AMD's next-gen RDNA 2 rumor: 40-50% faster than GeForce RTX 2080 Ti

[EarthDog]

No need to be sorry, just show me that card. the card which is more efficient at a higher clock speed.

You're missing the point I think?

All I am saying is that you cannot simply lower the clocks to achieve what you want. Power doesn't drop linearly with clocks. Yes, it goes down... that's DUH obvious, but it doesn't scale linearly.

 

[moproblems99]

Sure, but each card is different man.. you just can lower the clocks and accomplish that goal. Sorry.
I haven't come across that rumor... however, I've seen you (someone) say it twice now in this thread.

3 hours ago... https://www.pcgamer.com/nvidia-ampere-12pin-power-connector/

they aren't buying it.......

And think about it... what kind of a BEAST would Ampre have to be at over 350W? If we consider what we have now at 225W... plus a node shrink plus new arch, plus 50% more power to use.... I don't think any competitor stands a chance on the performance front if that is true.

So many rumors... so many people taking these as the gospel and standing by them... god bless the lemmings...... every one!

I believe there was a pictures card, connector, and pin out... I'll see if I can find it. I may not have been paying close enough attention.

Edit:. Yeah, I'll retract this. Definitely not enough substance. The pic of the connector is not convincing enough to think it wasn't shopped. No card pic, no pinout.

 

[cucker tarlson]

You're missing the point I think?

All I am saying is that you cannot simply lower the clocks to achieve what you want. Power doesn't drop linearly with clocks. Yes, it goes down... that's DUH obvious, but it doesn't scale linearly.

same as cu count doesn't scale.
even nvidia found that out painfully.
4300 cuda to achieve 22% over 3072 cuda.
ouch.

well,ouch for us.they probably made tons of money off TU102 2080Ti's,Titans and Quadros.

 

[EarthDog]

I believe there was a pictures card, connector, and pin out... I'll see if I can find it. I may not have been paying close enough attention.

Edit:. Yeah, I'll retract this. Definitely not enough substance. The pic of the connector is not convincing enough to think it wasn't shopped. No card pic, no pinout.

I have no idea if it is true or not but think about it.........

225W now... (2080 Ti)

350W... a 50%+ increase of power + node shrink + IPC increases from the new arch.... This thing would be a 4k/120 card if all that were true.

 

[cucker tarlson]

I have no idea if it is true or not but think about it.........

225W now... (2080 Ti)

350W... a 50%+ increase of power + node shrink + IPC increases from the new arch.... This thing would be a 4k/120 card if all that were true.

dude,2080Ti is 280 at least

 

[EarthDog]

dude,2080Ti is 280 at least

At reference speeds, its 225W. FE = 250W. That doesn't change my point, however.

EDIT: Oops... 250/260W according to NV website (reference and FE). So let me update for accuracy.................

250W now... (2080 Ti)

350W... a ~40% increase of power + node shrink + IPC increases from the new arch.... This thing would be a 4k/120 card if all that were true.

 

[cucker tarlson]

At reference speeds, its 225W. FE = 250W. That doesn't change my point, however.

EDIT: Oops... 250/260W according to NV website (reference and FE). So let me update for accuracy.................

250W now... (2080 Ti)

350W... a ~40% increase of power + node shrink + IPC increases from the new arch.... This thing would be a 4k/120 card if all that were true.

well,a 10nm node shrink.64MT/mm
it's not 7nm tsmc at close to a 100

 

[EarthDog]

well,a 10nm node shrink.64MT/mm
it's not 7nm tsmc at close to a 100

In English, plz..

 

[cucker tarlson]

In English, plz..

they're going with samsung's 8nm (10nm extension). 64 MegaTransistor/mm2. tsmc is much denser.

 

[EarthDog]

they're going with samsung's 8nm (10nm extension). 64 MegaTransistor/mm2. tsmc is much denser.

Gotcha... but still improvements to be had versus current process, yes?

 

[cucker tarlson]

Gotcha... but still improvements to be had versus current process, yes?

yuge

almost twice.tsmc's 12nm is 16nm extension at 33 mt/mm.
still more to gain than amd from 7 duv to 7 euv

 

[HD64G]

Efficiency=Perf/W and Performance is the sum of all things that result in it (arch, IPC, core and VRAM clockspeeds, VRAM bandwidth, etc).

