Arm Pioneer: Nvidia's Grace CPU Is Proof That It Will 'Compete Unfairly' (tomshardware.com) 122
RealNeoMorpheus writes: Arm pioneer Hermann Hauser has once again criticized Nvidia's plan to acquire the semiconductor design company, with The Telegraph reporting Sunday that he believes Nvidia is "clearly showing it will compete unfairly" if the deal is approved. Hauser's concerns reportedly centered on the Grace processor Nvidia announced at GTC 2021. The company's first Arm-based CPU will connect to high-end GPUs via NVLink, which purportedly offers data transfer speeds up to 900 GBps. That's significantly faster than other technologies -- it's also exclusively available to Nvidia.
This is why Hauser told The Telegraph that he believes using a proprietary interface like NVLink could end up "locking customers into [Nvidia] products," which "clearly shows that they will compete unfairly with other Arm-based server companies such as Amazon and Fujitsu," rather than retaining Arm's neutrality. [...] Nvidia told The Telegraph that Hauser "does not understand what Grace will do or its benefits to Arm" and that "we have been working on Grace using off-the-shelf Arm technology, available to all Arm licensees, long before we agreed to acquire Arm."
This is why Hauser told The Telegraph that he believes using a proprietary interface like NVLink could end up "locking customers into [Nvidia] products," which "clearly shows that they will compete unfairly with other Arm-based server companies such as Amazon and Fujitsu," rather than retaining Arm's neutrality. [...] Nvidia told The Telegraph that Hauser "does not understand what Grace will do or its benefits to Arm" and that "we have been working on Grace using off-the-shelf Arm technology, available to all Arm licensees, long before we agreed to acquire Arm."
"agreed to acquire Arm." (Score:1)
Re: "agreed to acquire Arm." (Score:2)
Re:"agreed to acquire Arm." (Score:5, Interesting)
Nvidia is a problem.
RISC-V is the solution.
Dell, HP, Lenovo, Acer, and Google should form a consortium to optimize RISC-V for laptops, desktops, servers, and Android phones/tablets.
They can get out from under the yoke of both Intel and Nvidia, and be able to compete with Apple's M1/M2.
Re: "agreed to acquire Arm." (Score:1, Interesting)
Re: (Score:3)
It isn't just the CPU's, it is the entire ARM ecosystem that RISC-V will take a long time to replicate. To do that will require major company backing. Yet any company or companies doing that will lock down their IP.
Re: (Score:3)
RISC-V hasn't been available for years either. Several specs for features that are taken for granted in a desktop CPU, which 64-bit ARM chips have had since 64-bit ARM was spec'd, are still not finalised. Only one or two semiconductor companies have dared to implement a spec that is still in draft stage.
The reason why Apple M1 is so fast relative to the competition is not just because it is ARM or wide issue, but also largely about the RAM interface -- which so far requires the RAM to be in the same package
Re:"agreed to acquire Arm." (Score:4, Insightful)
It's unfortunately not that simple. To replicate the work that ARM has done to get performance, both in terms of processing speed and power consumption, is a major long term undertaking. Probably one that is worth doing, but in the medium term it's not going to help much if Nvidia decides to get nasty.
There are probably patent issues too. Patents seem to be the reason why we aren't seeing something like the big.LITTLE architecture on x86, where you have a mix of performance cores and low power cores. Not just the general concept, but a lot of the implementation details that make it work seamlessly are patented.
Re: (Score:2)
Re: (Score:2)
Yes, not to mention that having a dominant architecture backed by "Dell, HP, Lenovo, Acer, and Google" is not an improvement over what exists today.
Claiming that "Nvidia is a problem" and "RISC-V is the solution" is totally arbitrary. The purchase hasn't been made, Nvidia hasn't done what is claimed they will do, and RISC-V is not currently competitive. All these things can change...and change again and asserting that what we need is a bunch of technology developed and released into open source is not an
Re: (Score:2)
Ahem. [techbriefly.com]
AMD might release a hybrid x86 CPU with big.LITTLE architecture soon
Re: (Score:2)
Re: (Score:2)
It's mostly the organization of the cores and the fine detail of how that works that is patented.
The first version of big.LITTLE only allowed all the performance cores or all the low power cores to run at once, no mixing. The OS had to switch between the two sets, and it was quiet slow.
