• 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

Think I'm settling on the founders edition 3090 if I can get one. Better not be anymore than $1400 -.- you'd hope it would be £1049-1099 with the exchange rate.
 
So its 12% reality down!!!
This much extra works!

It's actually 12% buzzwords expanded to 68% with DLSS3.0!

:p

And I foolishly thought that the FE was big but this is positively enormous.

Looks like 2.5/3 slot and I would guess around 340-360mm long. :O

Well its very weird Nvidia is rating the GPUs for so much power with a node shrink. Even the RTX3080 looks more relatively cut down too(if you compare the Titan RTX and the RTX2080TI).
 
Well if you look at the RTX3070 and RTX3070TI,the GDDR6X and extra shaders is only adding 30W more board power. But if you look at the RTX3080 and RTX3090,despite the latter having 2.4X the amount of GDDR6X and far more shaders,the board power is only 30W higher.

Transitors are not the same - this affects power usage particularly, that's where binning comes in.
 
Transitors are not the same - this affects power usage particularly, that's where binning comes in.

The RTX3080 is much more cut down compared to the RTX3090 than the RTX2080TI was compared to the 24GB Titan RTX. The same goes with the Titan X/Titan Xp against the GTX1080TI.

I think the GDDR6X is a red herring,when people keep saying its the main reason power consumption is so high,as Micron says otherwise. I think the core is having yield problems,and the RTX3090 just has the best binned cores,but the GA102 in the RTX3080 is going to closer to what is actually doable in volume. It is nearly 20% cutdown. If it is indeed Samsung,then they don't have as much experience as TSMC in making large dies.

Going by the 34.5 billion transistors leak and that the 2080ti has 18.6 billion transistors. It seems to me more likely to be on 7nm somehow?

Samsung have 7NM/5NM nodes too. They ditched Qualcomm as a customer,and bought a ton of EUV machines,so it means someone needs a lot of capacity.
 
It would make sense to use GA102 with the 3080 if the halo product being the 3090 needs to be so big (cores wise) in order for it to be competitive with the out going tech, and process failure would mean there are plenty of non-perfect GPU's left over for the lower down part.
 
Think I'm settling on the founders edition 3090 if I can get one. Better not be anymore than $1400 -.- you'd hope it would be £1049-1099 with the exchange rate.

It does work out at £1050 converted, but then you have to add 20% VAT
 
Darn it the 3080 doesn't have enough ram for my gaming, thats a second generation of card with not enough ram as my 1080Ti :/

Rumour has it AIB models will offer a 3080 in a configuration of either 10Gb or 20Gb.
It's a shame they didn't think to just go to straight to 16Gb for the 3080 and just leave it at that.

At this time I have pretty much ruled the 3090 out as the cost including an PSU upgrade is just not worth it (imho).

But I am more interested in the 3080 (20Gb).
As spec wise that is a closer successor to the 2080ti than the 3090 which is basically a cheaper version of the Titan.

And of coarse I'm still open minded about a 16Gb RDNA2 card, as priced right that could be real temping.
 
Last edited:
It is not the clock speeds. It is not the memory. It isn't anything else other than the poopo Samsung GPU die.

This has been common knowledge for over a year.

I kinda feel like Roddy Piper in "They Live" sometimes. You know? where he tries to make the guy put on the glasses so he can see what is going on?

https://www.youtube.com/watch?v=NTGkW9cRUKI

Pls. Watch.that.video.

This isn't a recent discovery. He has been saying the same for over a year about it. Everything you will ever need to know will be in that video, and has been for two weeks. All of the process issues, the processes themselves, the power draw, ETC ETC ETC.

Thanks I had seen it but watched it again. To be fair, most of the conclusion is sprinkled in only a few sentences in that video.

So you think it's on Samsung 5nm(7nm) (previous leaks had suggested it was on Samsung 8nm(10nm) and there was speculation that the 7nm parts were still being manufactured at TSMC).

Ok that's fine, makes sense then.

Like AdoredTV says, no one really cares if the card consumes more power if it blows away previous generation performance.

To conclude from that that the Samsung die is poopo is a bit of a stretch. Poopo compared to what? We're talking 350 W against previous gen 250 W and many people have run the 2080 Ti in SLI anyway.

Despite the smaller node, and poopo process, if the rumours are true the new 3090 has twice as much memory, running on bigger and faster bus, and with 20% more cores than the 2080 Ti running at a faster clock. So it's 100W more power, does that really matter? Feels not unreasonable to me.
 
ZOTAC-RTX-30-Series.jpg



1933 Boost Clock? 45c temp?? Seems the leaks are coming boys!
 
I think the core is having yield problems,and the RTX3090 just has the best binned cores,but the GA102 in the RTX3080 is going to closer to what is actually doable in volume. It is nearly 20% cutdown.

Just because it's cut down does not mean all disabled cores had defects - it's just transistors can use more power than some more luckier ones, so you can't use all cores at same high frequency even if there were no defects, that's why (I think) power usage isn't that massive between two variants of the same chip.
 
Remember that's only the 3080. The 3090 may not be as well behaved. Mind you, its cooler is much bigger so hey there may be some surprises :)
 
Back
Top Bottom