Saturday, June 6th 2020

NVIDIA GeForce RTX 3080 Pictured?

Here are what could be the very first pictures of a reference NVIDIA GeForce RTX 3080 "Ampere" graphics card revealing an unusual board design, which is the biggest departure in NVIDIA's design schemes since the original GeForce TITAN. It features a dual-fan aluminium fin-stack cooler, except that one of its fans is located on the obverse side, and the other on the reverse side of the card. The PCB of the card appears to extend only two-thirds the length of the card, ending in an inward cutout, beyond which there's only an extension of the cooling solution. The cooler shroud, rather than being a solid covering of the heatsink, is made of aluminium heatsink ridges. All in all, a very unusual design, which NVIDIA could implement on its top-tier SKUs, such as the RTX 3080, RTX 3080 Ti, and in a cosmetic form on lower SKUs. We get the feeling that "Cyberpunk 2077" has influenced this design.
Sources: ChipHell Forums, HXL (Twitter), VideoCardz
Add your own comment

225 Comments on NVIDIA GeForce RTX 3080 Pictured?

#51
efikkan
ARF
If legit, less VRAM than an RTX 2080 Ti with similar specs. Performance should be around RTX 2080 Ti, too.


wccftech.com/nividias-rtx-3080-getting-spec-upgrade-might-ship-with-ga-102-200-and-4352-cuda-cores/
That's a very bold prediction. Do you know the performance of the Ampere architecture and the clock speeds it will be running?
And BTW, more VRAM doesn't necessarily yield more performance. More VRAM allows you to run with higher details.

I would be more worried about memory bandwidth. Are there faster than 18 Gb/s GDDR6 yet?
Posted on Reply
#52
cucker tarlson
efikkan
That's a very bold prediction. Do you know the performance of the Ampere architecture and the clock speeds it will be running?
And BTW, more VRAM doesn't necessarily yield more performance. More VRAM allows you to run with higher details.

I would be more worried about memory bandwidth. Are there faster than 18 Gb/s GDDR6 yet?
a 102 die that goes from 5370 to 4350 ? we ever had a big die sku with 20% disabled ? they don't even do this much on 104s.
Posted on Reply
#53
nguyen
Looks like 7nm yield hasn't reached maturity parity with 12nm that Nvidia decide to use the GA 102 die for 3 SKUs (Titan, 3090 and 3080) while TU 102 is only used for Titan RTX and 2080Ti.
Let just hope there are enough functioning cores for 3090 and Titan...
Also hope the new PCB design wont delay waterblock release...
Posted on Reply
#54
efikkan
cucker tarlson
a 102 die that goes from 5370 to 4350 ? we ever had a big die sku with 20% disabled ? they don't even do this much on 104s.
We have seen similar things, like TU104s sold as RTX 2060 (KO) etc., but it's rare.
Since we have so many conflicting rumors about these specs, at most one of them can be true, possibly none of them, so I'm not going to assume any of them are true until we have something more substantive.

But anyway, in my opinion the best reason for using a "GA102" as "RTX 3080" would be to achieve extra memory bandwidth.
Posted on Reply
#55
EarthDog
Damn thats fugly...come onnnnn board partners!
Valantar
That die and VRAM layout looks unrealistically close to the display outputs. There would be no room for the circuitry necessary to make the outputs work unless they're doing something different
FTFY. :p

That said, I dont think it's too close at all...
ARF
Let's hope this is relatively slow and Navi 21 completely obliterates it.
Where do you get your stuff at? Me want!
Posted on Reply
#56
cucker tarlson
efikkan
We have seen similar things, like TU104s sold as RTX 2060 (KO) etc., but it's rare.
Since we have so many conflicting rumors about these specs, at most one of them can be true, possibly none of them, so I'm not going to assume any of them are true until we have something more substantive.

