Vendor lockout of GPUs? *spawn*

[DavidGraham]

At the end of the day Intel still achieved their end goal which was to stop Nvidia from producing integrated graphics chipsets ...

Yeah, a mutual goal that NVIDIA wanted to reach in the first place. Once more NVIDIA couldn't care less about their shrinking chipset business.

Nvidia could've very well pressed Intel into court but they didn't so they decided to chicken out instead because they thought that the risk/reward benefit of continuing litigation wasn't worth it ...

Nope, they took a very large sum of money from Intel, a sum that probably compensated them well enough not to care about chipsets anymore.

 

[Lurkmass]

Nope, they took a very large sum of money from Intel, a sum that probably compensated them well enough not to care about chipsets anymore.

$1.5B in installments of 6 years is virtually chump change for Intel who was making well over $10B annually in profit during that period. Nvidia accepted a relatively raw deal since they weren't even confident that they'd get better terms in court ...

 

[JasonLD]

Now that Intel has entered the discrete graphics market, I cannot see how it will be politically acceptable for their CPUs to retain compatibility with GPUs from other competitors in the future. In order to gain any genuine traction it would have to come at the expense of their competitors so conditions would need to be changed in following that line of thought ...

The question now is when graphics acceleration will be exclusively tied with the same CPU vendors ?

It will be the direct violation of the antitrust law lol. Not going to happen.

 

[Lurkmass]

It will be the direct violation of the antitrust law lol. Not going to happen.

It really isn't since a similar scenario played out in the past. There's no law where they have to stick to using PCIe for the indefinite future or make their proprietary interfaces compatible with other hardware vendors. Both AMD and Intel blocked Nvidia from producing integrated graphics because they could do it on their own. Now that both AMD and Intel can produce discrete graphics hardware the only natural outcome is that others are going to be block out again ...

I don't imagine regulators will even bat an eye because the move isn't reducing the total amount of competition either. Instead of the options being just AMD + Nvidia in the past, going forward it would be AMD + Intel. The end result would still be 2 vendors so it isn't really anti-competitive to replace a vendor ...

 

[pcchen]

It really isn't since a similar scenario played out in the past. There's no law where they have to stick to using PCIe for the indefinite future or make their proprietary interfaces compatible with other hardware vendors. Both AMD and Intel blocked Nvidia from producing integrated graphics because they could do it on their own. Now that both AMD and Intel can produce discrete graphics hardware the only natural outcome is that others are going to be block out again ...

You know, back then Intel produced their own "proprietary interface" specifically targeting graphics cards, and they did have their own graphics chip, specifically for using the interface. And they didn't use that to block other graphics vendors out.
Today Intel's GPU is not at a better position compared to the competitors any more than i740 used to be, not to mention that Intel's advantage on CPU is much worse than in 1998. I really don't see why Intel would want to do this, unless they are being stupid (but Pat Gelsinger is not a stupid man).

Maybe some years later when Intel's GPU is competitive with NVIDIA's it might be possible, but if it happens NVIDIA would be in a much larger trouble even without this.

 

[Bondrewd]

Now that both AMD and Intel can produce discrete graphics hardware the only natural outcome is that others are going to be block out again ...

Yes but applies to DC parts only (sans PVC which has no CXL host attach or anything like that) and NV is bolting a BIG APU of their own anyway.
Absolutely irrelevant for DG2 etc etc.

 

[Lurkmass]

You know, back then Intel produced their own "proprietary interface" specifically targeting graphics cards, and they did have their own graphics chip, specifically for using the interface. And they didn't use that to block other graphics vendors out.
Today Intel's GPU is not at a better position compared to the competitors any more than i740 used to be, not to mention that Intel's advantage on CPU is much worse than in 1998. I really don't see why Intel would want to do this, unless they are being stupid (but Pat Gelsinger is not a stupid man).

