News Intel GPUs - waiting for B770

[mikk]

Tom Petersen talks about the Arc cancellation

 

[IntelUser2000]

Has anyone posted the Semiaccurate article about Intel's driver teams in Russia? It has a date of Sept. 2nd but I only read it today.

I didn't link the article but I commented on it in this thread a while ago.

@maddie Tom Petersen says the same thing about the gaming GPU in his interview in the post right above me.

"...had we not doing the discrete GPU, none of this would have been known, and wouldn't have been factored in the Battlemage GPU and it's across the stack".

He talks about how the client dGPU gets the most attention. No one cares about the integrated graphics and server folks are hard to wedge out of an established mindset.

Because of the vocal and demanding nature of gaming workloads and people who play them WHILE caring about power use, WHILE caring about noise, WHILE caring about cost(this is most important), it forces the designers to very carefully balance all aspects, which affects the thinking elsewhere as well.

Fundamentals DO NOT change and this is one of them. The non-client GPU vendors are dead. D-E-A-D, period!

 

[IntelUser2000]

On Twitter:

John Ospanov: There are rumours of Intel dropping it's video card business. Any update on that?
Dr. Ian Cuttress: They're not.
John Ospanov: Are you sure? Moore's law is dead claims so.
Dr. Ian Cuttress: Sorry I forgot MLID was the authority.



Hopefully this takes MLID down a notch, or ten!

 

[jpiniero]

I think you guys are underestimating how hard it's going to be for Intel to even pretend to continue to be a leading edge foundry while being cash flow negative, even with the free Government money. And I'm sure they really, really, really want to keep the dividend.

 

[mikk]

One more A730M review: https://www.hardwareluxx.de/index.p...-medion-erazer-major-x10-im-test.html?start=2

 

[coercitiv]

I think you guys are underestimating how hard it's going to be for Intel to even pretend to continue to be a leading edge foundry while being cash flow negative, even with the free Government money. And I'm sure they really, really, really want to keep the dividend.

Giving up the consumer dGPU business isn't a simple binary argument. Even a mediocre dGPU business can add up to foundry volumes, which in turn helps their momentum. The same applies to their iGPU efforts, the increased mind share and R&D from dGPU may push Intel's consumer mobile segment in the medium & long term. Whatever they choose between dividends and Arc, I think there's a potential for unintended consequences in the future.

 

[Aapje]

@IntelUser2000

MLID has been walking his claim back a bit. Now he's saying that Intel will only release one chip for the next gen.

 

[Heartbreaker]

On Twitter:

John Ospanov: There are rumours of Intel dropping it's video card business. Any update on that?
Dr. Ian Cuttress: They're not.
John Ospanov: Are you sure? Moore's law is dead claims so.
Dr. Ian Cuttress: Sorry I forgot MLID was the authority.



Hopefully they takes MLID down a notch, or ten!

MLID audience doesn't actually seem to notice that he continually gets stuff wrong.

 

[darkswordsman17]

Honestly it seems like Intel wants to keep with Arc just so the former tech site people they hired can trash the "YouTubers and social media influencers" out of spite. But that also seems like it has potential to further lead them into baffling propaganda like userbenchmark or whatever where they're spouting such obvious nonsense it undermines them. If Intel ever gets directly linked to that, or worse starts spouting that type of drivel directly, it will completely trash their rep with enthusiasts. And it sure seems like they might be intent on doing just that over these dGPUs.

The reality though, is that if things don't drastically improve its gonna bring heat on Gelsinger. For IntelUser's claim that they absolutely had to do client dGPU because whatever reason he keeps changing to, I don't know if he realizes that as it is now, Intel is shooting themselves in the foot with their dGPU. But if things don't improve but they still insist on trying to make it a thing, it could end up being more like shooting themselves in the femoral artery, as the only way that Intel will be able to make inroads is to push Intel dGPU/CPU deals to OEMs, but OEMs are not gonna put up with people returning systems because of driver issues, so it could mean OEMs decide they could also start looking to get away from Intel CPUs. AMD is not only viable but seems like it carries a bit of a premium cachet these days (at least CPU wise), but the bigger issue is it might push OEMs to actually really start moving to ARM stuff for their volume shipments. And the timing is poor for Intel as I think the deal between Microsoft and Qualcomm for Windows ARM is over (or just about). Intel might get lucky as there's some rift in the ARM space due to Qualcomm seemingly wanting to make performance ARM designs targeting PCs and the licensing deal that ARM is suing over (which that might be due to the Qualcomm/Microsoft situation which has stifled ARM there), but is also possible that gets resolved without too much issue. But there's plenty of others (including both Nvidia and AMD) and so things could pretty quickly change.

