AMD Navi 20 Faster Than An Nvidia RTX2080Ti?

[Participant]

I'm not sure what the problem is with having the pinnacle of gaming GPUs at a pinnacle price point. The very best in the world at something is always really expensive and hardly anyone buys them relative to 1660Tis. Like Ferrari vs VW. I'm ok with it.

Price up the best floor standers and stereo for example...

 

[Zernath]

I think the entry and mid range is exempt.

Yes, at the moment. But that can always change, once upon a time high end didn't mean £1,000 or more either!

 

[Troezar]

I'm not sure what the problem is with having the pinnacle of gaming GPUs at a pinnacle price point. The very best in the world at something is always really expensive and hardly anyone buys them relative to 1660Tis. Like Ferrari vs VW. I'm ok with it.

Price up the best floor standers and stereo for example...

Cool £5k GPUs it is then, I'll take...err...none

There is also perceived value, the target market, longevity, depreciation etc. to take into account. A GPU is just a commodity, devalues rapidly, is quickly replaced, is not a status symbol and is used to play games that may cost up to £50.

Probably why at £1200-£1500 they aren't selling that many.

 

[TNA]

They have gone too far already, people are not buying the really expensive cards like the RTX Titan.

Even EVGA are not pricing their Kingpin 2080 Ti anywhere near the RTX Titan.

Having said that GPU prices are at ridiculous levels and in the end NVidia will be the losers as gamers lose interest and go for consoles or other hobbies.

+1

I am hoping the 3080 Ti is priced like the 1080 Ti, if it is then I will buy one. Much more and I am going PS5 and just buying a mid range card from the next gen and keep that for a long time to be able to play my current steam library.

 

[FoxEye]

I'm not sure what the problem is with having the pinnacle of gaming GPUs at a pinnacle price point. The very best in the world at something is always really expensive and hardly anyone buys them relative to 1660Tis. Like Ferrari vs VW. I'm ok with it.

Price up the best floor standers and stereo for example...

Prices across the range have trended upwards. With an xx60 being ~£300 and xx80 being £650 at various points.

I don't care if the top-end card is £5 thousand or £5 hundred thousand. I'm not paying either But I baulk at the idea of paying £300 for a 4th/5th tier "mid-range" card.

 

[Kaapstad]

I'm not sure what the problem is with having the pinnacle of gaming GPUs at a pinnacle price point. The very best in the world at something is always really expensive and hardly anyone buys them relative to 1660Tis. Like Ferrari vs VW. I'm ok with it.

Price up the best floor standers and stereo for example...

EVGA are asking $1900 for the new Kingpin 2080 ti and the cheapskates are selling it with an AIO cooler instead of a proper water block, what a joke.

Apparently water block versions will be available later but that will be extra.

For that sort of money you may as well buy an RTX Titan and put a waterblock on that with custom bios, at least you get 24gb of Samsung VRAM that clocks higher than the 11gb of the Kingpin lol.

Seriously cards like the 2080 Ti should be £700 or less so they sell in large numbers, without those sales there will be no Titan type cards in future architectures from NVidia or AMD.

NVidia have produced a really bad product in Turing, they need to move onto something better fast for their own sake.

 

[Participant]

Seriously cards like the 2080 Ti should be £700 or less so they sell in large numbers, without those sales there will be no Titan type cards in future architectures from NVidia or AMD.

Is this statement being made from an ideological or business point of view? Because from the former I completely agree, but there'll be teams of marketing and pricing strategists as well as cost accountants being involved in determining the profit maximising pricing structure.

There are so many different factors at play that need to be scientifically analysed before being able to determine the 'correct' price (normally the one that maximises shareholder wealth), such as supply and demand, product lifecycle, market segmentation, luxury pricing and brand perception, macro economic variables, real price effects, etc.

Yes you could say shareholder wealth is an interesting point with NVDA equity in its current state, however this would be the cum hoc fallacy hard at work and pricing the 2080Ti at £700 may have further secreted value.

 

[Kaapstad]

Is this statement being made from an ideological or business point of view? Because from the former I completely agree, but there'll be teams of marketing and pricing strategists as well as cost accountants being involved in determining the profit maximising pricing structure.

There are so many different factors at play that need to be scientifically analysed before being able to determine the 'correct' price (normally the one that maximises shareholder wealth), such as supply and demand, product lifecycle, market segmentation, luxury pricing and brand perception, macro economic variables, real price effects, etc.