The next version presented the OS with 4 virtual cores, each of which was made up for a performance and an efficient core. Each virtual core could only have one CPU active at once. Switching was seamless but left a lot of per
Re: (Score:2)
There are probably patent issues too. Patents seem to be the reason why we aren't seeing something like the big.LITTLE architecture on x86, where you have a mix of performance cores and low power cores.
Both AMD and Intel (with Alder Lake) are releasing such things.
Re: (Score:2)
Yeah, RISC is good
Why the surprise? (Score:5, Insightful)
It's frickin' 40 BILLION DOLLARS. You don't spend that kind of cash to then let ARM run business as usual.
Of COURSE Nvidia is going to leverage their acquisition to provide themselves with a commercial advantage.
Anyone who says they thought otherwise is either lying or mentally deficient.
Re: (Score:2)
Re: (Score:2)
NVidia is not paying 40 billion, nVidia doesn't even have 40 billion - it's just stock value which means nothing.
nVidia has maybe, depending on who you ask 3-5 billion USD of actual net cash.
So if you were offered 1 million USD worth of stocks in for example Apple you would decline to take that? Because it means nothing?
Re: (Score:2)
That is not what they were saying. Derplord was trying to get across that $1m in stock is NOT the same as $1m cash.
Using your own metaphor:
Re: (Score:2)
$1 million in Apple stock is effectively the equivalent of $1 million in cash because you can sell it with no significant effect on the market for AAPL. The company has a market cap of over $2 trillion so your stock is a drop in the bucket.
$40 billion in NVIDIA stock is nothing like $40 billion in cash. (The actual deal is for 10% of NVIDIA plus an unspecified amount of cash, but given current market values it's not likely to be over $5 billion in cash.) The company's current market cap is about $376 billio
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Except that most M&A deals aren't executed solely in cash, they're usually a combination of cash and shares, so regardless of what you think of NVIDIA's market capitalization, it still certainly benefits them in the transaction.
Re: (Score:2)
it's just stock value which means nothing.
Aptly named, I see.
Re:Why the surprise? (Score:5, Insightful)
Unless there is reason to believe nvidia is lying about this specifically, and I don't see any, nvidia was working on this before they even acquired ARM. And frankly, they likely don't need anything from the acquisition to continue.
This is about a proprietary connection to their GPU from their CPU. And guess what? That connection (nvlink) is not part of ARM. It's not part of the ARM acquisition. They were going to do it whether they were an ARM licensee, or an arm licensor. So this has nothing whatsoever to do with the acquisition.
Frankly, I do not believe for a second that nvidia won't license nvlink. They want to be out in front with it with their own solutions. It's difficult to get upset by that, unless one is trying really hard.
Re: (Score:2)
Unless there is reason to believe nvidia is lying about this specifically, and I don't see any, nvidia was working on this before they even acquired ARM.
More to the point they haven't acquired ARM and they've developed this product. You're also right about NVLINK as this is nothing to do with ARM and this whole product has been developed as an ARM licensee.
Re: (Score:2)
The whole point of mergers is to provide "synergies" - efficiencies of scale or integration, which make the merged company more efficient and thus lower the price for consumers. If a merger won't improve a company's efficiency, but rather is designed to degrade competitors' efficiency and raise the price for consumers, why should we allow the merger to take place? "OF COURSE" they want it, but of course we should not allow it. This is exactly what antitrust law is intended to prevent.
Re: Why the surprise? (Score:2)
Nvidia likes to introduce artificial barries (Score:5, Informative)
I have tried using nvidia cards in a VM, and quickly learned about the "Error 43", and the very large (and helpful) communities that tries to solve that problem. Basically nvidia goes out of their way to make sure you pay more for "enterprise" features.
So, you have the desktop GeForce 1070 ($300), and want to use it on a workstation? They would not even give you Windows Server drivers for it. (You need to by Quadro P4000 for $600, which is the same chip with better cooling and ECC RAM).
So, you said "no worries, I can put that in a VM", well, then you run into Error 43, when they detect the "consumer" card running under virtualization. Even Quadro will not cut that time. You will need P100 data center card, which was a cool $6,000 back in time.
They even offer multiple VMs sharing a single card, again only for the data center version.
