Finally "upgraded" GPU

[cardboard_killer]

I finally felt like it was time to step up my old, old, old RX 580 (8 gig) card. I've intended to now for two years, but the prices have been insane and there was always "the next generation" about to come out. Sigh.

 

Don't laugh, but I went cheap and got an RX 6650xt. A two year old card, but it will do for now. I almost got caught again in the "but the RXT 4060 Ti is about to be released!" trap.

[Voyager]

Definitely give them both a go in the Syn_Vander benchmark. We've got a lot of nVidia hardware but AMD stuff is much rarer and we've still got tons of questions to answer there. 

[AngleOff66]

After the reviews of the  4060 on yt, I wonder what Jenson is trying to pull?  Drop the gpu market to concentrate on cards that run AI?

 

[Voyager]

20 hours ago, AngleOff66 said:

After the reviews of the  4060 on yt, I wonder what Jenson is trying to pull?  Drop the gpu market to concentrate on cards that run AI?

 

 

Yes. That is exactly their strategy. 

 

NVidia is extremely good at monolithic GPUs, but AMD has figured out how to build competitive GPUs on far cheaper nodes and processes. 

 

In 3nm, nVidia is expected to max out the TSMC reticle limit for their 5090. It will be very high performance and very good energy efficiency.

 

Meanwhile AMD is subdividing their GPUs into things they can crank out the way they're cranking out Zen GPUs, so when TSMC halves the reticle limit on 2nm, nVidia won't be able to build a 6090 on the scale of the 5090, but AMD will be able to build that class of card on multi chiplette designs for less BOM cost. 

 

NVidia will still likely have significantly higher AI performance per watt, which means in power constrained AI applications, they will still be able to charge a big premium for their products. 

 

It is a risk, but it plays to nVidia's strengths so it could work. Or it could blow up in their faces. It's too soon to tell. 

[AngleOff66]

I just read that Nvidia stock price had a 26% spike based on demand for AI chips.

 

So should game devs start looking at optimizing for AMD? Or Intel(assuming the gpu line up survives)? 

 

Perhaps some GPU company in the far east will see an opening.

 

No, not being serious.

[Sokol1]

I remember at some years people complaining of AMD VGA/Drivers bad performance in IL-2 Sturmovik: Great Battles, this still a issue?

 

[AngleOff66]

Never mind misread. More coffee.

Edited June 2, 2023 by AngleOff66

[LuftManu]

On 6/1/2023 at 6:06 PM, Sokol1 said:

I remember at some years people complaining of AMD VGA/Drivers bad performance in IL-2 Sturmovik: Great Battles, this still a issue?

 

Hello Sokol1,

As an owner of NVIDIA and AMD cards in the last 2 years I can assure you the 6XXX gen of AMD performs a little bit subpar in comparision with Nvidia 3000 Gen counterparts.

Not taking into account VR performance, they should have a little tiny bit more performance.

I.e. If the 3070 and a 6700XT are 5% apart (in general) in favour of the 3070, in Il-2 it might be a 7-8%.

 

You can see some examples in the community benchmark.

 

This was checked during 2022. Don't know how it performs now or how the new 7000 series perform.

 

Kind regards,

Edited June 2, 2023 by LuftManu

[chiliwili69]

On 6/2/2023 at 6:13 PM, LuftManu said:

You can see some examples in the community benchmark.

 

Looking at the SYN_Vander GPU test in 4K (where only GPU should become bottlenecked) there are a few tests of the AMD cards:

 

 

In theory both 79000XT and 6900XT are superior to a 3080 (according to GPU public passmark), but in IL-2 the 3080 is giving better fps.

 

for 7900XT: (123-92)/92= 33% more by the 3080

 

for 6900XT: (123-89)/89= 38% more by the 3080

 

So unfortunately the AMD still don´t give their full performance in IL-2. We don´t know why. This has been reported to IL-2 team several times.

 

If you have both cards you can make your own tests witht he SYN_Vander test.

 

 

Edited June 4, 2023 by chiliwili69