Hacker News Reader

Comment by causi | Hacker News Reader

Yeah, with all this "GPU prices are falling woo-hoo" people seem to be forgetting they're still paying above MSRP for cards that are both two years old and were overpriced on launch. I bought an EVGA 1070 SC shortly after launch and paid $400. If Nvidia thinks I'm going to fork over $600 for a two year old 3070 they can sit and spin.

View on HN

Consider AMD.

If you don't need NVIDIA specific features (e.g. ray tracing performance, CUDA), you can get the same level of performance [0] offered by a 3070 for around 350$-450$ in the form of an RX 6700/RX 6750 XT [1][2]. You also get more VRAM with these cards - 12GB instead of the 8GB of the 3070.

[0]: https://www.tomshardware.com/reviews/gpu-hierarchy,4388.html

[1]: https://www.newegg.com/p/pl?d=6700+xt&Order=1&N=4841%2010000...

[2]: https://www.newegg.com/p/pl?d=6750+xt&Order=1&N=100007709&is...

for just gaming AMD is the way to go if you are on a tight budget. they had driver issues which are now fixed so they offer both good performance and price. too bad all ML stuff require CUDA :(

Sadly I do much of my gaming over Moonlight which is Nvidia-exclusive and beats the pants off Steam Link. Also I find the AMD naming conventions somewhat confusing and I can never judge the relative performance of an AMD card by its model number.

I'm a huge Moonlight fan so this is an issue for me as well, but just an FYI, a lot of people have good luck with Sunshine https://github.com/loki-47-6F-64/sunshine - which is a Moonlight compatible host that uses ffmpeg (and works w/ AMD cards).

I'm in the market for an upgrade and will wait until Nov 3 to see what RDNA3 brings to see if it's worth paying for a new card vs last-gen used card pricing.

How does the latency compare to Moonlight? I'm lag-sensitive to the point GeForce Now is immediately nauseating in first-person games.

> Also I find the AMD naming conventions somewhat confusing and I can never judge the relative performance of an AMD card by its model number

It's really quite simple and very similar to NVIDIA's naming scheme.

First digit is the generation (currently 6XXX)

Second digit is the relative performance (64XX < 65XX < 66XX < 67XX < 68XX < 69XX)

"XT" suffix - similar to NVidia's "Super" designation. Non-Zero third digit is like Nvidia's "Ti" designation: 6600 < 6600 XT < 6650 XT.

I've used Steam's remote play/streaming setup for years without issue. What makes Moonlight better?

AMD is also the way to go for non-Windows users. There are hopes that NVIDIA drivers will become usable in wake of the wake of Lapsus$ hijinx, but for now AMD cards work so much better with GNU/Linux

The lapsus hack doesn't matter, no body is going to Read trade secret code.

What actually matters is the announcement of an open source kernel driver, similar to how AMD does it, which will allow for third party drivers like Mesa to exist, instead of manually reverse engineering everything.

https://developer.nvidia.com/blog/nvidia-releases-open-sourc...

The nvidia drivers work great on linux in my experience, so long as you stay away from wayland.

Recently bought a RX 6600 for (€/$ 285), very happy besides not being able to (easily at least) try out stable fusion ;)

You may be able to by building PyTorch with ROCm bindings.

That’s what I meant with "easily" ;) With the effort for it, the 6600 apparently only being borderline able to run it, and my general hate for everything python related, I’d rather not try it.

> Our strategy is to reduce the sell-in - reduce the sell-in this quarter, next quarter to let channel inventory correct. Obviously, we're off the highs and the macro condition turned sharply worse. And so, our first strategy is to reduce sell-in in the next couple of quarters to correct channel inventory. We've also instituted programs to price position our current products to prepare for next-generation products.

-Jensen Huang, Nvidia Q2 Earnings Call to Investors. Paraphrased here: https://nvidianews.nvidia.com/news/nvidia-announces-prelimin...

JayzTwoCents has a video that speculates on what it means: https://www.youtube.com/watch?v=15FX4pez1dw

Short version though, Jay thinks NVidia's trying to reduce the supply of 3xxx cards on the market so that they can hopefully keep prices high-ish and position the 4xxx cards at a higher price bracket then the 3xxx, selling both at the same time.

*eyeroll*

Looks like I will not be buying a new Nvidia anytime soon

I think most people are aware that the prices need to come down another 50% or more until we're approaching "cheap".

MSI's 3090 Ti was hovering around $1099 on Amazon, new-in-box, last week. Nobody should buy a used 3070 for $600 - brand new ones are ~$550 right now: https://www.amazon.com/s?k=rtx+3070

They don't think you're going to buy it, but they know someone is going to.

Yup, iirc i paid $1600 (or maybe $1800? I forget) for a new 3080TI. It hurt. However i'd do it again if/when i need to. Ideally i won't for another few years, but you get me.

Clearly i can afford to do this though. But this isn't just gaming either, i create CG on this machine, digitally sculpt, etc.

I vaguely recall like 10 years you could get a then-decent GPU for like $120. But maybe I'm comparing against, like, the "best-buy special" slightly out of date card or something. I wonder if some of that market has been wiped out by the slow creep of iGPU performance.