Let's not even get into what they do with low quality silicon (hint: sell as crypto miner cards). IF they would be able to distinguish Machine Learning, I am pretty sure, they would block even that on consumer cards, and sell a different version only for ML.
Overall, nvidia will squeeze the customer to the fullest extent. If history is any indication, ARM under nvidia will most likely see similar schemes (want the latest features? sure price has gone 10x)!
Re:Nvidia likes to introduce artificial barries (Score:4)
If AMD solved the reset bug [nicksherlock.com] then there would be some kind of competition.
Re: (Score:3)
Yeah, I always love AMD proprietary windows drivers - thousands of annoying bugs that turn your compute powerhouse into a guessing game of peeking around in various internet web boards.
If this kind of bugs still exist in the Linux open-source drivers , then there is no hope for AMD users!
Re: (Score:2)
Re: Nvidia likes to introduce artificial barries (Score:1)
The $600 price tag didnâ(TM)t include only the additional costs for better cooling and ecc-ram it included the software engineering costs to develop and maintain the windows server drivers and possibly other drivers and libraries that are enabled for use with those cards, that the average consumer would not need nor usually want.
Similarly quadro cards include additional software and libraries for those market segments as well as some hardware differences.
And yes if Nvidia could figure out how to block
Re: Nvidia likes to introduce artificial barries (Score:5, Informative)
Yes, it would be reasonable if they sold premium features separately. For example, multiple guest virtualization on a single card really requires effort on their end. Cloud providers would be happy to pay for that feature.
However nvidia's treachery goes much further.
They actively block features if they can squeeze more money out of you. Blocking drivers on VFIO is one example, since if you can successfully hide your VM, the driver just works.
And they block open source drivers. By default cards limit the clock speeds to minimum (idle desktop), so your state of the art gaming GPU will act as a 10 year old potato with Nouveau. There is some hope (again with lots of manual labor):
https://www.phoronix.com/scan.... [phoronix.com]
Yes, they are innovative. But they are much closer to Oracle level tactics than a regular open source friendly hardware vendor.
Re: (Score:3)
Re: Nvidia likes to introduce artificial barries (Score:2)
VM support for consumer cards is now supported at least (in beta anyway, or in other words they don't want to bothered if it doesn't work).
Re: (Score:2)
IF they would be able to distinguish Machine Learning, I am pretty sure, they would block even that on consumer cards, and sell a different version only for ML.
But they include Tensor Cores on the consumer cards specifically for machine learning. If they didn't want consumer cards to do ML they wouldn't include Tensor Cores on them.
Re: (Score:2)
I have tried using nvidia cards in a VM, and quickly learned about the "Error 43", and the very large (and helpful) communities that tries to solve that problem. Basically nvidia goes out of their way to make sure you pay more for "enterprise" features.
So, you have the desktop GeForce 1070 ($300), and want to use it on a workstation? They would not even give you Windows Server drivers for it. (You need to by Quadro P4000 for $600, which is the same chip with better cooling and ECC RAM).
So, you said "no worries, I can put that in a VM", well, then you run into Error 43, when they detect the "consumer" card running under virtualization. Even Quadro will not cut that time. You will need P100 data center card, which was a cool $6,000 back in time.
They even offer multiple VMs sharing a single card, again only for the data center version.
Let's not even get into what they do with low quality silicon (hint: sell as crypto miner cards). IF they would be able to distinguish Machine Learning, I am pretty sure, they would block even that on consumer cards, and sell a different version only for ML.
Overall, nvidia will squeeze the customer to the fullest extent. If history is any indication, ARM under nvidia will most likely see similar schemes (want the latest features? sure price has gone 10x)!
Did NVIDIA teach APPLE or is it the reverse, Apple showed Nvidia how to squeeze the last pennies from the consumer?
Re: (Score:1)
Re: (Score:1)
NVLink is the replacement for old SLI bridges (see RTX 3090). even if you have dedicated skots on the motherboard for nvlink, the card still plugs into a standard pcie,
https://www.servethehome.com/d... [servethehome.com]
AMD will work just fine
Re: (Score:3)
SLI bridges? No. NVLink is a hell of a lot more than that.
https://www.gpumag.com/nvlink-... [gpumag.com]
NVLink is moving in the direction of heterogeneous compute. In fact, NV is adopting CXL to enhance their existing Unified Memory system:
https://developer.nvidia.com/b... [nvidia.com]
Re: (Score:2)
Does it or does it not use SLI bride replacements to communicate with other cards?