The i740 also existed well before AMD purchased ATi so political circumstances during that time would not give Intel enough bargaining power to control the discrete graphics market in totality because there were other x86 CPUs vendors who didn't yet produce their own graphics technology. The other x86 vendors at the time had no choice but to accept compatibility for graphics hardware vendors in order to stay competitive which kept Intel from taking over total control of the discrete graphics market. Similarly when AMD purchased ATi well after Intel had pulled out discrete graphics market, it gave them no bargaining power since Intel was was the most popular x86 vendor by far so AMD had to tolerate being compatible with other graphics vendors just like Intel to stay competitive ...

In a way Nvidia's prosperity was based on delicate political conditions of being able to secure compatibility with x86 platforms so they kept hoping that they wouldn't have to compete with all of the significant x86 vendors in order for at least one of them to bail them out. There's no turning back to the previous state, the reality is that all other discrete graphics vendors are now at the absolute mercy of either AMD or Intel. They can decide at their own whims whether the others will fail or succeed and what happens to them is totally out of their control. The old notions that we once held aren't going to apply anymore ...

Maybe some years later when Intel's GPU is competitive with NVIDIA's it might be possible, but if it happens NVIDIA would be in a much larger trouble even without this.

I would think that if Intel came closer to AMD's offerings that they would start locking graphics acceleration to their own GPUs because at that point what Nvidia offers becomes an irrelevant comparison point when they can be blacklisted since they can't easily secure compatibility with an x86 platform. There's simply too many numerous tactical advantages to not make this move ...

1. The leading competitor's software/driver stack advantage becomes redundant overnight since no one can run their drivers anymore!
2. There'd be more wafer capacity available all around for which AMD/Intel would be able to place more orders of ...
3. 80+% of the market would be immediately up for grabs and even if Intel isn't able to capitalize on all of it would sure beat likely staying in the single digits for a while.

Not many people realize just how profound the changes in the balance of power would be with Intel's entry in the current context ...

 

[entity279]

Can this silly "Intel should block other's GPUs" get its own thread?

 

[CarstenS]

It will be the direct violation of the antitrust law lol. Not going to happen.

What they can do though is orchestrate support for new features in GPUs and CPUs in a temporally synced fashion, i.e. PCIe 6 for GPU and CPU, in order to drive up attach rates.

And speaking of antitrust - there's always the chance of giving (massive) rebates to their favourite customers for pairing their brand in the socket and the PCIe slot. This has been pretty hard to prove in the past and massive rebates were modus operandi in the server space as well for years.

 

[JasonLD]

Can this silly "Intel should block other's GPUs" get its own thread?

Looking at Lurkmass's posts so far, it reads more like anti-Nvidia campaign than "Intel should block other's GPUs" lol. Not even worth discussing imo.

 

[troyan]

And speaking of antitrust - there's always the chance of giving (massive) rebates to their favourite customers for pairing their brand in the socket and the PCIe slot. This has been pretty hard to prove in the past and massive rebates were modus operandi in the server space as well for years.

Qualcomm was fined for it.

 

[trinibwoy]

Looking at Lurkmass's posts so far, it reads more like anti-Nvidia campaign than "Intel should block other's GPUs" lol. Not even worth discussing imo.

His proposal is intriguing but misguided based on an assumption of unlimited Intel leverage over the market. I can’t think of an example where a company thrived by intentionally fragmenting an established standards based industry. The examples he’s using (integrated graphics, chiplets) aren’t relevant as consumers have never had an expectation of being able to choose which chiplet they use on a motherboard. Unless Intel is going to start soldering GPUs to motherboards the idea is a thought experiment at best.

 

[DavidGraham]

Nvidia accepted a relatively raw deal since they weren't even confident that they'd get better terms in court ...

You have it backwards, Intel was the one who requested the settlement, their position was weak, and NVIDIA's leverage over graphics IP was much stronger.

so it isn't really anti-competitive to replace a vendor ...

It is, if you do it by blocking access to your platform. Intel will be fined and forced to give access to others either free of charge or through a fee. It happened once and it will happen again.

