News Latest Nvidia drivers boost synthetics by 6-8% on RTX 50 GPUs — Users still report stability woes

[Admin]

Nvidia's 576.02 drivers are said to increase (synthetic) performance by up to 8%, but user reports suggest stability issues still persist.

Latest Nvidia drivers boost synthetics by 6-8% on RTX 50 GPUs — Users still report stability woes : Read more

 

[King_V]

Adding to the mix is increased synthetic performance... Alas, it's not all sunshine and rainbows, as it appears a new set of stability woes has surfaced with these new drivers, as indicated by user reports.

Increased performance in NON-real-world tests, at the cost of new stability problems? Where do I sign up? /s (if it isn't obvious)

Seriously, what's going on over there at Nvidia?

 

[Thunder64]

Wait I thought everyone always said AMD had the crap drivers? Oh well at least now I can play 3DMark faster!

 

[abufrejoval]

Increased performance in NON-real-world tests, at the cost of new stability problems? Where do I sign up? /s (if it isn't obvious)

Seriously, what's going on over there at Nvidia?

With me, the symptoms were very clear: All "Blackwell-ready" drivers cause massive issues with anything older than Blackwell, while I actually didn't have a lot of trouble with those newer drivers on the single Blackwell generation card I bought.

My main mistake was to roll out the newest drives across all machines, which cover quite a few generations right back to one GTX 980ti still used by an in-law.

I'd say there has been a major amount of refactoring in the driver code and any backward compatibility testing which involved manual actions (e.g. testing with various monitors, resolutions, KVM etc.) that can't be automated, was at best done rudimentary and with a focus on the latest hardware.

Apart from the huge variety of GPU ASICs Nvidia is still supporting, there has been an major multiplication of features and interface capabilites, from what display ports support in protocols and bandwidths to PCIe v5.

You can't just hire GPU driver programmers off the street for a few weeks to prepare for a launch, while you also can't start that early writing code when you're still far away from working silicon (in sufficient numbers).

I'd just say that the complexity of the driver code is almost better at matching Moore's curves than the chips they drive and money alone isn't enough to fix that.

Once I knew I only had to go back to the old drivers for the older hardware, that was quite ok with me. Pretty sure they'll fix it eventually.

It's only the weeks that I spent thinking that something was wrong with my KVMs or that this was due to OS migration issues which still irk me. That's why on the job, we always test ourselves and always partition the deployment of new versions.

 

[Pemalite]

The irony is... AMD was always tarnished with bad drivers.

Should be interesting to see what kind of impact this has on AMD's marketshare in the short term, but I doubt not much.

 

[Ryrynz]

The irony is... AMD was always tarnished with bad drivers.

Should be interesting to see what kind of impact this has on AMD's marketshare in the short term, but I doubt not much.

A lot of bad press recently for Nvidia, AMD are doing well, the GRE will further cement inroads.

 

[Thunder64]

The irony is... AMD was always tarnished with bad drivers.

Should be interesting to see what kind of impact this has on AMD's marketshare in the short term, but I doubt not much.

The thing is AMD hasn't had bad drivers in a long time. It was true at one point but just got repeated over and over as if it was still true.