Hacker News Reader

Comment by ks2048 | Hacker News Reader

ks2048

This is maybe a dumb question, but why is it so hard to buy Nvidia GPUs?

I can understand lack of supply, but why can't I go on nvidia.com and buy something the same way I go on apple.com and buy hardware?

I'm looking for GPUs and navigating all these different resellers with wildly different prices and confusing names (on top of the already confusing set of available cards).

View on HN

OK so there are a handful of effects at work at the same time.

1. Many people knew the new series of nvidia cards was about to be announced, and nobody wanted to get stuck with a big stock of previous-generation cards. So most reputable retailers are just sold out.

2. With lots of places sold out, some scalpers have realised they can charge big markups. Places like Amazon and Ebay don't mind if marketplace sellers charge $3000 for a $1500-list-price GPU.

3. For various reasons, although nvidia makes and sells some "founder edition" the vast majority of cards are made by other companies. Sometimes they'll do 'added value' things like adding RGB LEDs and factory overclocking, leading to a 10% price spread for cards with the same chip.

4. nvidia's product lineup is just very confusing. Several product lines (consumer, workstation, data centre) times several product generations (Turing, Ampere, Ada Lovelace) times several vram/performance mixes (24GB, 16GB, 12GB, 8GB) plus variants (Super, Ti) times desktop and laptop versions. That's a lot of different models!

nvidia also don't particularly want it to be easy for you to compare performance across product classes or generations. Workstation and server cards don't even have a list price, you can only get them by buying a workstation or server from an approved vendor.

Also nvidia don't tend to update their marketing material when products are surpassed, so if you look up their flagship from three generations ago it'll still say it offers unsurpassed performance for the most demanding, cutting-edge applications.

The workstation cards have MSRPs. The RTX 6000 Ada’s MSRP is $6799:

https://www.techpowerup.com/gpu-specs/rtx-6000-ada-generatio...

Nvidia (and AMD) make the "core", but they don't make a "full" graphics card. Or at least they don't mass produce them, I think Nvidia tried it with their "founders edition".

It's just not their main business model, it's been that way for many many years at this point. I'm guessing business people have decided that it's not worth it.

Saying that they are "resellers" isn't technically accurate. The 5080 you buy from ASUS will be different than the one you buy from MSI.

Nvidia also doesn't make the "core" (i.e. the actual chip). TSMC and Samsung make those. Nvidia designs the chip and (usually) creates a reference PCB to show how to make an actual working GPU using that chip you got from e.g. TSMC. Sometimes (especially in more recent years) they also sell that design as "founders" edition. But they don't sell most of their hardware directly to average consumers. Of course they also provide drivers to interface with their chips and tons of libraries for parallel computing that makes the most of their design.

Most people don't realize that Nvidia is much more of a software company than a hardware company. CUDA in particular is like 90% of the reason why they are where they are while AMD and Intel struggle to keep up.

It seems that they have been tightening what they allow their partners to do, which caused EVGA to break away as they were not allowed to deviate too much from the reference design.

That was mostly about Nvidia's pricing. It's basically impossible to compete economically with the founders editions because Nvidia doesn't charge themselves a hefty markup on the chip. That's why their own cards always sell out instantly and then the aftermarket GPU builders can fight to pick up the scraps. The whole idea of the founders edition seems to be to make a quick buck immediately after release. Long term it's much more profitable to sell the chip itself at a price that they would usually sell their entire GPU for.

This years founders edition is what I really want from a GPU. Stop wasting my 2nd PCIe slot because you've made it 3.5/4 slots BIG! It is insane that they are now cooling 575W with two slots in height.

I would suggest getting a case that has a set of inbuilt (typically vertically-oriented) expansion card slots positioned a distance away from the regular expansion card slots, mount your graphics card there, and connect it to the motherboard with a PCI-E riser cable. It's what I did and I kicked myself for not doing it years prior.

I have no experience with PCI-E 5 cables, but I've a PCI-E 4 riser cable from Athena Power that works just fine (and that you can buy right now on Newegg). It doesn't have any special locking mechanism, so I was concerned that it would work its way off of the card or out of the mobo slot... but it has been in place for years now with no problem.

Can you link to an example case and riser cable?

I shouldn't have to link to the cable given that I said "Athena Power" and "Newegg", but sure, here you go. [0] Their Newegg store is here. [1] (They also sell that cable in different lengths.)

