Report Suggests "Extreme" Stock Limits for GeForce RTX 5090 & 5080 GPUs in Germany

[Beermotor]

That cards far too slow

One reason they dropped out of the very high end, they can't keep up.

Possibly but I'm not sure their goal is to beat Nvidia. Their plan seems to be to strangle Intel Arc in the crib to protect their low-end AIB sales.

We absolutely do not want a situation where Nvidia is dominant. They're already showing signs they'll be worse than Intel ever was when they had complete dominance over x86.

 

[bonehead123]

'nuff said

 

[kondamin]

and you cant get anything? how is that in any way a realistic scenario....come on man the heck

in the sense that you want a 5090 and can't get a 5090, you could opt for something entirely different like a 4090

 

[spiral718]

If the stock reports are as fake as the AI generated frames per second the 50 series produces, no need to worry.

 

[mb194dc]

Probably because there's loads of 4 series GPUs in stock still. At least that's the case in the UK.

Who'll buy them once 5 series is out ?

 

[Chrispy_]

Linus may be stupid and not be able to understand that we have hit the silicon limit on the hardware side and the raster improvements on the software side.

LOL, the 5090 has 33% more CUDA cores and seems to offer ~37 more raw raster performance, so the 4090 was not even close to the silicon limit.

Nvidia have chosen to dedicate huge amounts of their silicon area towards datacenter focused tensor cores and raytracing. Tensor cores alone take up more room than the INT cores that Nvidia's raster performance depends on. If they'd made a GTX 4090 instead of an RTX 4090 that couldn't do raytracing and used FSR instead of DLSS because there were no tensor cores, it'd probably have 32000+ cores instead of the 16384 they shipped it with.

So much of the transistor budget in a 4090 isn't really intended for gaming. DLSS is just a way Nvidia have found to sell their LLM/AI datacenter silicon to the (largely gullible) gaming audience. Does AMD have tensor cores? Does Intel have tensor cores? Do GTX 16-series cards have tensor cores? Yet they can all still do upscaling, and as far as I can tell, AMD's 7000-series framegen actually looks and feels better than Nvidia's 40-series framegen, despite no tensor cores.

 

[gffermari]

LOL, the 5090 has 33% more CUDA cores and seems to offer ~37 more raw raster performance, so the 4090 was not even close to the silicon limit.

Nvidia have chosen to dedicate huge amounts of their silicon area towards datacenter focused tensor cores and raytracing. Tensor cores alone take up more room than the INT cores that Nvidia's raster performance depends on. If they'd made a GTX 4090 instead of an RTX 4090 that couldn't do raytracing and used FSR instead of DLSS because there were no tensor cores, it'd probably have 32000+ cores instead of the 16384 they shipped it with.

So much of the transistor budget in a 4090 isn't really intended for gaming. DLSS is just a way Nvidia have found to sell their LLM/AI datacenter silicon to the (largely gullible) gaming audience. Does AMD have tensor cores? Does Intel have tensor cores? Do GTX 16-series cards have tensor cores? Yet they can all still do upscaling, and as far as I can tell, AMD's 7000-series framegen actually looks and feels better than Nvidia's 40-series framegen, despite no tensor cores.

We have hit silicon limit at 5nm. 3nm was not available because Apple bought them all.
So nVidia had to introduce the software gimmicks to compensate. A GTX 4090 would be a glorified 7900XTX. No thanks.
The tensor cores etc. are necessary if you want to run RT/PT today at decent fps and not in 10 years natively.

In general, if TSMC does not find a solution after the 2nm, nVidia will be giving us only DLSS updates.
Unless a groundbreaking multi ccd is invented that performs in gaming.

AMD fails because of DLSS. Not because of their gpus are slow.
FSR 4 will be running in AMDs equivalent tensor cores and most likely will be competitive to DLSS.
If that happens, 9070XT will be a success regardless the native performance.

 

[Bomby569]

what was said above, they want to move the 4000 gpus first, there is still stock everywhere.

