Question are video card prices headed down yet?

[marees]

Useful video

 

[igor_kavinski]

Welcome back @VirtualLarry !

 

[igor_kavinski]

Useful video

View attachment 126627

View attachment 126631

You can take clearer screenshots in Firefox by just pausing the video and right clicking twice. First right click is Youtube's menu. Second right click gives the Firefox menu with option to save screenshot.

 

[igor_kavinski]

Heads up, folks!

LTT and AMD team up to giveaway TWO 9070 XT Nitro+ GPUs!!!

https://gleam.io/7fzFs/linus-tech-tips-and-amd-gaming-red-team-win-sapphire-nitro-amd-radeon-rx-9070-xt Open to residents of the United States, Canada, Mexico, UK, EU*, Japan, South Korea, Australia, New Zealand, and India. So enjoy participating @Io Magnesso @poke01 @Red Squirrel @WelshBloke...

forums.anandtech.com

 

[Io Magnesso]

To be honest, I don't understand the significance of the existence of RTX5050...

 

[gdansk]

To be honest, I don't understand the significance of the existence of RTX5050...

It is cheaper. Cheaper even than anything modern AMD offers. But it has all those Nvidia features that consumers value at hundreds of dollars elsewhere.

I think that's the entire reasoning behind GB207. Cheap, feature-rich, green sticker. I'm disappointed there is no 75W slot-powered variant.

 

[Heartbreaker]

I think that's the entire reasoning behind GB207. Cheap, feature-rich, green sticker. I'm disappointed there is no 75W slot-powered variant.

It's clocked a little high just to get near 4060 performance, so a 75W version would be significantly behind the 4060, so I guess they could make a 5040 or 5030 out of it, but it might be a hard sell, as they really aren't saving on the BOM, just selling it for less.

 

[Io Magnesso]

It is cheaper. Cheaper even than anything modern AMD offers. But it has all those Nvidia features that consumers value at hundreds of dollars elsewhere.

I think that's the entire reasoning behind GB207. Cheap, feature-rich, green sticker. I'm disappointed there is no 75W slot-powered variant.

Well, cheap is cheap, but that's true
It's disappointing that the power consumption is too much. I think
Are you making a fool of it that it's more than the RTX4060…？(The difference is only 15w, but...)
After all, I want a graphics card that uses the latest architecture that works without auxiliary power...

 

[Io Magnesso]

Well, cheap is cheap, but that's true
It's disappointing that the power consumption is too much. I think
Are you making a fool of it that it's more than the RTX4060…？(The difference is only 15w, but...)
After all, I want a graphics card that uses the latest architecture that works without auxiliary power...

I'd be happy if Intel is the XE3 generation and releases a new low-end model ARC3 Series.

 

[gdansk]

It's clocked a little high just to get near 4060 performance, so a 75W version would be significantly behind the 4060, so I guess they could make a 5040 or 5030 out of it, but it might be a hard sell, as they really aren't saving on the BOM, just selling it for less.

It's understandable that they didn't do what I want. They have so many options with GB207 supporting both GDDR7 and 6. But it seems they plan not to make anything but a slightly cheaper 4060. And I'm not sure the low power variant would have to be any cheaper to sell well. People are mainly buying it for features, not performance.

But ultimately the existence of a $250 GPU from Nvidia this generation is a step towards lower prices. A tiny one but I'll take it.

 

[jpiniero]

The point of GB207 is more for Laptops.

 

[gdansk]

The point of GB207 is more for Laptops.

I suspect that's not relevant to the thread. Any laptops with GB207 video cards? In video cards GB207's point is to be cheap.
Nvidia could have easily not launched it in a video card. But they did and its point is to be cheap.

 

[DAPUNISHER]

https://www.techpowerup.com/review/gigabyte-geforce-rtx-5050-gaming-oc/

$50 is $50 as the girl at the bowling alley used to say. The old decoy effect/asymmetric dominance strategy. Don't want 8GB? Then you have to buy the large popcorn.

Hopefully we get a cut down 5050 6GB 75W version. I really like my 3050 variant I got open boxed from Best Buy for $140+tax. DLSS 4 TM has made using upscaling at 1080p much more palatable.

 

[jpiniero]

Hopefully we get a cut down 5050 6GB 75W version.

Yields too good for that I think. They'll just dump any busted dies on mobile at some point.

 

[igor_kavinski]

Nvidia AI dominance looking to suffer soon:

https://twitter.com/x/status/1945438104745324835

A MUST READ INTERVIEW with a current Director at $MU on HBM & GPU/ASIC market ( $NVDA, $MU, Samsung struggles ):1. The hyperscalers $AMZN, $MSFT, $GOOGL are achieving significantly higher GPU utilization rates than they had in the past years. In 2022 and 2023, the GPU utilization rates were less than 40% The ROI on that in terms of hyperscalers getting their investment back used to be 12-18months. Now, the utilization rates are 70%-80 %, and the ROI has accelerated to only 6-9 months.

2. When $NVDA's Rubin comes out, he expects even higher GPU utilization rates - 90%. With Blackwell next year, he expects the cost of training on hyperscalers to be in the $60-$70 per hour range and for inference, smaller cases to be in the $5-$10 per hour range.

