NVIDIA keynote and GeForce RTX 40 GPU / DLSS 3.0 announcements

YES.

Up to now the exchange rate average was €1 to $1.14. In EU we have 20-24% VAT. So let's say 20% which is the avg sold on almost all Eurozone countries.

A $1200 3080Ti has EU MSRP €1089 with VAT. At 1.14 exchange rate we had in 2021, that was $1241 (with VAT), without VAT was $1035.

So Nvidia did a 13.7% (1200* 86.3% = 1035) discount for the EU market, to keep us paying the similar final price in USD with the Americans. You will find the same rate was for 2000 and all previous Nvidia cards. Nothing do with with exhange rates etc. Just to keep the Europeans paying the same price with the Americans.

Let's move to today.

RTX 4080 12 GB, American MSRP $900, in EU is $/€1100, without VAT $917.

Nvidia decided to remove the 13.7% discount on the EU marker to keep the prices in line with the Americans, even added an extra 2% on the price for the 4080 12GB.

If it had kept the discount, Nvidia would have charged us around €/$940 (with VAT) not €/$1100

The 4090 has net €/$1624 (US MSRP $1599, +1.17% extra in EU) and with VAT €/$1949.
The 4080 16GB has net €/$1224 (US MSRP $1199, +2.04% extra in EU) and with VAT €/$1469.

So the Nvidia is charging the Europeans now on average 15.7% higher prices with RTX4000 series than 3000 series PLUS the VAT on top. (we pay VAT on that 15.7% price increase).

Sorry for the long post with the maths.

 

As you see on the above post, Nvidia was doing a 13.7% discount on Europe to level off the VAT increase and have the Europeans pay the similar price with the Americans. With the RTX4000 series, not only removed the discount, but added a 2% extra so in total added 15.7% on the price of their GPUs for the EU market + VAT compared to RTX3000 series and previous generations.

Nothing to do with the EUR-USD exchange rate as I have factored it in.

 

You don't have to "imagine"... the specs are pretty clear... each 8pin supplies maximum 150W.

So that means your card only uses 150W from that plus I think 75W from the PCI bus itself.

2 x8pin = 300W
3x8pin = 450W

I think the new single power connector allows a maximum of 600W

 

I want to see the RAW RTRT muscle at MAX AIR performance against the 3xxx RTX series and AMD options without the DLSS assistance.

Let's see the first GPU to push 4K RTRT @ 120 FPS on all horsepower!

 

no it doesn't.

 

Yeah the more i think about latest DLSS the more of a con-job it appears to be...

Buyers will be delighted to see their game running at "4K 120FPS".... but in reality it is being rendered at 1080P, 60FPS and then doubled up in resolution and a fake frame inserted between each real frame.

I wonder how the latest Futuremark type benchmarks will handle this... will they allow this fake double frame-rate to be compared directly with older cards all-true frame-rate?

 

Hah. I was thinking at that tbh.

If Nvidia cheats on Futuremark with the "AI created frames" to boost the FPS, and the suit is patched in 3-4 months' time to solve the issue, the original numbers from the benchmarks will show a distorted performance for years to come.

But knowing Futuremark they will leave it as it.
Do you remember the issue with the code path for the acync compute on AMD vs Nvidia GPUs in Timespy having Nvidia showing higher numbers than AMD because of that?

 

Saw this REDDIT post at TPU and it reeks of puke that a 4050 might be faster at DLSS3 than a 3080 or 3090 because of reasons...

 

I just want that stuff they smokin', must be indeed a mind blower

 

More DLSS junk, now with worthless interpolation .

DL"SS" has been a joke from the start. Faux performance uplifts for people that are too proud to reduce settings and want to pretend they aren't dropping resolution. I hope FM won't allow DLSS for benchmark scores as that would defeat the purpose.

 

lol

the number one trend in pc's is ITX, then watercooling (incl aio)

plus, there are things called PCIe risers that allow better fitment .
there are several cases (all under $125) that still are SFF(ish) and have three slots for standard or vertical placement - and have good air flow

 

How is that reasons? Seems like a ~4x uplift in tensor performance and an upgraded optical flow accelerator makes it possible. What other reasons would you want?