Yes you could say shareholder wealth is an interesting point with NVDA equity in its current state, however this would be the cum hoc fallacy hard at work and pricing the 2080Ti at £700 may have further secreted value.

I understand what you are saying about pricing but NVidia have found themselves in a position with Turing where they have to charge more than is reasonable to recoup their costs.

Turing chips are way too big and offer very little extra conventional performance for the asking price and features like RTX and DLSS can not make up for this.

This is a classic R & D foul up that has priced the cards way too high.

 

[D.P.]

I understand what you are saying about pricing but NVidia have found themselves in a position with Turing where they have to charge more than is reasonable to recoup their costs.

Turing chips are way too big and offer very little extra conventional performance for the asking price and features like RTX and DLSS can not make up for this.

This is a classic R & D foul up that has priced the cards way too high.

The large die is due to still being on 12nm. 7nm was just not ready and even although die sizes would be smaller the costs would end up far higher.

Some recent IR shots of the Turing shows that RTX and Tensor cores are about 8% of the die area, so they have very little no added cost.

Given 12nm is more or less just the same as 16nm, for Turing to be a successor to pascal required the die sizes to go up a lot. Unlike pascal there was no real compute stuff to take out and Pascal was already very lean.

I think you are also way overestimating the die size costs. Even with the extra 8% for RTX & Tensor cores, the Turing chips would be cheaper than Pascal to produce for a given number of compute units etc because yields are higher on the more mature process.

The price difference doesn't come form the dies, but the increasing R&D costs, increased memory costs, and increased margins.

If the RTX & Tensor cores were not added total manufacturing costs would be a few percent less at most, basically indistinguishable. Having 8% more CUDA cores might seem nice but in all likelihood is just not possible because core counts always have to scale in a particular way (normally following a power of 2).


When RTX is impacting prices it is purely due to marketing. Nvidia believe they can charge more due to the features. The market will decide and if they disagree with Nvidia then NV will simply reduce margins. Nvidia's statement to investors is so far Turing has beaten Pascal in revenue, that is all shareholders typically care about. Also, Turing was the fastest growing GPUs on steam survey, and nvidia;s market share is only increasing. So most of the siogns point towards Turing selling quite well in and of itself.

 

[TNA]

But I baulk at the idea of paying £300 for a 4th/5th tier "mid-range" card.

Exactly.

 

[Participant]

Exactly.

I agree it's a lot of money when you can get an entire PS4 Pro for £350.

Don't get me wrong, I have an intense dislike of the way midrange card pricing has outstripped general inflation by a factor of about 10, and I live in hope that at some point the market won't sustain it and manufacturers will have to reassess.

 

[TNA]

I agree it's a lot of money when you can get an entire PS4 Pro for £350.

Don't get me wrong, I have an intense dislike of the way midrange card pricing has outstripped general inflation by a factor of about 10, and I live in hope that at some point the market won't sustain it and manufacturers will have to reassess.

Yeah. Hope it is sooner rather than later. All that is happening now is we are lining shareholders pockets and buying even more leather jackets for Jensen. It is obvious when you look how there share price did a 10x in a few years.

They could likely sell these cards at half the price and still turn a decent profit. Just hope they realise they went too far with Turing and go back to pascal pricing.

 

[LePhuronn]

I'm not sure what the problem is with having the pinnacle of gaming GPUs at a pinnacle price point. The very best in the world at something is always really expensive and hardly anyone buys them relative to 1660Tis. Like Ferrari vs VW. I'm ok with it.

But it's not the pinnacle of gaming GPUs, that's the point. Very few games utilise RTX tech, and those that do see woeful performance. And traditional raster performance is not really any better than the previous generation. And it costs a massive premium to partake in this "privilege". So where is that pinnacle?

Using your car example, we're not talking Bugatti Veyron here. That beast understandably and justifiably cost £1M because VW really did create the pinnacle of automotive engineering. But imagine Ferrari took a standard 458 (3 year old design with 3 year old performance) added a tiny KERS system that only offered an extra 20BHP in certain circumstances but then claim it's the greatest thing yet and charge £1M for it because "Ferrari". That is not the pinnacle of automotive engineering and does not warrant the asking price. But some numb nut Ferrari fan will rationalise it and buy one anyway.

RTX 2000 is that Ferrari.