Frankly, it seems like Intel never should have released this first Arc as a product, other than possibly as a dedicated media processing card. Which actually could have built some interest, if they designed a streaming card (or NUC like device). Instead they should have sent these out to devs, getting feedback and bug fixing while they worked to improve, also building a rapport with them in the process. Instead we're getting them fighting with friggin rumor mongers, seemingly unaware that's making them look like even bigger clowns.

 

[VirtualLarry]

but the bigger issue is it might push OEMs to actually really start moving to ARM stuff for their volume shipments. And the timing is poor for Intel as I think the deal between Microsoft and Qualcomm for Windows ARM is over (or just about). Intel might get lucky as there's some rift in the ARM space due to Qualcomm seemingly wanting to make performance ARM designs targeting PCs and the licensing deal that ARM is suing over (which that might be due to the Qualcomm/Microsoft situation which has stifled ARM there), but is also possible that gets resolved without too much issue. But there's plenty of others (including both Nvidia and AMD) and so things could pretty quickly change.

Quite frankly, being a long-time PC (x86/x64) enthusiast, it frightens the crap out of me that, if the winds blow where they may, that we might just see "Windows-on-ARM" PC Desktops on the budget PC shelf at Walmart, for example. That there would be a split between the "average joe's PC" and our enthusiast PC, and not just a matter of grade of part class, but a wholly different taxonomy of parts.

Granted, I know that Intel experimented with ATOM-based Celeron and later, "Pentium Silver". but they've been pretty good about it lately (or OEMs have), about keeping the abominations of processors like Pentium Silver in laptops, and the occasional budget AIO PC, where low-power due to slim profile and nearly no cooling suits it as well, better than your average 65W Desktop TDP CPU.

 

[igor_kavinski]

Quite frankly, being a long-time PC (x86/x64) enthusiast, it frightens the crap out of me that, if the winds blow where they may, that we might just see "Windows-on-ARM" PC Desktops on the budget PC shelf at Walmart, for example.

Same here. There will be compatibility sacrifices made for going with WoArm. It wouldn't be ideal. But then, what if we get 25 hour battery life with 99Whr battery? That might make it worth the trade-off. But it might kill budget x86. And in future, x86 itself for good. That is not something I want to see in my lifetime.

 

[Revolution 11]

Remember when they took a 10 billion dollar plus loss trying to penetrate the mobile market, then failed and laid off 12.000 employees? After blowing golden opportunities the decade before by selling XScale. And passing up a chance to power the iPhone. That's probably what trying to supplant AMD would be like this late in the game. They missed their window, and would have to play catch up. All to win a market with low margins? Unlikely.

God, anyone remember the complete fail purchase of McAfee by Intel for $7.5 billion back in 2011? That money could have been used to better effect in almost any other product field. Imagine if Intel had slowly and steadily developed a dGPU business for datacenters with that money. Imagine if it had invested that money more in mobile or spent the money on doing IFS a decade early.

So many lost opportunities and for what, freaking McAfee.

 

[Revolution 11]

Also isn't that the cost over 5 years. So 700 Million/year is Not that big of a deal for Intel. Especially considering They still need to design the core architectures, and still need to design the drivers, and UI, so killing discrete GPUs probably wouldn't even save half of that. Maybe 200 Million.

200 million/year savings isn't make or break for Intel. It would be stupid to can it over that.

Stupid? Yes it is.

But Intel has a bad habit of over-emphasizing what would make investors happy and not enough on what its long-term vision will be. Even with Pat as CEO, who has done a good job advocating for a long-term vision, you still see hints of the classic Intel that will sacrifice key investments in strategic product segments just to improve short-term investor impressions or concerns.

Imagine if Nvidia had divested or shut down CUDA development because it was consuming too many resources in the first few years of the concept. Where would it be?

In the same place as Intel right now.

 

[Heartbreaker]

But Intel has a bad habit of over-emphasizing what would make investors happy and not enough on what its long-term vision will be.

How many years did intel spend on 3D Xpoint (AKA Optane)?

How many years on SSDs?

How many years on McAfee?

How many years on drones?

How many years on modems?

It would seem Intel habitually spends a great many excess years on side projects.

None of those has the core importance of GPU, nor are they as well aligned with Intel expertise (compute).

Given past history Intel should be spending many more years on trying to make their GPU business work.

 

[moinmoin]

It would seem Intel habitually spends a great many excess years on side projects.

Yeah, in those excess years Intel absolutely dominated the market in the core business but had zero idea how to expand beyond it.

 

[Heartbreaker]

Yeah, in those excess years Intel absolutely dominated the market in the core business but had zero idea how to expand beyond it.

Just pointing out, that a claim that Intel gives up easy, doesn't really track with history.

You could claim we are in uncharted territory, but that's different than holding up history as evidence of things that really didn't happen.

 

[Revolution 11]

Yeah, in those excess years Intel absolutely dominated the market in the core business but had zero idea how to expand beyond it.