Ehh - like 90% of game optimizations are 'faux performance uplifts' by that definition but I don't think you'd ask the devs to turn off that, would you?

I do agree that using DLSS for benchmarks, especially when it's unclear what is/isn't enabled is bullshit -- I think nvidia should be punished for that performance slide, it's garbage. But DLSS itself is just an extension of techniques that are baked into most games -- you just don't have the option to disable/enable them.

 

NV said the 4090 has a 4x8pin adaptor for the new style connector

 

That`s why we are waiting for proper reviews.

Still, i have no problems believing that the cards are going to be very fast, but the thermal figures are going to be problematic, to say the least, because no one deploys a 4 slot cooler unless they really need it...

And then, there is the pricing issue...

Interesting, i didn`t know that.

So europeans are being milked even harder...

 

very true. i'm one of those folks (though "rich" is a relative term and i can assure you that $150k/yr is not rich in Silicon Valley, but that actually underlines your point).
Americans have a strong dislike for regressive taxation like VAT, but we know it's a factor in EU and other places.

however, $1.5k is still a lot of money because the cost of living here is the highest in the USA (especially rent/mortgages). most of Silicon Valley is populated by renters with an average rent of $4k, that's right $48K/year for rent when in most of the USA (outside of California) you could have a mortgage on a palatial villa for the same.
a reality here is the so-called "hacker house", with people sharing every room except the kitchen and bath for sleeping and with cars parked all over the yard. this is the entry level world for programmers/IT specialists and they're still paying a hefty price to live in these conditions.
and those folks are the ones earning $100-125K.

$150K allows for a better situation, although it will typically involve a commute from the less expensive bedroom communities across the Bay. above that point, you still aren't "rich", but you can afford to live within a short commute (in the Valley desirable areas are measured by the minutes it takes to go to work.

 

They said the 30x0 series cannot have it because it is slow, not that the HW does not exist, even on the 3080/Ti/3090/ti?!?!?!

Seriously? If a 4090 running DLSS3 at 120 FPS @ X resolution, then 3090Ti can do it at 30 FPS so why not offer it?

Why not allow the 3000 RTX series to run DLSS3 for better QUALITY but those users have to drop resolution to sustain FPS or accept lower FPS than 4000 RTX series, which is the main difference from DLSS2?

So if a 4050 @ 1/4 of the performance of a 4090 would probably be in the same performance ballpark as a 3080/3090/Ti in terms of DLSS3 performance so why not offer it?!

They are artificially NOT offering DLSS3 because of "reasons"...

 

The problem with EU pricing is the collapse of the Euro. Nvidia is simply pegging their prices to the greenback. Polish MSRP prices have been calculated straight from the dollar with 23% VAT tacked on. Almost to the cent, if you take the current exchange rate into account.

 

Few things here.

A. We have no idea if a 4050 is going to have DLSS3. When turing came out there were low end Turing cards that didn't have RT/DLSS.

B. They said the performance AND quality of the optical flow accelerator is improved in ADA.

Even with that upgrade in quality in this video:



Pause the video at any point when spiderman is running up the wall at 1:33 seconds in. There's artifacting in DLSS3 galore:


DLSS 3.0


No DLSS.

So even with an upgraded, high quality auto-encoder accelerator, it still is artifacting. Based on what the post said and what we know about image stabilization on an autoencoder (powerpoint presentation that nvidia did on auto-encoders back in 2017) - on a 3K series that artifacting would be significantly worse.

So again, it could very well just be that DLSS 3 would be dogshit on 3000 series and that's why it's not enabled. Honestly after looking at that video it's arguably bad on a 4K series in it's current state.

You also have to keep in mind that DLSS3 isn't going to improve input latency of native frame generation over 2/no DLSS. So running a game at 30fps with DLSS 3 on would still feel like an abysmal experience due to latency impact. So potentially older cards with less native performance wouldn't yield an acceptable framerate to make DLSS3 worth enabling in those instances.

 

Honestly, I doubt we will be seeing the lower stack any time soon. Seems to me Nvidia is content to have the 4090/4080 and the fake 4080 with Ampere serving as the - ahem - budget offerings for the time being.

 

I rather take a gamble with AMD drivers.