Wouldn't we expect a similar price increase over the last 5 years, if only due to inflation and the 25% tariff?

You'd expect that increase to be applied to a significant discount for the card being two years old.

Also the tariff was paused 6 months ago, retroactively refunding back to 11 months ago. It should have minimal effect on current stock.

Nvidia only sets a MSRP not the selling price which the manufacturer sets.

Nvidia does set the price for Founders Edition cards sold under their own name though, although they only account for a small part of the overall card supply.

Nvidia sets the price for all cards that use their chips as part of their OEM agreements. Partners used to get around this by doing things such as adding more memory, OC, and other features which allowed them to charge a premium in those cases. Nvidia has clamped own on this over the years. It's gotten so bad that EVGA famously stopped all GPU development this year in reaction to the declining margins and increasing hostility from Nvidia.

The point is MSRP used to be a relatively safe assumption of selling price (AFAIK it also has been for quite a while), and now it's way lower than what we actually end up paying.

Given the recent evga split, this is actually not the case. Nvidia has contracts to provide them with a great deal of control over end user pricing.

Nvidia literally mandates the pricing their "partners" can sell each card ( it's an interval )

It's widely known Nvidia is a bitch to work with. It has been "fine" because Nvidia was pretty much the only game in town and because of Bitcoin/Covid/floods/locusts/etc the prices have been ridiculous for years and everybody made their sweet money, now things are starting to become sane and the usual crocodiles are starting to cry..

I don't feel pity for anyone, everybody other the the final real consumer made a lot of money.

> I don't feel pity for anyone, everybody other the the final real consumer made a lot of money.

And most of the final real consumers just lost out on inexpensively running Minecraft shaders in 4K or whatever, so eh, life moves on.

I heard in the recent coverage of EVGA ending their partnership with NVIDIA and pulling out of the GPU market that NVIDIA not only sets the MSRP, they also set floor and ceiling for the selling prices.

Also, EVGA CEO claimed that it wouldn't be feasible compete with NVIDIA's pricing for Founder's Edition cards, which are usually cheaper.

That's not true. NVIDIA sets a min and max price range you can sell at (both to board partners AND to retail partners). LTT has talked about it at length in the past, including on last week's WAN Show. It was also cited by EVGA last week as one of the reasons they opted to stop selling GPUs.

Han told the YouTubers that NVIDIA had stopped treating EVGA as a true partner over the years despite their history. https://www.engadget.com/evga-is-leaving-the-gpu-market-and-...

1.NVIDIA had allegedly stopped informing EVGA about new product announcements and important information, such as suggested retail pricing and availability

2.NVIDIA's first-party Founders Edition cards undercut the prices of their counterparts from EVGA and other partners, forcing the vendors to mark down their prices.

Do you have a source?

Sure, NVIDIA sets an MSRP, but that MSRP is a sticker price given to consumers and not representative of what the manufacturers pay for the chips themselves. Manufacturers need to make somewhat of a profit, but NVIDIA has encroached on this profit from its manufacturers since the 10-series came out. JayzTwoCents on YouTube recently discussed this while reporting on a meeting he had with EVGA's CEO about EVGA's recent break with NVIDIA [1].

When the 10-series came about, NVIDIA launched the a limited run of the founder's edition graphics cards at $100 above MSRP. Before the 10-series graphics cards, NVIDIA didn't sell cards themselves and went out of its way to make this seem as if this was a limited edition sort of product. Then the 20-series came out with a continuous run of founder's edition cards which were priced at MSRP, directly competing with their previous board partners.

Tie this along with the previous attempts to strongarm manufacturers via GPP and chip supply allocations, and I tend to give the gift of charity to the manufacturers and not NVIDIA.

[1] https://www.youtube.com/watch?v=12Hcbx33Rb4&t=61s

I love this alt-history, because the contemporary view at the time was that the 10-series MSRP was "fake", that it was impossible to find cards under the $700 MSRP, and that the whole thing was a scheme from NVIDIA to market a lower number than you could pay in practice and upcharge for the blower cooler.

Partner cards were stacking up in the $725-800 range for 6 months after launch, so they came in above the MSRP, which people justified because they were "premium cards". Very few vendors offered any products at the actual MSRP, EVGA had one that was "only" $20 above MSRP and Zotac had something similar and those two low-cost models were impossible to find for a long time.

Now it has gone from "NVIDIA wants to upcharge you for a blower cooler" and it's a dastardly plan from NVIDIA to undercut partners.

everyone is real riled up by that EVGA ceo's tantrum but even GN told you to take it with a grain of salt, and other board partners have confirmed that in the last 2 years they've been running 10x their usual profits due to mining. The EVGA ceo spinning it like partners been losing hundreds of dollars per card for years now is false and untrue. Being a partner is generally around 10% margin and even a smaller partner like EVGA should be able to get around 5%. Losing hundreds of dollars per card is something that happened in like, the last month, partners made very healthy margins selling into the mining boom, and the CEO is being disingenuous by pretending that they've lost a bunch of money on it/are running super low margins. Mining was great for margins.

https://www.igorslab.de/en/evga-pulls-the-plug-with-loud-ban...

Reply on news.ycombinator.com