 

[robal]

Just ignore this news (and RTX 5090), and move on with your life.
You don't need those for anything.

 

[lexluthermiester]

If the Germans are getting hosed this mean we're all getting hosed =/

Not if special situation exists to explain this problem.

This begs the question:
Why just Germany?

 

[Waldorf]

@gffermari
lol, if that was the case, i would have seen zero improvement switching from 8C to 16C (5800 vs 5950, same hw for everything else)
but fact is it did, to the point that i can even play most older games (outside online/shooters) with power saving profile, as my ram thru output is almost double because of 2 ccds.
and i dont play anything newer than 2019, so yeah, obsolete joke to say they dont perf better.

@Beermotor
Nv is already "dominating" since rtx2xxx series, or they wouldnt be able to get away with cards costing 4 digits.

 

[dyonoctis]

LOL, the 5090 has 33% more CUDA cores and seems to offer ~37 more raw raster performance, so the 4090 was not even close to the silicon limit.

Nvidia have chosen to dedicate huge amounts of their silicon area towards datacenter focused tensor cores and raytracing. Tensor cores alone take up more room than the INT cores that Nvidia's raster performance depends on. If they'd made a GTX 4090 instead of an RTX 4090 that couldn't do raytracing and used FSR instead of DLSS because there were no tensor cores, it'd probably have 32000+ cores instead of the 16384 they shipped it with.

So much of the transistor budget in a 4090 isn't really intended for gaming. DLSS is just a way Nvidia have found to sell their LLM/AI datacenter silicon to the (largely gullible) gaming audience. Does AMD have tensor cores? Does Intel have tensor cores? Do GTX 16-series cards have tensor cores? Yet they can all still do upscaling, and as far as I can tell, AMD's 7000-series framegen actually looks and feels better than Nvidia's 40-series framegen, despite no tensor cores.

Actually, since RDNA 3 AMD got AI cores and Intel always had them with XMX. And AMD is also going to sell to gamers their Datacenter silicon with UDNA. RDNA 4 is the last "gaming only" focused arch.
AMD's FSR 4 will use machine learning but requires an RDNA 4 GPU, promises 'a dramatic improvement in terms of performance and quality' | PC Gamer
AMD RDNA is coming to an end with UDNA taking the torch | Club386

AMD is once more combining its GPU development efforts into a single design, mixing consumer and data centre needs. The new architecture should combine CDNA and RDNA, opening the way back for high-end consumer cards.

 

[Xaled]

This is just illegal. Announcing something's price and selling it at a different price is just illegal. Nvidia has been doing this for years, nobody, no goverment etc. has taken any action against them yet, so weird and suspecious.

 

[Visible Noise]

This is just illegal. Announcing something's price and selling it at a different price is just illegal. Nvidia has been doing this for years, nobody, no goverment etc. has taken any action against them yet, so weird and suspecious.

When has Nvidia ever sold something over MSRP?

Answer: Never.

What would really be illegal is Nvidia telling board manufacturers what they can charge.

But why do you care? It‘s obvious you would never buy an Nvidia card. You just want them to force AMD’s prices lower.

 

[LittleBro]

Meanwhile in middle EU, some e-shops have already listed RTX 5070 Ti, RTX 5080 and RTX 5090 with prices:

ASUS ROG Astral LC RTX 5090 no price yet
ASUS ROG Astral RTX 5090 = 3080€ (2505€ without VAT)
ASUS TUF Gaming RTX 5090 = 2880€ (2342€ without VAT
ASUS ROG Astral RTX 5080 = 1650€ (1342€ without VAT)
ASUS TUF Gaming RTX 5080 = 1550€ (1260€ without VAT)
ASUS PRIME RTX 5080 = 1400€ (1138€ without VAT)
MSI and Gigabyte also listed but no prices yet.