If so, then I'm right, and AMD users have nothing to fear from nvlink GPU interconnect!
I don't give a shit what kind of advanced features it supports - as long as nvlink is not replacing the pcie slot for each card, then I'm right, and the author if a fucking FUD-spewing idiot!
Re: (Score:2)
I don't give a shit what kind of advanced features it supports - as long as nvlink is not replacing the pcie slot for each card, then I'm right, and the author if a fucking FUD-spewing idiot!
They're replacing the PCI-e slot. The ARM core will talk to the GPU via NVLink, not PCI-e. Whether or not they will support multi-GPU at all in this configuration is unknown. They've made no mention of it. If they are, it would use NVLink everywhere, both CPU-GPU and GPU-GPU if that's still necessary in that architecture. It may not be.
Re: (Score:3)
It has nothing to do with upgrading of dGPUs. It has everything to do with HOW you will buy nVidia dGPUs in the future. Or how you will buy ARM server hardware. Or produce it for yourself (Amazon/Google/Microsoft).
If NV has their way:
When you want to buy the best GPGPU accelerators on the market with the best GPGPU software stack in the industry, you buy NVidia. Like it or not. Sorry AMD, you have a lot of work to do. So you want to buy NV cards for your datacenter? NV will make sure that you have to
Re: (Score:2)
CUDA = lockin
So no
AMD offers HIP that lets your translate CUDA software over to non-NV accelerators somewhat, but it's not perfect.
So what? (Score:4, Informative)
Arm pioneer Hermann Hauser has once again criticized Nvidia's plan to acquire the semiconductor design company, with The Telegraph reporting Sunday that he believes Nvidia is "clearly showing it will compete unfairly" if the deal is approved. Hauser's concerns reportedly centered on the Grace processor Nvidia announced at GTC 2021. The company's first Arm-based CPU will connect to high-end GPUs via NVLink, which purportedly offers data transfer speeds up to 900 GBps. That's significantly faster than other technologies -- it's also exclusively available to Nvidia.
It's also something NVIDIA can build without acquiring ARM. I'm not sure how that's relevant.
Re:So what? (Score:5, Insightful)
Nothing about this plan is worse than what Apple has been "locking" everyone into for years. Apple's ARM devices, from their phones to their laptops, are festooned with all manner of exclusive, proprietary stuff. The ability to do this is exactly why ARM has succeeded.
And no amount of interweb fanboi mental masturbation will convince me that Apple is in any way morally or ethically superior to Nvidia.
Re: (Score:2)
Confused. Is that a whataboutism?
Re: (Score:2)
Complaining "whataboutism" here, as almost always, is an attempt to distract from a double standard. If what Apple does is so different from what is speculated about Nvidia and NVLink, please point out the relevant difference.
Re: So what? (Score:2)
The difference is that Apple does not control the ARM license and therefore cannot advantage itself relative to other licensees except by better execution. As owner of ARM, Nvidia could be in the invidious position of being able to stall integration of technologies that would benefit licensees in preference to keeping them for itself or by having outsize costs, effectively hobbling their licensees. Unless you are Apple, you probably want a high degree of interoperability with other ARM licensees e.g. if you
Re: (Score:2)
People don't complain about Apple lock-in to ARM cores, but to Apple's walled garden, which Apple does control. For example, their requirements on who can build software that is allowed to run on their devices, their decisions about vertical lock-in efforts, and the like.
Re: (Score:2)
The difference is that Apple does not control the ARM license and therefore cannot advantage itself relative to other licensees except by better execution. As owner of ARM, Nvidia could be in the invidious position of being able to stall integration of technologies that would benefit licensees in preference to keeping them for itself or by having outsize costs, effectively hobbling their licensees.
What precisely are you referring to? You mean they could make their own extensions to the ARM ISA? Because licensees (including nvidia and Apple) can already do that. Doing system level architecure things like having NVLINK as the system bus has nothing to do with the ARM ISA or licensees and they did that as a licensee of ARM and not an owner.
Re: (Score:2)
I am not suggesting that they canâ(TM)t already have their own extensions. I am saying that they can hobble the âoecoreâ ARM just enough to make their offerings better than those of their licensees. Basically, where an independent ARM might have looked to add something like NVLink, NVidia might not want to.
