Is VR dead?

[Deleted member 197115]

Yeah right, gpu pricing the last couple of years have been completely normal, nothing to see here.

GPUs when in stock were sold at MSRP, scalpers is completely different issue, as demand for current gen cards went down and market saturated, there is little incentive to hoard them for resale purpose. That doesn't mean we won't see the same situation with 4xxx series when it gets out, may be even worse.

 

[BrunoBæ]

Generally its good to be optimistic - but I guess some of you will get disappointed.
First when you realise that the GPU prices probably will stay at a too high level for years to come.
And second when future benchmark tests of a comming 4000 series GPUs does show that they are only slightly faster than the 3000 series.
I hope Im wrong - but I fear Im not

 

[RCHeliguy]

And second when future benchmark tests of a comming 4000 series GPUs does show that they are only slightly faster than the 3000 series.
I hope Im wrong - but I fear Im not

Nooooooooooooo!!!!!!!!!!!

 

[blekenbleu]

Nooooooooooooo!!!!!!!!!!!

View attachment 558840

There have been multiple hints that a top model or so will deliver
substantially improved performance with equivalently increased power demands,
but e.g. x060 model may get the usual bump, roughly equivalent to current 3070.
One can easily imagine inflated premium pricing for those halo models.

 

[BrunoBæ]

There have been multiple hints that a top model or so will deliver
substantially improved performance with equivalently increased power demands,

I will not try to prove anything here that cannot be proved - because none of us does actually know if this substantially hype is anything else than yes hype - but..
But when I hear beforehand that a comming GPU need substantially more juice to get a useable output then I grab my wallet - hard.
Because then I instantly get a suspicion that this new GPU is just an old architecture (3000ish) which is extremely overclocked from the manufacturer.
Because if the architecture really was next gen then the power demands would not be astronomical.
OK?

 

[blekenbleu]

extremely overclocked

I agree that lots of power suggests pushing an older technology more than improving fundamentals.
In this case, Nvidia seems to be going with more parallel processing units.

 

[BrunoBæ]

Be aware that I would have been much more optimistic if the hype and/or believable rumors had told that the comming 4000 series would be based on (at least partly) new architecture.
And further parallel processing units isnt exactly this.

 

[RCHeliguy]

If I need to buy a new 1200-1600W PS with ATX 3.0 handling 600W with peaks of up to twice that, I BETTER GET SOME DAMN PERFORMANCE!

I will be buying a minimum of a 4080/5080 however they name it.

We better get that doubling or better of performance!

Keep in mind that they are still dropping from 10nm to 7nm process, so it should run more efficiently per cuda core than before and if it needs that much power with a smaller process, it had better be a dragon slayer!

 

[wajdi]

Wake me up when the 4080ti is out!

 

[Bob73]

Forgive me for bring up what is possibly an old point… but why hasn’t the idea of using two video cards been revisited? Or is it simply the games don’t support that particular route?

 

[RCHeliguy]

For VR SLI has been very problematic and/or unstable. I believe that the software has to support multiple cards as well. At one point I think Pimax was considering having one GPU per eye but they abandoned that idea.

 

[Bjarne Hansen]

The idea has come up but you see the issue is if you could run 2X GTX 970 in Sli and get better fps than a RTX3090 who would then buy the top end GPU's?

Sli and X-fire used to work fairly good but it wasn't in financial interest of the CPU makers that it worked.
I am sure there are also some technical issues that was used as excuse but.
Anyway SLI is not working in VR.

 

[RCHeliguy]

Good point!
If people will more than double what they spend to get an extra 15% performance, why let them double the performance of a lessor card.

That said there are some technical challenges to keep two cards synchronized and to split the work between them. When the 20 series was released the Nvlink it was touted as something more than it was. NVidia came out with a way for their high end professional Quadro cards to look like one very large card. It took a very high speed connection between the cards to accomplish this. Nvlink for the RTX cards did not come close to what the professional cards were doing.

However here was a test of just this. None of these were being run in VR.
He says that multi-card gaming is dead after showing some games actually saw more than a 90% improvement in speed with SLI.

Of course with these cards being 3 slots wide and most computers only having one fast slot to put them in, it would require some kind of frankenmonster computer to handle this.

 

[RCHeliguy]

This is interesting, but it says that there are very few VR titles that actually support this.

 

[anton_Chez]

You fail to recognise that people will also buy two of the HIGHEST tier card for unparalleled performance. The death of multi card configs is sad. I was a user of both Crossfire and SLI in the past. I did have 2 x GTX 970's and NOTHING was faster than what I was running. Except for..... 2 x GTX 980's, of course. Which I'm sure people were also running. It worked fine for me for as long as I had the setup and I am extremely disappointed that it's just faded into obscurity over the past few generations. I mean, Nvidia made ONLY the top 2 tier cards SLI compatible, not long ago. So if profit is their concern, which lets face it, of course it's their ONLY concern, then making people buy the top cards to enable SLI would have solved that problem easily.

 

[Bob73]

Sounds like a clever marketing plan….. limit the options so the only choice is to spend spend in the hope limited gains keep you hanging in for the next update

At the moment I am building a DIY rig and VR has always been the route I wanted to take for immersion and also the smaller footprint over triples.

 

[BrunoBæ]

when I hear beforehand that a comming GPU need substantially more juice to get a useable output then I grab my wallet - hard.
Because then I instantly get a suspicion that this new GPU is just an old architecture (3000ish) which is extremely overclocked from the manufacturer.

Maybe there is still hope.
Because I wonder if the same upping of efficiency for the comming power hog in the 4xxx series can be made.
Some kind of trade of by given up on the last fps to get a manageable power consumtion.

NVIDIA GeForce RTX 3090 Ti Graphics Card Turns Into A High-End Efficiency King, Beats AMD 6900 XT In Games at 300W

https://wccftech.com/nvidia-geforce-rtx-3090-ti-graphics-card-turns-into-a-high-end-efficiency-king-beats-amd-6900-xt-at-300w/

 

[blekenbleu]

Intel’s new Arc GPUs will reportedly use TSMC’s 3nm and 2nm process nodes
as they attempt to take on market leaders Nvidia and AMD.
TSMC is to start high-volume production of 2-nanometer technology starting in late 2025.
https://gizmodo.com/apple-intel-in-line-for-tsmc-2nm-chips-1848830019

 

[RCHeliguy]

On that note, Nvidia will be going to 4nm for the next GPU's rather than 5nm.

 

[BrunoBæ]

On that note, Nvidia will be going to 4nm for the next GPU's rather than 5nm.

If that is correct (guess it is) then it means a lot more than just a number.
Mainly because smaller nodes/finer resolution means (generally) better/lower power use.