These look like Founders Edition cards:
PNY RTX 5090 = 66k CZK (around 2650€, 2190€ without VAT)
PNY RTX 5070 = 19k CZK (around 760€, 628€ without VAT)
PNY RTX 5070 Ti = 26k CZK (around 1040€, 860€ without VAT)

Mainstream RTX 5070 Ti for more than 1k € ...
Expect AIB RTX 5060 cards to cost no less than 500 € incl. VAT.

 

[chrcoluk]

This is just illegal. Announcing something's price and selling it at a different price is just illegal. Nvidia has been doing this for years, nobody, no goverment etc. has taken any action against them yet, so weird and suspecious.

Happens all the time here, supply and demand, only way manufacturers can enforce pricing is print price on the label (which usually needs a deal with retailer that they pay less to keep their margins) or sell direct to consumers.
In Nvidia's case they do it via Founder Edition cards which are only sold at authorised retailers who of course honour the price. Got my 3080 FE at MSRP whilst AIB's were gouging like crazy and selling direct to miners. 4080 FE Super also MSRP.

 

[lexluthermiester]

This is just illegal. Announcing something's price and selling it at a different price is just illegal. Nvidia has been doing this for years, nobody, no goverment etc. has taken any action against them yet, so weird and suspecious.

When has Nvidia ever sold something over MSRP?

Answer: Never.

What would really be illegal is Nvidia telling board manufacturers what they can charge.

But why do you care? It‘s obvious you would never buy an Nvidia card. You just want them to force AMD’s prices lower.

The both of you need to stop with the nonsense drivel. You're embarrassing yourselves.

 

[Chrispy_]

Actually, since RDNA 3 AMD got AI cores and Intel always had them with XMX.

But FSR3 works on RDNA and RDNA2, Intel GPUs without XMX and Nvidia cards like the 1660S without Tensor cores, so it clearly doesn't need them for upscaling. Maybe they help unburden the shader cores in some way, but do they help as much as doubling the shader count would help?

XeSS works on literally any GPU or IGP as long as it supports Shader Model 6.4 which goes back several generations for all GPU brands.

 

[Xaled]

When has Nvidia ever sold something over MSRP?

Answer: Never.

What would really be illegal is Nvidia telling board manufacturers what they can charge.

But why do you care? It‘s obvious you would never buy an Nvidia card. You just want them to force AMD’s prices lower.

Why do you lie? Since the mining craze in 2020 Nvidia has never sold anything at the announced MSRP. They've always being lying and giving fake msrps. Not to mention the Founder edition that almost never existed.
Your second lie is that i won't buy an Nvidia card, Both of my PC and laptop have Nvidia GPUs. And I'll buy whoever i want and I'll still bash them if they do anything wrong

 

[dyonoctis]

But FSR3 works on RDNA and RDNA2, Intel GPUs without XMX and Nvidia cards like the 1660S without Tensor cores, so it clearly doesn't need them for upscaling. Maybe they help unburden the shader cores in some way, but do they help as much as doubling the shader count would help?

XeSS works on literally any GPU or IGP as long as it supports Shader Model 6.4 which goes back several generations for all GPU brands.

AMD themselves said that FSR 3 was limited in term of improvement in Image quality, hence why they went the nvidia road with FSR 4. And it's been knonw that Xess work best on Intel GPU when it's accelerated. Better IQ, and higher performance gains.
With the way that everyone is moving, and the money that they've been making since we entered the upscaled/AI era, I doubt that we'll get back to the era of the pure gaming GPU. AMD explained that this strategy got in the way of the democratization of ROCm and HIP. And there's seems to be other stuff happening in the research world that call for specialized hardware. CG researchers who don't work at nvida/AMD/intel have been looking at ways that AI/machine learning could be used to augment raster/ray traced graphics in real time.

Now the discussion about weither or not things would have been better if nvidia just went balls to the wall and filled the RTX 4090 with cuda is too speculative for me, the 4090 already use a LOT of power as it is, even in native gaming, and I can't imagine that doubling the CUDA core count would have given us free performance without any impact on power and heat.