• 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 ‘Ampere’ 8nm Graphics Cards

RTX Titan is 754mm^2 on 12nm with 4608 cuda cores

4608 vs 7680 = +65%

The claim is "Double RT and Tensor cores"

Half of the die is Cuda cores, the other half RT and Tensor Cores.

RTX Titan:
337mm^2 Cuda Cores + 65% = 622mm^2
337mm^2 RT and Tensor Cores X2 = 674mm^2

Die Size @ 12nm = 1296mm^2

Ampere is 7nm

12nm is 71% lager than 7nm. So 1296mm^2 on 7nm would be 758mm^2 (758 + 71% = 1296)

The 12nm 754mm^2 die is massive, the same sized die on 7nm would be even more massive, but probably doable.

Its not going to be cheap tho...

Shaders also don't scale 100%, more like 70 - 80%, so the actual performance uplift from +65% shaders is probably around +50%. and £1800.

Enjoy your new price overlords......

oshaqk5.jpg.png

it wouldn't just be shaders there would be nice IPC gains too and a rework of the rtx and dlss algorithms but yes cores and clock speed don't necessarily have a linear scale. A card with these specs would be too expensive at 7nm for gamers
 
it wouldn't just be shaders there would be nice IPC gains too and a rework of the rtx and dlss algorithms but yes cores and clock speed don't necessarily have a linear scale. A card with these specs would be too expensive at 7nm for gamers

That is generally where a lot of the speed up comes (ignoring shrinks/frequency difference due to node changes, etc.) - reworking of caches and so on to improve throughput, etc. traditionally the number of SMs, etc. wouldn't be directly comparable to the generation before (unless a refresh) due to architectural changes.
 
RTX Titan is 754mm^2 on 12nm with 4608 cuda cores

4608 vs 7680 = +65%

The claim is "Double RT and Tensor cores"

Half of the die is Cuda cores, the other half RT and Tensor Cores.


I'll stop you right here. RTX + Tensor cores combines amounts to 8-9% of the die area of Turing, and removing Tensor cores requires the addition of FP16 units such as Turing GTX cards.
 
Wccftech is a tabloid tech news site and like any tabloid news site has some good articles and gets some stuff right and some bad articles where it gets stuff wrong. To ignore it is just silly and counter-productive.
Anyone could do that though, it adds no more value to the topic than anyone here could muster up if they felt so inclined.

I could claim to know Samsung has started Ampere production in volume, release a tweet and then generate a news article about said tweet with a caveat. Would be complete self-generated ****** but there is a 20% chance it might be true, and if it was they’d claim the source to be historically reliable. Utter garbage if you ask me....
 
Anyone could do that though, it adds no more value to the topic than anyone here could muster up if they felt so inclined.

I could claim to know Samsung has started Ampere production in volume, release a tweet and then generate a news article about said tweet with a caveat. Would be complete self-generated ****** but there is a 20% chance it might be true, and if it was they’d claim the source to be historically reliable. Utter garbage if you ask me....

So are you saying that there is nothing useful or truthful on that site? That's a rhetorical question by the way, because obviously there are useful and truthful articles on the site. It's a news and rumours site so just accept it will be used, choose not to click or read it and stop whining about it because I assure you no-one cares about any distress you feel when they are linked to.
 
it wouldn't just be shaders there would be nice IPC gains too and a rework of the rtx and dlss algorithms but yes cores and clock speed don't necessarily have a linear scale. A card with these specs would be too expensive at 7nm for gamers

IPC gains?

Is that a fact?, or did you just make that up and assert it as a fact?
 
The new Xbox's specs were just shown off by Phil Spencer. This new console features an RDNA 2.0 GPU that is 24% faster than the 5700xt! Plus it also does Variable Rate Shading and Hardware Accelerated Ray Tracing.

So clearly AMD has more up it's sleeves than they showed off last year.
 
The new Xbox's specs were just shown off by Phil Spencer. This new console features an RDNA 2.0 GPU that is 24% faster than the 5700xt! Plus it also does Variable Rate Shading and Hardware Accelerated Ray Tracing.

So clearly AMD has more up it's sleeves than they showed off last year.
Trying to remember who it was who was adamant that the consoles would be at best a 5700 XT :p
 
Its stretching it though isnt it, what I mean by that is the more the time ticks on into 2020, the more time they have of dropping in a card better than a 5700. If its accurate and they do throw in a meaty GPU, it will be Sony taking the hit or rather stealthing it out of the consumer.

People would have been guessing 5700 because of the price point. If you want a console to be between £4-500 it would have a far cheaper GPU component - imagine having that as your Gaming rig budget!
 
The new Xbox's specs were just shown off by Phil Spencer. This new console features an RDNA 2.0 GPU that is 24% faster than the 5700xt! Plus it also does Variable Rate Shading and Hardware Accelerated Ray Tracing.

So clearly AMD has more up it's sleeves than they showed off last year.

Is Kirsty doing the PS5?
 
Not recently but they did on the XB360 and PS3 for many years, I remember reading the time they finally started to make a profit on them but AFAIK since then they have launched with no loss leaders.
 
The new Xbox's specs were just shown off by Phil Spencer. This new console features an RDNA 2.0 GPU that is 24% faster than the 5700xt! Plus it also does Variable Rate Shading and Hardware Accelerated Ray Tracing.

So clearly AMD has more up it's sleeves than they showed off last year.



The 24% is going to be under some artificial best case scenario, not an across the board change.

Vega supposedly had VRS but there was a hardware issue so was never implemented in drivers. People way over hype this technology, you get a few percentage boost in performance but it requires active developer support.
 
Back
Top Bottom