Re: (Score:2)
I am not suggesting that they canâ(TM)t already have their own extensions. I am saying that they can hobble the âoecoreâ ARM just enough to make their offerings better than those of their licensees. Basically, where an independent ARM might have looked to add something like NVLink, NVidia might not want to.
But NVLINK has nothing to do with ISA that ARM licensees actually license, licensees can use NVLINK or PCIe or whatever they like. "Core" arm is the instruction set architecture and anybody can extend that however they like, and they do exactly that.
Re:So what? (Score:4, Interesting)
I also don't see it as much of an issue. Providing direct nvlink integration from the SoC to the GPU only becomes anti-competitive if e.g. NVIDIA refuses to license nvlink to other ARM licensees. The fact they've developed a proprietary interconnect is neither here nor there. Let's also not pretend like they're the only ones doing this: https://www.tomshardware.com/n... [tomshardware.com]
Re: (Score:2)
Re: (Score:2)
The difference is that Nvidia will not be playing on a level field, they will have monopoly control of the upstream platform, which enables all kinds of anticompetitive shenanigans. For example, Nvidia could introduce ARM v9 with their proprietary bus hooks built in, and refuse to license the core without them.
Re: (Score:2)
Re: (Score:3)
Which they don't particularly care about because the name of the game to them is market control. A few defections to RISC-V matters nothing when they can keep their knee on the neck of the entire handset industry.
Re: So what? (Score:2)
Re: So what? (Score:2)
They can deny everyone use of NVLINK regardless.
Re: (Score:2)
Re: So what? (Score:2)
NVIDIA will almost certainly keep selling existing licenses, cores coming with AMBA CHI/ACE to tie your own I/O to.
Re: So what? (Score:2)
But Grace doesn't make a difference, NVIDIA is right about that. It's proof of nothing in the context of future ARM licensing except that some people like grasping at straws.
Well duh (Score:4, Insightful)
Re: (Score:1)
Money can be made on an arms-length basis, without any system integration at all, based on regular sales and dividends.
If it was obvious that Nvidia would take steps that hinder competitors, by way of increased integration or development that reflects their ownership, where is the document describing the ways in which this will be done to hinder competitors?
Re: (Score:2)
IBM made a POWER9 with NVLink [ieee.org] quite a few years ago.
Re: (Score:3)
Right. They want to buy ARM in order to engage in trust-making activity. That's the problem.
First ARM CPU? (Score:3)
I keep seeing this talk of the "first ARM CPU" by Nvidia. Does the Tegra line just not exist anymore? The Nvidia Shield? The Nintendo Switch?
Also, there are already CPUs with nvlink tech. And as other's have pointed out, there is nothing stopping Nvidia from adding nvlink to ARM CPUs without acquiring them, or they could add it to RISC-V as well.
Also, proprietary interconnects have always been common in computing. HyperTransport?
Re: (Score:1)
Tegra uses ARM V8 core, not designed by Nvidia.
Re: (Score:2)
Re: (Score:2)
What? The Tegra CPUs most certainly are designed by NVIDIA, and also have a number of tightly coupled NVIDIA-specific accelerators. I'm unaware of any Tegra that shipped with a standard off-the-shelf ARM core.
Re: (Score:2)
Re: (Score:2)
That is irrelevant to the point being made.
The CPU was designed by nvidia around an existing core.
Therefore this is NOT nvidia's first ARM-based CPU. AT ALL. It's not even their second!
Re: (Score:1)
Those were all system-on-chips. Integrated CPU, GPU, power management and peripherals.
Nvidia is now producing a CPU only product that is designed for server use.
There are mutiple RISC designs available already (Score:2)
Re: (Score:2)
None of those RISC designs enjoy a de facto monopoly in the mobile space.
Re: (Score:2)
Not now, perhaps, but when ARM was still just a toy CPU MIPS pretty much dominated the mobile space. ARM basically only stopped being a joke in the last decade.
Re: There are mutiple RISC designs available alrea (Score:2)
Re: (Score:2)
ARM was faster than MIPS by the late nineties.
ARM has ALWAYS dominated the smartphone market.
The last phones to use a broad mix of processors were flip phones. Motorola was still using PowerPCs in their BREW devices.
MIPS essentially died with SGI, in that it fell behind in performance then, and never recovered.
