talk like Raja was in there re-arranging the transistors - he's a high level manager, he's not designing silcon, he makes no design tweaks. He sets high level strategy, makes high level hires, fights for his orgs place in the Intel pecking order, and is the face for investors/major customers to talk too. As far as that job went - he did ok, I mean Intel went from nothing to something that works (even if not up to Nvidia levels), and got it selling which is a huge step. Certainly a lot more impressive then that Chinese gpu or Larrabee.

 

[igor_kavinski]

he's not designing silcon

Not sure about that. He's described as a chief architect here: https://www.electronicdesign.com/te...koduri-chief-architect-of-amds-graphics-chips

 

[Exist50]

Look at the guys who is replacing him, he is a “supply chain sales” expert. That guy isn’t going to let ship dates slip. That guy also sounds like he has a big accounting background which likely will suck. Big companies I have worked for when the accountant focused guys lead it becomes about reducing (business) costs and increasing profit. Those guys suck to work with.
Above comes with the full disclosure I’m a sales & service guy. Accountant goals typically don’t align with my goals or problem solving manner.

You got a few things mixed up. Stu Pann is taking over IFS, not AXG. Sounds like Raja hasn't really been replaced.

 

[Mopetar]

I don't get why he made so much money being such a failure.

Once you get far enough towards the top there aren't a lot of people in charge that are competent at evaluating people for fitness in those positions. No one person can have the expertise in all of the different areas they'd need to determine if someone will be good.

Even if you hire someone who isn't particularly good, they can still get by if they have a good team underneath them and they don't screw everything up, which can make them look at a lot better than they actually are.

Raja might have been good enough at lower positions to rise to a point where he could even be considered. Once you can get in a top level position you're now in a small list of people who've had that kind of role and companies looking to fill a position will give a lot of weight to past experience in that role even if the results have been mediocre. It seems less risky than promoting someone new who doesn't have any track record.

It basically boils down to "No one ever got fired for buying IBM" but applied to personnel. Who's going to come after a CEO for hiring someone with experience for a high level position if that person doesn't work out? Whereas if they appoint someone new and they fail a lot of people are going to be demanding answers about that kind of reckless maneuvering.

 

[IntelUser2000]

Certainly a lot more impressive then that Chinese gpu or Larrabee.

Can't compare with the Chinese GPU though. Intel has decades of iGPU base. Hence why their compatibility is so much better since they are not a startup starting from scratch.

A GPU startup is unrealistic for that reason, because lots of games need hand-tuning.

 

[igor_kavinski]

A GPU startup is unrealistic for that reason, because lots of games need hand-tuning.

Unless someone perfects an AI that plays games 24/7 and tunes the drivers accordingly.

 

[Exist50]

@Exist50 Will Gelsinger(and Team) be the person that most(including me) hope that he would be, or fail anyways, either because Intel is beyond the point they can recover or that he wasn't enough?

Hah, if I had any real confidence in that answer, I'd speak with my wallet instead of babbling on tech forums. But no, lending an ear here and there to some venting is woefully insufficient to answer that question.

I don't personally know any "old Intel" folk, but second or third hand, they seem to universally respect Gelsinger. The ones currently at Intel seem to genuinely trust that he knows what he's doing, while one I indirectly heard from that left long ago was skeptical that he alone is enough to right the ship. A common thread from the former is that Gelsinger is both capable and willing to make quick, decisive technical decisions of the sort that e.g. Swan would waffle over endlessly and BK would just reliably do the stupidest possible thing. In short, they view him as a proper engineer. Now is that enough? Heck if I know.

My outsider's read on things is that Gelsinger's priorities go something like: IFS >> server CPUs > [everything else]. If he has to make sacrifices in Intel's design side to make IFS a success, he will, or rather, he has. Quite frankly, it's an enormous gamble, and my biggest reason for not touching the stock. I don't think he's neglecting the design side per se, but he can't afford to fully fund both.

 

[Aapje]

My outsider's read on things is that Gelsinger's priorities go something like: IFS >> server CPUs > [everything else]. If he has to make sacrifices in Intel's design side to make IFS a success, he will, or rather, he has. Quite frankly, it's an enormous gamble, and my biggest reason for not touching the stock. I don't think he's neglecting the design side per se, but he can't afford to fully fund both.

I think that Intel has to split up. Both the manufacturing and design side have become so complex that they require the full attention of management. Worse, both sides of the company seem to making each other fail. The design problems result in the foundries having to wait really long for products to be ready for manufacturing, while the foundry problems cause issues for the design teams. The IFS plan is undermined by the foundries being focused on helping internal customers, which means that there is a lack of good tooling and access for external customers, and also apprehension by external customers to choose Intel, for fear of helping their competing products. External customers may also fear that they get a 2nd rate experience, with the best stuff going to internal customers.

By splitting up the company, either side can't hide behind the failings of the rest of the company anymore. Intel Design would become a direct competitor to Nvidia and AMD, and has to start shaping up so they can meet market demand with fewer designs and require a single or at last not 12 steppings for their design.

IFS has to learn how to attract customers that aren't forced to use them.

 

[Strangerer]

I think that Intel has to split up. Both the manufacturing and design side have become so complex that they require the full attention of management. Worse, both sides of the company seem to making each other fail. The design problems result in the foundries having to wait really long for products to be ready for manufacturing, while the foundry problems cause issues for the design teams. The IFS plan is undermined by the foundries being focused on helping internal customers, which means that there is a lack of good tooling and access for external customers, and also apprehension by external customers to choose Intel, for fear of helping their competing products. External customers may also fear that they get a 2nd rate experience, with the best stuff going to internal customers.

By splitting up the company, either side can't hide behind the failings of the rest of the company anymore. Intel Design would become a direct competitor to Nvidia and AMD, and has to start shaping up so they can meet market demand with fewer designs and require a single or at last not 12 steppings for their design.

IFS has to learn how to attract customers that aren't forced to use them.

they've mentioned they want to do a mobileye like spin-off for the foundry BiZ( where its a public company with the majority owned and controlled by intel,intel owns 90-98% of mobileye after the ipo iirc)
and talked about the transistion to an "internal foundry model" multiple months ago https://www.intel.com/content/www/us/en/newsroom/news/intel-embraces-internal-foundry-model.html

 

[Rannar]

can he go to NVIDIA next? he would fit right in with Jensen.