3. The supply of $NVDA's H100 has dramatically increased. The shortage, in his view, has been reduced mostly because of $TSM, which has improved their 4nm yield a lot as the technology has matured and as HBM availability has been good. In his words, when it comes to HBM, we are now moving from »call scarcity« to »high-volume pricing«

.4. HBM has been almost 2X-2.5X year on year. Both SK Hynix as well as $MU first had yield challenges as the biggest problem is TSV technology (through-silicon via), because you have to put eight of these chips on top of each other. If you have to put 12 of these on top of each other the connection between them is the biggest challenge. Both $MU and SK Hynix have become very proficient in that area. He continues to see the pricing going up slightly on the ASP for HBM.

5. He gives the example of HBM use in B100. If you say eight-high stack, that's around $400-$500; if you're using five or six of those, that's around $2500-$ 3600. It is 30% of the GPU cost in most cases.

6. The reason why HBM capacity has caught up in his view is not because of Samsung, but because the TSV technology has matured, and both SK Hynix and $MU have converted most of their CapEx into expanding HBM. He thinks this year might be the last year where we have a scarcity. He thinks 2026 will be even better with availability.

7. The reason why Samsung has struggled with $NVDA's certification process is because of thermal issues. $NVDA B100 and H100 have very tight thermal margins because they're using more than 700W of power. Each signal requires eight Gbps per pin.

8. Samsung has been struggling with the DPPM rate (defect parts per million rate), as it is significantly higher on Samsung parts compared to their competitors. The second thing they struggle with is the eye diagram. The eye diagram is the margin, and according to him, that margin has always been poor compared to SK Hynix and $MU. The reason is that $NVDA uses NVLink. The poor margin also affects the data transfer.

9. According to him, it makes sense that $AVGO approved Samsung as they use less than 300W of power and have more margin, $AVGO uses four Gbps-six Gbps per pin, $NVDA uses twice that, both in power as well as data transfer rate.

10. He thinks that by 2028, $NVDA's market share will be reduced to 50-60% with most of it going to hyperscalers doing their own ASICs. He mentions $GOOGL, noting that their TPUs are at the forefront of this, as they have highly customized FP4 optimization for inferencing and have already deployed TPUs for Gemini, as well as for internal AI training. Second is $AMZN's with Trainium, while both $META and $MSFT are still a bit behind and need to catch up. If he had to make an educated guess, he thinks that $GOOGL and AWS are already serving 50% of their own internal inferencing with their custom ASICs.

Spoiler: FULL INTERVIEW

 

[Hitman928]

Nvidia AI dominance looking to suffer soon:

https://twitter.com/x/status/1945438104745324835

Spoiler: FULL INTERVIEW

View attachment 127207

View attachment 127208

View attachment 127209

View attachment 127210

Seems fake.

 

[igor_kavinski]

Seems fake.

Wouldn't it be lovely if fake AI generated interview/headlines lead to an Nvidia market crash? Nvidia getting a taste of its own medicine.

 

[jpiniero]

Seems fake.

I can buy the idea that hyperscalers using their own AI processors ends up being a large portion of the volume. That's kind of what happened with CPU servers.

 

[LOUISSSSS]

Hi testing 123

 

[igor_kavinski]

Hi testing 123

Test failed. $$$ required to pass.

 

[Golgatha]

Kinda disconcerting that prices at my local Micro Center continue to be well above MSRP while there's plenty of stock just sitting on the shelf. I guess they and/or video card manufacturers have finally just decided to put them out there, and then slowly sell while bringing the prices down even more slowly. Cheapest 5090 is still $2400 + tax.

 

[marees]

Kinda disconcerting that prices at my local Micro Center continue to be well above MSRP while there's plenty of stock just sitting on the shelf. I guess they and/or video card manufacturers have finally just decided to put them out there, and then slowly sell while bringing the prices down even more slowly. Cheapest 5090 is still $2400 + tax.

The past trend has been to wait 1 year for the super versions to release (this year end)

Even better wait for a console to release. That rubs Jensen's ego the wrong way
So by next year end the xbox will have a gpu as powerful as the 5080 / 5080 super

 

[blckgrffn]

Kinda disconcerting that prices at my local Micro Center continue to be well above MSRP while there's plenty of stock just sitting on the shelf. I guess they and/or video card manufacturers have finally just decided to put them out there, and then slowly sell while bringing the prices down even more slowly. Cheapest 5090 is still $2400 + tax.

As an importer, sourcing uncertainty is a big deal right now. New crap (duties, international turmoil, etc.) could drop tomorrow that messes up everything, and then you wouldn't have wanted to sell through that stuff.

Also, new product might cost a lot more due to tariffs than existing stock but you still have to get the margin today to pay for tomorrows resupply, so being conservative, if you can, might not be the worst strategy.

And... consumer purchasing power seems to be not great atm. This is always a slow month for my industry but it is dragging worse than ever right now.

Just my observations and if I was running a MC I'd be in no hurry right now so long as I was still doing OK cash flow wise.

 

[dlerious]

Probably won't last long, but PNY 5070Ti for $750

Amazon.de

 

[VirtualLarry]

For comparison, the venerable 1080P-class GTX 1660 Super was $260-280 upon release.

I bough a few Ti models for my mining rigs before prices went up, for $230 factory-refurb at Newegg.