 

[Kaapstad]

The large die is due to still being on 12nm. 7nm was just not ready and even although die sizes would be smaller the costs would end up far higher.

Some recent IR shots of the Turing shows that RTX and Tensor cores are about 8% of the die area, so they have very little no added cost.

Given 12nm is more or less just the same as 16nm, for Turing to be a successor to pascal required the die sizes to go up a lot. Unlike pascal there was no real compute stuff to take out and Pascal was already very lean.

I think you are also way overestimating the die size costs. Even with the extra 8% for RTX & Tensor cores, the Turing chips would be cheaper than Pascal to produce for a given number of compute units etc because yields are higher on the more mature process.

The price difference doesn't come form the dies, but the increasing R&D costs, increased memory costs, and increased margins.

If the RTX & Tensor cores were not added total manufacturing costs would be a few percent less at most, basically indistinguishable. Having 8% more CUDA cores might seem nice but in all likelihood is just not possible because core counts always have to scale in a particular way (normally following a power of 2).


When RTX is impacting prices it is purely due to marketing. Nvidia believe they can charge more due to the features. The market will decide and if they disagree with Nvidia then NV will simply reduce margins. Nvidia's statement to investors is so far Turing has beaten Pascal in revenue, that is all shareholders typically care about. Also, Turing was the fastest growing GPUs on steam survey, and nvidia;s market share is only increasing. So most of the siogns point towards Turing selling quite well in and of itself.

Full fat Pascal GP102 = 11,800,000,000 transistors

Full fat Turing TU102 = 18,600,000,000 transistors

57.63% increase.

Efficiency wise there is not much difference between Pascal and Turing.

We don't see anywhere near 57.63% increase in performance when comparing a Pascal Titan to a Turing Titan. A lot of those extra transistors are going toward Ray Tracing and DLSS and adding to the cost.

It is interesting to point out that a 2080 Ti is about 57.63% more expensive than a 1080 Ti was.

 

[TNA]

But it's not the pinnacle of gaming GPUs, that's the point.

True. DLSS in particular has been very disappointing imo. In it’s current form it is pointless. Hope they either improve it a lot by end of the year or don’t waste transistors on it.

 

[D.P.]

Full fat Pascal GP102 = 11,800,000,000 transistors

Full fat Turing TU102 = 18,600,000,000 transistors

57.63% increase.

Efficiency wise there is not much difference between Pascal and Turing.

We don't see anywhere near 57.63% increase in performance when comparing a Pascal Titan to a Turing Titan. A lot of those extra transistors are going toward Ray Tracing and DLSS and adding to the cost.

It is interesting to point out that a 2080 Ti is about 57.63% more expensive than a 1080 Ti was.

Transistor have zero cost. The cost is in the die area. Turing does have many more transistors, but the increase in die area due to RTX is well under 10%.


And you are also wrong, Turing actually has a really big improvement in efficiency. There are big changes to the CUDA cores, CUDA functionality, new features like variable rate shading, improved scheduling and load balancing. There is a 50% efficiency increase per CUDA core in Turing.

 

[Kaapstad]

Transistor have zero cost. The cost is in the die area. Turing does have many more transistors, but the increase in die area due to RTX is well under 10%.


And you are also wrong, Turing actually has a really big improvement in efficiency. There are big changes to the CUDA cores, CUDA functionality, new features like variable rate shading, improved scheduling and load balancing. There is a 50% efficiency increase per CUDA core in Turing.

Transistors have everything to do with it, the less you have the smaller the die. If so many were not dedicated to RTX and DLSS the dies would be smaller and cheaper.

As to efficiency I am talking raw performance as measured in game or synthetic benches. For this there is very little to separate Pascal Turing or Volta.

NVidia have screwed up badly with Turing offering features that are extremely poor value for money.

 

[Gerard]

NVidia have screwed up badly with Turing offering features that are extremely poor value for money.

And you punished them accordingly by throwing in the region of £8K+ at them.

 

[TNA]

And you punished them accordingly by throwing in the region of £8K+ at them.

Lol.

 

[Kaapstad]

And you punished them accordingly by throwing in the region of £8K+ at them.

At least I am honest enough to say when things are not right.

I am not into defending vendors come what may, if there is something wrong I say it.

For the record the RTX Titan is the best card I have owned, yes the asking price is appalling and RTX/DLSS leave something to be desired but it is still the best gaming card out there.