• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

NVIDIA 5000 SERIES

hobib_ali said:
well my colleague sold his Suprim 4090 for 1450 and it was only 8 months old, he provided the OG invoice for warranty purposes and it was snapped in the 30 mins, on average they go used for 1500ish, I dont think they will go lower because when the true benches gets published it will shy away from the 5090 by 10-15% max, that about sums it up in terms of price, 1500 for a 24GB 4090 or 2400 for a 32GB 5090....
Click to expand...
The thing that has gone bananas with used prices is people are obviously not adjusting for the fact that, well... it's used. 99% of the time, you have no idea how the card was used and for how long. There's also the warranty question which, in the best case involves only a year or two and the seller providing an invoice and doing work which is no garantee in itself. And then, of course, there's the absolute that the card will very likely fail sooner than new.

The equation to figure used cost is basically one of risk/reward.

I've been buying and selling cards for 30 years now. It is a viable way to get your hands on hardware. But after having a few fail and being ÷/£& out of luck, you need to factor that in to how much the card is actually worth and be ready for the loss. I see people still buying used 4090s today for £1500 and think they have in no way considered these variables.

I won't get into the risk as a seller which is more and more fraught with lost investment and frustration.
 
Richdog said:
Yeah, people really do spend their time watching some really trash Youtubers.
Click to expand...

He’s not wrong in some of his his logic though. Double the bandwidth does not equal double the performance. It’s about balance and all double the bandwidth will do is make something else a bottleneck.
 
Last edited:
ICDP said:
He’s not wrong in some of his his logic though. Double the bandwidth does not equal double the performance. It’s about balance and all double the bandwidth will do is make something else a bottleneck.
Click to expand...
Lots of the first party benchmarks have used DLSS2 (including at the performance setting), right? Aka not native 4K. I wonder if the bandwidth will help much at native 4k. Look forward to seeing those benchmarks in 8 days.
 
Absolutely Clueless said:
Lots of the first party benchmarks have used DLSS2 (including at the performance setting), right? Aka not native 4K. I wonder if the bandwidth will help much at native 4k. Look forward to seeing those benchmarks in 8 days.
Click to expand...

Not the reviews I read no. They use 4K native and then you get separate DLSS results, or you extrapolate DLSS performance by looking at the native lower resolution benchmarks. For example 1440p gives me an idea of my 4K quality DLSS settings.

So increasing bandwidth only helps up to a point where the rest of the GPU specs (or CPU) become the bottleneck. In all things PC it’s about balance. Overclocking a 4090 VRAM yields about 3% - 4% extra performance before it becomes pointless.

So the guy in the video may be annoying, but he is correct. Double the bandwidth does not equal double the performance if the rest of the specs only increase 30%.
 
Last edited:
ICDP said:
Not the reviews I read no. They use 4K native and then you get separate DLSS results. So increasing bandwidth only helps up to a point where the lack of other parts of the GPU (or CPU) becomes the bottleneck. In all things PC it’s about balance. Overclocking a 4090 VRAM yields about 3% - 4% extra performance before it becomes pointless.

So the guy in the video may be annoying, but he is correct. Double the bandwidth does not equal double the performance of the rest of the specs only increase 30%.
Click to expand...

Bandwidth is correct - so many people seem to think that memory bandwidth increases performance, it does not. All memory bandwidth does is remove bottlenecks on data transfer speed with the GPU cores if they are bottlenecked, and if there is no bottleneck then no matter how much more bandwidth you add, framerate won't change whatsoever.

Memory overclocking on RTX4000 is a good example of that, the cards can do +2000mhz overclocks easily but the performance stops increasing well before reaching that.

As a result of limited changes to the core design; it's clear that the 5090 doesn't need 1.8TB/s bandwidth - a lot of this bandwidth will be unused and so memory overclocking on the 5090 probably won't be a thing worth doing.


The only thing where the 1.8TB/s will be useful is Compute, mining and AI tasks, not gaming.
 
Last edited:
Perhaps the additional VRAM bandwidth was required for the AI parts to function sufficiently well enough to be worthwhile?

Only being devils advocate, don’t shoot me!
 
Grim5 said:
I think Nvidia could kill off the AIBs if it wants to, especially if its offering superior coolers and lower prices. The AIB's contribute nothing apart from supply chain, Nvidia to date is unable or unwilling to supply enough volume globally and AIB's are left to fill the gap
Click to expand...

I agree. I get the impression that Nvidia's been ramping their supply chain/manufacturing capability to eventually become the sole supplier. Why else pour so much into R&D for FE cards.

I think the ultimate goal is so not sell GPU's - but to sell a complete PC, with custom Nvidia ARM CPU with some kind of x86 emulation, or agreement from developers to write specifically for Nvidia PC's.
 
RanxZy said:
It's fairly easy to cripple a 12GB card at 4K in select modern titles. It would be interesting to see how the 4x Frame Gen works for the 16GB cards vs the 4090 when they run out of memory.
Click to expand...
I think you're underestimating the difference the new compression tech in the 50 series will have on vram plus lower vram requirement overall for things like Ray tracing.
 
ne0 said:
So you think the 5080 will just conk out as soon as you switch on 4x frame gen because it has 16gb vram. Cool.
Click to expand...
No I was just suggesting that vram compression techniques being advertised as meaning you need less vram have all proved to be rubbish. 4x frame gen will of course run fine as that's the only reason it will be equal to a 4090 :D
 
ne0 said:
I think you're underestimating the difference the new compression tech in the 50 series will have on vram plus lower vram requirement overall for things like Ray tracing.
Click to expand...

Developers actually have to make proper use of some of these new techniques for it to mean much and so far they've not raced to embrace such technology historically.
 
Devious said:
I sold my 4090 expecting to try and get a 5090 but not too impressed withe the uplift for the requured power.

I might go for the 5080 now to put me on until until the 60 series as that will be probably when GTA 6 etc will be coming out. That will at least let me try the new features out.

I think the 5090 might end up a depreciation disaster compared to the 4090 if the 60 series comes out on a new node.
Click to expand...
odd move, the roughly 30% uplift and power was known about for a long time.
 
You must log in or register to reply here.
Back
Top Bottom