Nvidia GeForce RTX 50-series product value

Question with the 5000 series not supporting Physx some people may want to purchase a low power / quiet older card to act as a dedicated Physx card, so whats the slowest card you could use for that ?
the slowest new card I can find on sale is
1740215005146.png
 
Question with the 5000 series not supporting Physx some people may want to purchase a low power / quiet older card to act as a dedicated Physx card, so whats the slowest card you could use for that ?
the slowest new card I can find on sale is
View attachment 13189

Ehhm.. it's an interesting question but I'm not sure if this is gonna lead to a satisfactory outcome. Unlike reusing some older card you happen to have lying around.

But ok if you check Nvidia's CUDA compatibility list that model number should have compute capability up to version 2.1 or 3.5 (as there are Fermi and Kepler versions of that card supposedly).
And driver support ends with 475.14.

A Pascal based GT1010 or 1030 shouldn't cost much more but supports all the way to 6.1 as well as still being supported by the latest Nvidia drivers. Might be a more sensible expense.
 
I could use an older card (and I guess it would be the same for a lot of people) but having a quite high end card (for the time) dedicated to just Physx is a bit much
I never thought about the driver side of things would for example having a rtx5070 with the latest driver and a gt710 with PhysX-9.13.0604-SystemSoftware-Legacy driver installed be enough to use that combo ?
or if not a gtx 9000 series card or later and NVIDIA PhysX System Software installed
1740230166690.png
 
Question with the 5000 series not supporting Physx some people may want to purchase a low power / quiet older card to act as a dedicated Physx card, so whats the slowest card you could use for that ?
the slowest new card I can find on sale is
View attachment 13189
You would definitely want a card supported by the same driver as your main card which means that anything older than Maxwell is out and if you plan on using the card for more than just playing the games now then you'd probably make a cut at Turing even - which means something like GTX 1630 or RTX 3050 I'd say.
 

Brilliant job by Tim and goes to show that nvidia literally outdoing native.
One of the best looks at DLSS4 so far and my limited personal testing confirms his findings.

The most noticeable flaw I saw in DLSS4 in Witcher 3 (driver override) was the disocclusion around Geralt. I'm suspect this is less of a problem at super high FPS but with base FPS around 45-60 it's definitely noticeable. Still better about this than any other upscaling solution I tested. Even the old way model was better than any other solution available in that game.

Something funny is that I mostly only see the disocclusion artifacting around Geralt's head, maybe since that's where my eyes rest. If there was increased quality around that small area of the screen, maybe it would have an outsized impact on my perception of the image quality.
 
Looks like Nvidia was caught selling 5090s with missing ROPs. As if the margins aren’t juicy enough, they’re still cutting corners everywhere. Horrible power connector design which represents a serious regression from the 3090ti. Now they’re selling incomplete hardware.

I mean, this is just ridiculous.
 
Would NVIDIA have to build support for 168 ROP 5090 in their driver for the driver to tell GPUZ that it has only 168 ROPs?

Also how did no AIB partners notice this? It's not hard to detect the issue and it clearly impacts performance.
 
Back
Top