The "away from motherboard expansion card slots feature" isn't particularly uncommon on cases. One case that came up with a quick look around is the Phanteks Enthoo Pro 2. [2] I've seen other case manufacturers include this feature, but couldn't be arsed to spend more than a couple of minutes looking around to find more than one example to link to.

Also, there are a few smaller companies out there that make adapters [3] that will screw into a 140mm fan mounting hole and serve as an "away from motherboard" mounting bracket. You would need to remove any grilles from the mounting hole to make use of this for a graphics card.

[0] https://www.newegg.com/athena-power-8-extension-cable-black/...

[1] https://www.newegg.com/Athena-Power/BrandStore/ID-1849

[2] https://phanteks.com/product/enthoo-pro-2-tg

[3] Really, they're usually just machined metal rectangular donuts... calling them "adapters" makes them sound fancier than they are.

Man, things are getting really large and unwieldy with these giant GPUs we have nowadays.

My theory is this is one of the ways nvidia is trying to force ML users to buy the $$$$$ workstation cards.

Can't put four 4090s into your PC if every 4090 is 3.5 slots!

You can do single slot 4090 cards using water cooling, so having enormous coolers is not forcing anyone to buy workstation cards to fit things. Alternatively, there are always cases designed for riser cables.

It is an ever uphill battle to compete with Nvidia as a AIB partner.

Nvidia has internal access to the new card way ahead of time, has aerodynamic and thermodynamic simulators, custom engineered boards full of sensors, plus a team of very talented and well paid engineers for months in order to optimize cooler design.

Meanwhile AIB partners is pretty much kept in the blind until a few months in advance. It is basically impossible for a company like EVGA to exist as they pride themselves in their customer support - the finances just does not make sense.

Which is why EVGA stopped working with Nvidia a few years ago... (probably mentioned elsewhere too).

https://www.electronicdesign.com/technologies/embedded/artic...

Yeah I should have said design, embarrassingly I used to work in a (fabless) semiconductor company.

Totally agree with the software part. AMD usually designs something in the same ball park as Nvidia, and usually has a better price:performance ratio at many price points. But the software is just too far behind.

AMDs driver software is more featureful and better than NVidia's offerings. GeForce Experience + the settings app combo was awful, the Nvidia App is just copying some homework, and integrating MSI Afterburner's freeware.

But the business software stack was, yes, best in class. But it's not so for the consumer!

I think they mean CUDA

I've bought multiple founders editions cards from the nvidia store directly. Did they stop doing that recently?

They still make reference founders editions. They sell them at Best Buy though, not directly.

Reference cards make up the vast minority of cards for a specific generation though. I looked for numbers and could not find them but they tend to be the Goldilocks of cards if you can grab one because they sell at msrp IIRC.

Yep, I scored a 3070 Founder's at launch and was very lucky, watching other people pay up to the MSRP of the 3090 to get one from elsewhere.

Didn't Nvidia piss of some of their board partners at some point. I think EVGA stopped making Nvidia based graphics cards because of poor behavior on Nvidia part?

Also aren't most of the business cards made by Nvidia directly... or at least Nvidia branded?

  it's not worth it.

I wonder how much "it's not worth it". Surely it should have been at all profitable? (a honest question)

The founders edition ones that I had were not great gpus. They were both under cooled and over cooled. They had one squirrel cage style blower that was quite loud and powerful and ran bascially at no speed or full blast. But being that it only had the one airpath and one fan it got overwhelmed by dust or if that blower fan had issues the gpu over heated. The consumer / 3rd party made ones usually have multiple fans at lower speeds larger diameter, multiple flow paths, and more control. TL;DR they were better designed, nvidia took the data center ram as much air as you can in there approach which isn't great for your home pc.

Founders cards being worse than board partner models hasn't been true in like 8 years. They switched to dual axial rather than a single blower fan with the 20 series, which made the value of board partner models hard to justify.

Since then, Nvidia is locked in a very strange card war with their board partners, because Nvidia has all the juicy inside details on their own chips which they can just not give the same treatment to their partners, stacking the deck for themselves.

Also, the reason why blowers are bad is because the design can't really take advantage of a whole lot of surface area offered by the fins. There's often zero heat pipes spreading the heat evenly in all directions, allowing a hot spot to form.

good to know, I have a 980gtx, I had to rma it after a summer of overheating. good to know they've gotten better on their own cards.