It's the curse of success (and also due to the Innovator's Dilemma). Many successful dominating companies have died to this hubris and willing blindness to the future.

 

[jpiniero]

How many years on modems?

Intel made a ton of money off of the modem. It's just that the business got 10 nm'd.

 

[Heartbreaker]

Intel made a ton of money off of the modem.

Where? After how many years of losses? It seems mostly to have been bundled with their massively money losing smartphone/tablet chips group.

WSJ said the selling iPhone modems was losing money:

Intel Finally Makes the Right Call on Modems

Intel’s news that it was effectively exiting the business of making modem chips for smartphones was not so surprising. Its iPhone business was losing money and straining capacity—and likely to go away soon anyway.

www.wsj.com

 

[jpiniero]

Where? After how many years of losses? It seems mostly to have been bundled with their massively money losing smartphone/tablet chips group.

WSJ said the selling iPhone modems was losing money:

Intel Finally Makes the Right Call on Modems

Intel’s news that it was effectively exiting the business of making modem chips for smartphones was not so surprising. Its iPhone business was losing money and straining capacity—and likely to go away soon anyway.

www.wsj.com

They did burn a ton of money trying to make the 5G modem.

 

[DrMrLordX]

Quite frankly, being a long-time PC (x86/x64) enthusiast, it frightens the crap out of me that, if the winds blow where they may, that we might just see "Windows-on-ARM" PC Desktops on the budget PC shelf at Walmart, for example.

They'll probably use the same peripherals and even internals as standard PCs. Only difference will be the board and CPU. They may also use ATX12VO PSUs but otherwise they'll still have PCIe slots, m.2 slots, maybe SATA slots for "legacy" SATA products (lol), external USB, that kind of thing. And they may use SO-DIMMs instead of DIMMs but we should be used to those by now.

Granted, I know that Intel experimented with ATOM-based Celeron and later, "Pentium Silver". but they've been pretty good about it lately (or OEMs have), about keeping the abominations of processors like Pentium Silver in laptops, and the occasional budget AIO PC, where low-power due to slim profile and nearly no cooling suits it as well, better than your average 65W Desktop TDP CPU.

Bite your tongue! Small Tremont and Gracemont-based machines would be a blessing, but there aren't many Tremont-based AiOs out there, and to date I haven't seen any Gracemont/Alder Lake-N units yet for desktop. You can get Tremont-based SFF PCs.

 

[VirtualLarry]

But at least, apart from the system chipset drivers being different for ATOM (a lot of SoC-type features built-in), at least the peripheral / add-on device drivers were binary-compatible between ATOM and Core, AFAIK.

Now, with ARM on Desktop, if you want to install a GPU, you no longer need "NVidia drivers", but "ARM-specific NVidia drivers" for that PCI device-id. It adds an additional layer of complexity to servicing those machines.

But if MS does things right, then Windows-on-ARM should be as familiar as ever.

 

[DrMrLordX]

Unless MS is a bunch of puds (and maybe they are), the WARM driver model should be very similar to standard Windows drivers. The only concern I would have with WARM is MS making the platform overly-dependent on the MS Store.

 

[Insert_Nickname]

Now, with ARM on Desktop, if you want to install a GPU, you no longer need "NVidia drivers", but "ARM-specific NVidia drivers" for that PCI device-id. It adds an additional layer of complexity to servicing those machines.

Not really different to x86/x64 versions 10-15 years ago. No worries there.

But if MS does things right, then Windows-on-ARM should be as familiar as ever.

Indeed. With Windows NT all that really -needs- to be architecture specific are drivers. All user programs should run fine through the compatibility layer.

Did you know NT was actually developed on the i860 architecture originally? It's actually quite portable between architectures, as it was developed that way. It has run on i860, DEC Alpha, Itanium, MIPS and PowerPC. ARM(64) is nothing special in that regard.

 

[moinmoin]

Just pointing out, that a claim that Intel gives up easy, doesn't really track with history.

You could claim we are in uncharted territory, but that's different than holding up history as evidence of things that really didn't happen.

I think you misunderstood my take, never intended to imply Intel didn't try or gave up to quickly. That wasn't the case. Intel just didn't know how to expand from its core business, it didn't know what expansions could be worth pursuing (I mean, McAfee?), how to approach them, and at what point to cut losses (Optane is the most recent example of that).

Historically I'd say Intel fell more often to the sunk cost fallacy than giving up to quickly. But I'm well aware that that's a conundrum Intel can't win either way. Easy solution: make it successful.

That all said, I fully agree GPUs being compute should have been considered a core business and should have gotten more investment earlier already. Can only guess that internally GPU compute was seen as direct competition to Core similarly to how Atom appears to have been seen and received less support, and that's why it wasn't pursued aside of the halfhearted Atom derived Larrabee/Knights stuff.