Moving to Linux Thread

32 GB of VRAM is still larger than 16 GB. That is unless AMD does release a 32 GB VRAM card this year.

1 Like

How high do you estimate the probability of that happening?

Are you planning on needing that much? Because the 7900 XTX has 24 GB of VRAM and the only consumer GPU with more is 5090. And that’s a very expensive GPU especially thanks to scalpers. Are you planning on buying that?

Even the 5080 has only 16 GB, which is pretty much an insult to gamers by Nvidia: https://www.reddit.com/r/buildapc/comments/1hr8sq9/now_everyone_is_mad_at_the_rtx_5080_having_16gb/

3 Likes

I wonder just how higher will these GB counts go over our lifetimes… I can guess into the Terabytes?

Maybeee… very unlikely. I think our current raster-raytrace-ML-mix could only every use maybe 256 GB of VRAM ever, and that’s quite extreme. I think it’s possible we’d invent more VRAM hungry rendering mixes but tbh all of our strategies we’ve ever invented are bandwidth limited, I’d expect a lot more bandwidth increase than raw VRAM boosts. Kinda like with CPUs the original i7 had 8 mbs of cache, a 9800x3d has 96MB of l3 cache + 8 mb of l2 .I don’t know if thats a good predictor logic because cpus are very different but I can imagine some wild memory architectures getting spotlighted well before anyone tries putting a TB in a single GPU. Datacenter AI GPUS…? Yes they’ll do that as soon as OpenAI asks Nvidia to make them one, but currently it’s just not cost effective.

2 Likes

Despite all odds, the AI bubble is yet to have caved in…

Well, it looks like AMD still ended up losing their opportunity to hit back at Nvidia since they also delivered a paper launch. Also, they decided not to make a top notch GPU like the RX 7900 XTX, which is another missed opportunity. What is wrong with the Radeon team?

belgium Radeon vs Chad Nvidia

It looks like v_irgin is censored.

2 Likes

Pretty sure there isn’t much of a reason for it not to be censored in a 13+ evo game forum.

My future workstation/PC:

CPU: AMD Ryzen Threadripper 7960X
SSD: 8 TB M.2 PCIe Gen 5.0 from any company willing to make one
GPU1: AMD Radeon RX 9070 XT (probably from Asus)
GPU2: Nvidia RTX 5090 FE
RAM: 64 GB (4 * 16 GB)
MBD: Any
Case: Lian-Li DK07
PSU: Any Fully Modular

I’m planning to watercool my build, but I’ll start with aircooling to figure out how to setup my dual GPUs on Fedora.

Budget Estimation: 12,000$CDN Maximum

That likely means I’ll have to wait till Christmas to make it, that is if I get a higher salary on a new job.

Also, why didn’t AMD make their own card this time?

1 Like

The filter list contains common insults.

What are you going to use the second GPU for?

I guess to be fair after trying various AI stuff over the past month, I do see the point in like putting together a second Linux box and using it as a home server that’s capable of running AI workloads. The reason why I said second box is that I’m kind of doubtful that it is reasonable to put so much high power GPUs into a single box as the power supply will need to be monstrous, the cooling extremely well done and, well you did spec out a Threadripper but that’s the only way to get full 16x PCI-e lanes for both GPUs.

2 Likes

The second GPU will be for AI workloads or sometimes for a Windows VM when a game doesn’t work with Proton (if the EULA allows it). I’m using PC Partpicker (Canada), which has always reminded me when the PSU didn’t have enough pins or didn’t have the right ones. Also, what do you mean by a box? You mean a second PC?

EDIT: I’ve heard that the RTX 50 series only have access to the open-source kernel module and not to the proprietary driver.

1 Like

Seeing that the PC is planned to be watercooled, I can imagine that it will generate quite the amount of heat for such a solution to be necessary long-term…
Though it’s not as expensive as helium/nitrogen cooling.

You need a case with a ton of airflow, otherwise both GPUs will thermal throttle due to not being able to get rid of the heat that builds up inside the case.

That’s probably untrue as I haven’t seen anything about that.

This is not a thing for computers at all. At least in terms of consumer hardware. So bringing this up is pretty much pointless as it couldn’t happen any case.

2 Likes

Ah not with that attitude. If you have a 12000 budget i’d barely consider that consumer anyways (IK workstations don’t really have anything more than watercooling either and 12000 is perfectly moderate for a workstation but whatever).

1 Like

Do you think watercooling will become more popular in the future? Perhaps even getting a chance to challenge/topple aircooling?

With my intended PC case, that won’t be a problem. Plus my computer is inside my basement, which is always cold.

The RTX 5090 has Blackwell architecture and thus needs the open-source kernel module to function on Linux.

It already challenges air cooling, but it needs more maintenance than air cooling. You need to replace the water coolant every 6-12 months. In terms of popularity, I’m pretty sure it’ll stay the same.

1 Like

I can imagine that requires temporarily deconstructing the entire computer, right?

That depends on the coolant you use. Most water coolants are colored and are either than translucent or solid (opaque). Solid coolants often form small lumps over time and you need to clean your waterblocks if you used a solid coolant. Normally, you need to drain the entire system with the drain valve before you can do just that.

And for the record, I’m planning to use EKWB parts for water cooling because I love supporting companies that don’t pay their employees.

2 Likes

Is solid coolant like ice and stuff?
(Also I hope that second part is satire…)

No, it’s just distilled water with an opaque colorant. Solid is synonym for opaque.

It’s almost satire. I don’t like the fact that employees were unpaid, but EKWB produces the best looking water cooling parts.

1 Like