 

[moproblems99]

At reference speeds, its 225W. FE = 250W. That doesn't change my point, however.

EDIT: Oops... 250/260W according to NV website (reference and FE). So let me update for accuracy.................

250W now... (2080 Ti)

350W... a ~40% increase of power + node shrink + IPC increases from the new arch.... This thing would be a 4k/120 card if all that were true.

I hope it is true and frankly I think they are both true-ish. If NV has 40% or more increase in performance over 2080ti then I don't think AMD can surpass them - the gap is just so large. They'll get to a very competitive spot and it will be wonderful for all.

 

[bug]

but ipc is a way to achieve higher perf/wat

It's tempting to think that, but no. The concepts are actually different. Look who's offering the highest perf/W: https://www.techpowerup.com/review/asus-radeon-rx-5700-xt-tuf-evo/29.html
It's mid-rangers which don't have the highest IPC by a long shot. And the explanation is very simple: as you increase clocks, IPC goes up, but power draw goes up faster. Inevitably you'll reach a point where IPC goes up (it always does with clock increase), but perf/W goes down. That's where high-end cards sit.

Edit: see below about IPC.

 

[EarthDog]

It's mid-rangers which don't have the highest IPC by a long shot.

But IPC wouldn't change with the same architecture. What does are clock speeds and ROPs/TMU/Shader counts.

When you increase clocks, IPC does NOT go up. Remember IPC is Instructions per CLOCK cycle. IPC doesn't change when you increase clock speeds. If I can do 100 instructions per clock at XXX MHz, It's still 100 instructions per clock at a higher MHz... the difference there is ONLY in clock speeds not how much work gets done within each cycle.

 

[bug]

But IPC wouldn't change with the same architecture. What does are clock speeds and ROPs/TMU/Shader counts.

When you increase clocks, IPC does NOT go up. The If I can do 100 instructions per clock at XXX MHz, It's still 100 instructions per clock at a higher MHz... teh difference there is ONLY in clock speeds.

Yeah, brain fart. IPC goes up when you add hardware resources, but it doesn't with clock changes.

 

[TheoneandonlyMrK]

AMD are going to win this round I think , until next year when Nvidia port ampere to Tsmc and call them supers.

Yeah Nvidia might get near or pip the performance of big Navi at the high end but at 50-100 watts more and pushed to it's limit's, I think AMD might hold back a ltd edition counter punch.

Nvidia would just roll with Samsung if they were that good but they Did run back to Tsmc in the end, but that'll take time to fit in now.

Mad rumours of 350-450 watt amperes abound.

 

[Fluffmeister]

The 2080 Ti is already as good as 2 years old, I'd hope for more than 40-50% more performance frankly.

 

[Metroid]

50% faster x nvidia 2080 ti rumor does not sound impossible, but given the misleading track record of amd in the past, however, remember that amd is 7nm and rnda 2 can indeed give that kind jump in performance. People should not underestimate AMD, look at ryzen.

 

[kings]

We have to wait and see, there is always a lot of hype with each new AMD generation. It will always be the greatest thing ever. Fury was going to be the Titan killer and the overclockers dream, Vega were the "Poor Volta" thing... In every new generation, Nvidia is "doomed" due to something special that AMD is preparing.

And frankly, in my opinion these rumors thrown into the air by random youtubers to get clicks end up damaging AMD's image more than benefiting. We've seen this in the past, people start to create an idea of what the product will be and then when it finally arrives, it disappoints, not because the product is bad, but because the expectations were unrealistic.

 

[cucker tarlson]

time to go back to amd
they'll beat 2080Ti by 50% and their drivers have no issues
they'll trounce nvidia's 450w cards

 

[GoldenX]

I expect the biggest bugfest since the HD2900 was announced, followed by 8 years of small and slow improvements.

 

[bug]

I expect the biggest bugfest since the HD2900 was announced, followed by 8 years of small and slow improvements fine wine.

Ftfy

 

[cucker tarlson]

Ftfy

lol,you won tpu for today.

 

[BoboOOZ]

they'll trounce nvidia's 450w cards

It's not 450W, it reads "up to 600W"