This is supply and demand at work. NVIDIA has to choose to either sell consumer or high end and they can reserve so much resources from TSMC. Also, Apple has outsold hardware before or it has high demand when it releases but for NVIDIA they have nearly constant purchases throughout the year from enterprise and also during consumer product launches.

It is frustrating speaking as someone who grew up poor and couldn't afford anything, and now I finally can and nothing is ever in stock. Such a funny twist of events, but also makes me sad.

Imagine how sad you'd be if you were still poor.

If you think it is bad for Nvidia, give AMD a try. Go ahead and try to guess which GPU is the most powerful by model number. They give so many old parts new model numbers, or have old flagship parts they don't upgrade in the next generation that are still more powerful.

GPUs are in demand.

So scalpers want to make a buck on that.

All there is to it. Whenever demand surpasses supply, someone will try to make money off that difference. Unfortunately for consumers, that means scalpers use bots to clean out retail stores, and then flip them to consumers.

Without thinking about it too deeply I'm wondering if GPU demand is that much higher than let's say iPhone demand. I don't think I've ever heard of iPhones being scarce and rare and out of stock.

Apple very tightly controls their whole value chain. It's their whole thing. Nvidia "dgaf" they are raking in more cash than ever and they are busy trying to figure out what's at the end of the semi-rainbow. (Apparently it's a B2C AI box gimmick.)

I read your question and thought to myself "why is it so hard to buy a Steamdeck"? Available only in like 10 countries. Seems like the opposite problem, Valve doesn't use resellers but they can't handle international manufacturing/shipping themselves? At least I can get a Nvidia GPU anytime I want from Amazon, BestBuy or whatever.

> At least I can get a Nvidia GPU anytime I want from Amazon, BestBuy or whatever.

You can? Thought this thread was about how they're sold out everywhere.

Maybe, it is simply a legacy business model. Nvidia wasn't always a behemoth. In olden days they must be happy for someone else to manage the global distribution, marketing, service etc. Also, this gives an illusion of choice. You get graphic cards in different color, shape, RGB, water cooling combinations.

One way to look at is that the third party GPU packagers have a different set of expertise. They generally build motherboards, GPU holder boards, RAM, and often monitors and mice as well. All of these product PCBs are cheaply made and don't depend on the performance of the latest TSMC node the way the GPU chips do, more about ticking feature boxes at the lowest cost.

So nvidia wouldn't have the connections or skillset to do budget manufacturing of low-cost holder boards the way ASUS or EVGA does. Plus with so many competitors angling to use the same nvidia GPU chips, nvidia collects all the margin regardless.

Yet the FE versions end up cheaper than third party cards (at least by MSRP), and with fewer issues caused by the third parties cheaping out on engineering…

I've always assumed their add-in board (AIB) partners (like MSI, ASUS, Gigabyte, etc) are able to produce PCBs and other components at higher volumes and lower costs than NVIDIA.

Not just the production of the finished boards, but also marketing, distribution to vendors and support/RMA for defective products.

There is profit in this, but it’s also a whole set of skills that doesn’t really make sense for Nvidia.

It depends on the timing. I lucked out about a year ago on the 4080; I happened to be shopping in what turned out to be the ~1 month long window where you could just go to the nvidia site, and order one.

Nvidia uses resellers as distributors. Helps build out a locked in ecosystem.

How does that help "build out a locked in ecosystem"? Again, comparing to Apple: they have a very locked-in ecosystem.

I don't think lock-in is the reason. The reason is more that companies like Asus and MSI have a global presence and their products are available on store shelves everywhere. NVIDIA avoids having to deal with building up all the required relationships and distribution, they also save on things like technical support staff and dealing with warranty claims directly with customers across the globe. The handful of people who get an FE card aside.

Nvidia probably could sell cards directly now, given the strength of their reputation (and the reality backing it up) for graphics, crypto, and AI. However, they grew up as a company that sold through manufacturing and channel partners and that's pretty deeply engrained in their culture. Apple is unusually obsessed with integration, most companies are more like Nvidia.

Apple locks users in with software/services. nVidia locks in add-in board manufacturers with exclusive arrangements and partner programs that tie access to chips to contracts that prioritize nVidia. It happens upstream of the consumer. It's always a matter of degree with this stuff as to where it becomes anti-trust, but in this case it's overt enough for governments to take notice.

Reply on news.ycombinator.com