Re: (Score:2)
MIPS licensing and EDA tools for customization was better than ARM in the late 1990's and early 2000's. It's why we ended up with so many wildly custom MIPS SoCs in our home routers and enterprise routers. The approach that the (slow) CPU runs some management software, and the real fast path is in the silicon is what makes a lot of the Internet possible today.
I had some MIPS based pen computers. It wasn't all that clear that ARM would dominate back in those days, and Microsoft offered Windows CE (and later
Re: (Score:2)
Even Intel designed ARM processors in the past - the StrongARM series.
This isn't really relevant to the rest of your post. Intel had an architectural license (that they sold to Marvell).
They could not compete with ARM, because they, like everyone else who makes ARM processors, was licensing the patents (or the cores themselves) from ARM, who themself does not make processors.
And like Apple, the StrongARMs (and later Marvells) blew everything else out of the water. Eventually everyone else caught up, but for a while, if you wanted real horsepower in an ARM, you wanted a Stro
Re: (Score:2)
Intel retained their architectural license when they sold the StorngARM/XScale business to Marvell because Marvell already had one.
However, both are only for 32-bit ARM so they are not as useful any more.
Re: (Score:2)
Yup. All architectural licenses are limited to the architecture generation you bought the license for.
Even the M1 is built on "old" ARM core licenses now.
Re: (Score:3)
Even Intel designed ARM processors in the past - the StrongARM series.
StrongARM was originally designed by DEC, and ended up with Intel after a lawsuit settlement.
Re: (Score:2)
And Intel further failed to curb its power consumption, which was around an order of magnitude higher than the competitors that came out to defeat it. It was the fastest ARM of its day, but it was the power-hungriest per MIPS.
It looks awfully like Intel is just bad at silicon now
Re: (Score:2)
Poor Amazon (Score:4)
> clearly shows that they will compete unfairly with other Arm-based server companies such as Amazon
I'm shedding crocodile tears that Amazon might get competition. Oh the horrors! Maybe Amazon should stop ripping off open source software vendors and product vendors, and generally, behave less as the 2400lbs gorilla, before "unfair NVDA fights AMZN" manages to evoke sympathy
In a decade ARM will be ... (Score:2)
... a lost and forgotten cpu platform. After nvidia seals the deal they'll block the licensing formula and nobody will develop arm anymore. Either nobody wants to bee dependend on nvidia and their products will fail to render a reasonable profit. So it will bleed to death just like hp's merceded platform.
Re: (Score:2)
they'll block the licensing formula and nobody will develop arm anymore
They can't, many companies already have a perpetual license that can't be revoked.
Re: (Score:3)
That license only exists for a specific architecture (ARMv7, ARMv8, etc.)
And further more, every core you make has to pass a compatibility suite.
So while the license to make your own ARMv8 compatible cores is valid in perpetuity, you will find that the "ARM" ecosystem will leave you behind.
the corporate world (Score:2)
âoeCapitalism is religion. Banks are churches. Bankers are priests. Wealth is heaven. Poverty is hell. Rich people are saints. Poor people are sinners. Commodities are blessings. Money is Godâ â" Miguel D. Lewis
Re: the corporate world (Score:2, Funny)
*crime
Damn touchscreens. Another abortion, brought to you by profit maximization instead of improving the world.
So what if it "was long before"? (Score:1)
Doesn't mean you're not trying to lock your victims in so you can rape their wallets.
What about AMD? (Score:2)
It sure would be a shame if a vendor that owned both the CPU and GPU introduced extensions that gave vendor-specific integration a competitive edge: https://www.tomshardware.com/n... [tomshardware.com]
Re: (Score:2)
Competition is the American Dream. (Score:2)
Not new. (Score:2)
At least he's a consistent nationalist. [bbc.com]
Personally, I suspect China is behind this at least in part - why? Because ARM is one of the few they're legally allowed to work with.
They have a huge incentive to try to stop a sale to the US. Biden has not repealed Trump's trade war stuff.
Don't you mean (Score:2)
Nope (Score:2)
How is this unfair? (Score:2)
A company develops a technology and decides to keep it proprietary. Companies do this all the time. In what sense is this unfair?
It's hard to compete with if you're not Nvidia but that doesn't make it unfair. Companies are under no obligation to make it easier for their competitors.