Addition to the very first/my comment:

• Using a clamshell VRAM design (It simply means doubling the 4 2GB GDDR6 chips also to the other side of the PCB and, as such, doubling the 8 GB VRAM to 16 GB VRAM. This would probably be the cheapest solution and maybe even the best solution [for the price] as the VRAM would also be more than 12 GB VRAM, which definitely should be enough for this kind of GPU performance and it would be the same as PS5, not that is has to have the same amount as a PS5, just more than 8 GB VRAM ;-). 12 GB VRAM would be fine, but the way it works out (see first comment), is that a clamshell VRAM design, again, would probably be the cheapest and best solution. And maybe VALVE gonna release a Steam Machine with such a clamshell VRAM design).

QuoteProvide an example of a mobile non Nvidia off the shelf dgpu part that comes with more than 8 GB of VRAM? Has to be reasonably priced, so no halo's involved.

It has nothing to do if the GPU chip is a mobile, or a desktop, one, it's the same chip off the silicon waver, maybe the mobile ones are better binned, aside being driven with a lower Wattage, and it's obviously not spam either: In YouTube, or in a search engine of your choice, just type: "are 8gb ram enough in 2025 for AAA games?" Obviously 8GB VRAM are enough for older games.

Ok, now let me give you and everyone else an example: The idea is simply using a clamshell VRAM design (I didn't think of it in my first post where I gave the 4 examples to increase the VRAM above 8 GB): The cheapest more-than-8GB-GB-VRAM GPU, according to en.wikipedia.org/wiki/Radeon_RX_7000_series, would be a 7600 XT in the clamshell VRAM configuration (= 16 GB VRAM).

But even better would be, if VALVE used RDNA4 9060 XT (or 9060 non-XT, if AMD would allow VALVE to use a clam-shell VRAM design). Then the official ML upscaling would also be supported (official support required RDNA4, unofficial INT8 support on RDNA3 (and RDNA2?) exists, but the quality and performance aren't the same.

It may be that any RDNA3 / RDNA4 128-bit chip could be used in a clamshell design, if allowed by AMD, and it 16 GB of VRAM. (same applies for the other 2 GPU makers)

Tada!

Quote from: Only 8GB VRAM big mistake on November 24, 2025, 09:39:49It has nothing to do with high-end,

You're still not addressing the core issue.

Provide an example of a mobile non Nvidia off the shelf dgpu part that comes with more than 8 GB of VRAM? Has to be reasonably priced, so no halo's involved.

I'll wait.

There's the 7800M/7900M I guess but that's a 200W part and they would likely have trouble cooling that in such a small form factor. It's also in much less supply and expensive (aka read as: 'high end').

You can complain all you want about it but it still doesn't change the facts. There are no real alternatives for the price bracket and SFF they are marketing to.

Quote from: Only 8GB VRAM big mistake on November 24, 2025, 09:39:49It has nothing to do with high-end, with only 8 GB VRAM, more and more games become a stutter show, even at low settings: youtube.com/watch?v=ric7yb1VaoA ("Gaming Laptops are in Trouble - VRAM Testing w/ ‪@Hardwareunboxed‬")

1/10 ragebait.

It has nothing to do with high-end, with only 8 GB VRAM, more and more games become a stutter show, even at low settings: youtube.com/watch?v=ric7yb1VaoA ("Gaming Laptops are in Trouble - VRAM Testing w/ ‪@Hardwareunboxed‬")

Quote from: Only 8 GB VRAM on November 17, 2025, 09:29:42would make it more expensive

Well, you should know by now that valve aren't aiming for high end yet and trying to keep costs down.

So case closed, I guess?

Quoteand 8 GB of GDDR6 VRAM

After years of proofs that 8 GB VRAM are not enough anymore, and things getting worse as games demand more VRAM (or hosting your favorite local AI / LLM), VALVE comes out with 8 GB VRAM..

Sorry VALVE, but to be able to actually call it custom silicon, you could have at the very least tried to make it at least 10 GB VRAM.

Here is how (and each solution would make it more expensive, but 8GB VRAM is just a no-go) (current hw: 128-bit memory bus width GPU chip / 32-bit per GDDR5 chip = 4 chips, 2 GB each):

• Using higher GDDR6 density: 2.25 GB per chip * 4 chips = 10 GB VRAM (uncommon density)

or

• Using higher GDDR7 density: 2.25 GB per chip * 4 chips = 10 GB VRAM (since GDDR7 is the successor, maybe the memory manufacturers would be "kind" enough to make such densities)

or

• Using higher GDDR7 density: 3 GB per chip * 4 chips = 12 GB VRAM (3 GB chips are already being used in the 5090 Laptop, see en.wikipedia.org/wiki/GeForce_RTX_50_series#Mobile) (and will be used in the SUPER refresh GPUs in 2026)

or

• Using a wider memory bus width: 32-bit per chip * 5 chips, 2 GB each = 160-bit memory bus width GPU chip (this would be a different class GPU, one that would be approx. 33% faster (1.33 = 160-bit/128-bit) than the current one