 

[Silent_Buddha]

His proposal is intriguing but misguided based on an assumption of unlimited Intel leverage over the market. I can’t think of an example where a company thrived by intentionally fragmenting an established standards based industry. The examples he’s using (integrated graphics, chiplets) aren’t relevant as consumers have never had an expectation of being able to choose which chiplet they use on a motherboard. Unless Intel is going to start soldering GPUs to motherboards the idea is a thought experiment at best.

Yup. Just looking at my current AMD motherboard, it uses Intel chips for both the LAN and WiFi. Often things like this happen because one vendor or another just don't want to make their own version of a component if there is already one on the market. Or MB makers might want to choose something other than a vendor provided solution if a solution from that vendor's competitor is either cheaper or better performing.

Hell, I remember for years MB makers would avoid AMD's USB solutions because they were up to half the speed of Intel's solution which just looks bad, especially when they showed up in online MB reviews. So they used USB chips from various 3rd party USB chip providers instead of using AMD's USB solutions for anything that wasn't a budget motherboard.

Regards,
SB

 

[Davros]

Just to point out if nvidia and amd gpu's stop working with intel cpu's, I stop working with intel cpu's

 

[Lurkmass]

You have it backwards, Intel was the one who requested the settlement, their position was weak, and NVIDIA's leverage over graphics IP was much stronger.

For a "weak position", $1.5B was enough to make Nvidia fold all of their disputes and share their graphics IP ...

It is, if you do it by blocking access to your platform. Intel will be fined and forced to give access to others either free of charge or through a fee. It happened once and it will happen again.

Source and when was this precedent established ?

 

[Qesa]

For a "weak position", $1.5B was enough to make Nvidia fold all of their disputes and share their graphics IP ...

It's not much compared to the size of the company today, but $1.5B then was ~5x their annual net income

 

[Lurkmass]

It's not much compared to the size of the company today, but $1.5B then was ~5x their annual net income

For Nvidia sure but Intel was making over $11B annually just before the settlement ...

It's highly doubtful that Intel are interested in playing the exact same game in the long-term seeing how hard it currently is for AMD to make inroads. If the current rules aren't working for AMD then what makes people think that those same exact rules will somehow work out for Intel ? Since Nvidia now has a conflict of interest between both AMD and Intel by directly competing with them, reevaluating the rules will start to become an ever attractive option to them rather than keeping the same rules as before ...

The management behind Intel are not going to be one bit happy to see that they're pouring in tons of resources only to see their new products failing so they're going to want to block the others sooner rather than later. Similarly, AMD's management are going to be annoyed of the fact that the sales of their products are going to be subverted to another competitor so the only way that they're going to tolerate keeping their competitor afloat is if they directly contribute to AMD's bottom line because otherwise they'd prefer just having 2 vendors over 3 vendors if the others aren't going to share their profits with the only platform available to them ...

 

[Deleted member 90741]

Vendor lockout has not been a good strategy for many reason cited before, either from legal perspective or otherwise.
Integration is however a different matter.
If Intel manage to put a good dGPU for mobile, they could possibly alter the GPU landscape.
This would be on top of the massive market share that Intel iGPUs already command
Intel spends a huge amount of marketing fund to help OEMs adopt their CPU designs, adding more money to support GPU is a given.

https://www.3dcenter.org/news/die-grafikchip-und-grafikkarten-marktanteile-im-zweiten-quartal-2021




In the very short term, with RDNA2 in Rembrandt and Xe in ADL, this could swing in many different ways

Longer term, Zen4 is shipping with RDNA2 and RPL with stronger Xe variants

Remains to be seen how the pie gets sliced.

 

[JoeJ]

The question now is when graphics acceleration will be exclusively tied with the same CPU vendors ?

I did not follow this discussion, but obviously the answer is: After iGPUs become mainstream, because discrete GPUs became too expensive in comparison.
I really see no other option to move on anyway. FPU merged with CPU too, and nobody complains anymore.