But anyway, in my opinion the best reason for using a "GA102" as "RTX 3080" would be to achieve extra memory bandwidth.
turing already achieved really good results with 8gb
at 4K 2070s seems to hold its own against 11gb 1080Ti
www.techpowerup.com/review/nvida-geforce-rtx-2070-super/27.html
Posted on Reply
#57
Decryptor009
cucker tarlson
how can you tell performance ?
new 80s always beat old 80Ti's with less cuda.sometimes by a lot when there's a new node.1080 > 980Ti by 25%
25% is nothing, back to Fermi, GTX 480 was ridiculously faster, 5870 was roughly on par with 2 4870's too.

Posted on Reply
#58
cucker tarlson
Decryptor009
25% is nothing, back to Fermi, GTX 480 was ridiculously faster, 5870 was roughly on par with 2 4870's too.


cause it was a bigger die on a smaller node,nothing magical.30% bigger.
they're not gonna make a 1000mm2 gpu now
and apples to apples,1080Ti at 470mm2 was almost 70% faster than 600mm2 980Ti.
EarthDog
Damn thats fugly...come onnnnn board partners!

FTFY. :p

That said, I dont think it's too close at all...

Where do you get your stuff at? Me want!
ugly or not,their constant progress with reference coolers drives the prices of aib coolers down.
let's hope amd don't embarass themselves with that friggin blower again.
I'll prolly go with trio again.great on 1080Ti,even better on 2070S.
Posted on Reply
#59
ARF
Decryptor009
25% is nothing, back to Fermi, GTX 480 was ridiculously faster, 5870 was roughly on par with 2 4870's too.
It's very weird why AMD did what they did with Navi 1st generation.
Why no Big Navi and why not competing with Vega 64/ Vega 64 derivatives in the below Big Navi segments.

They had a 256 mm2 HD 4870 but back then they had a card with 70-80% higher performance.

Now they have nothing.

And if the rumours are true, the smaller Navi 2X chips won't be out before 2021.
Posted on Reply
#60
cucker tarlson
ARF
It's very weird why AMD did what they did with Navi 1st generation.
Why no Big Navi and why not competing with Vega 64/ Vega 64 derivatives in the below Big Navi segments.

They had a 256 mm2 HD 4870 but back then they had a card with 70-80% higher performance.

Now they have nothing.

And if the rumours are true, the smaller Navi 2X chips won't be out before 2021.
you enjoy rumors,don't you ?
Posted on Reply
#61
Valantar
TheLostSwede
It's from some "guy on the internets" so I wouldn't read too much into it.
That said, I don't think it's too hard to make a V shaped VRM layout.
Not talking about the VRM, just all the other stuff that needs to go between the die and the I/O.
Posted on Reply
#62
nguyen
ARF
It's very weird why AMD did what they did with Navi 1st generation.
Why no Big Navi and why not competing with Vega 64/ Vega 64 derivatives in the below Big Navi segments.

They had a 256 mm2 HD 4870 but back then they had a card with 70-80% higher performance.

Now they have nothing.

And if the rumours are true, the smaller Navi 2X chips won't be out before 2021.
Easy answer, yield for 7nm was terrible (5600XT-5700-5700XT all use navi 10 die), it's still terrible now for bigger die, plus their CPU designs are winning against Intel, why waste precious 7nm fab when they know their high end GPU would never sell as well as Nvidia.
Posted on Reply
#63
TheLostSwede
Valantar
Not talking about the VRM, just all the other stuff that needs to go between the die and the I/O.
It's actually not all that much that goes there these days.
I see something that looks like a couple of chokes and what might be some signal amplifiers between the GPU and the I/O, the rest looks pretty empty to me.

www.techpowerup.com/review/evga-geforce-rtx-2070-super-ko/3.html

The reference 2080 has a few more bits there, but not much. I guess that's due to the USB-C port.

www.techpowerup.com/review/nvidia-geforce-rtx-2080-founders-edition/5.html
Posted on Reply
#64
Valantar
TheLostSwede
It's actually not all that much that goes there these days.
I see something that looks like a couple of chokes and what might be some signal amplifiers between the GPU and the I/O, the rest looks pretty empty to me.

www.techpowerup.com/review/evga-geforce-rtx-2070-super-ko/3.html

The reference 2080 has a few more bits there, but not much. I guess that's due to the USB-C port.

www.techpowerup.com/review/nvidia-geforce-rtx-2080-founders-edition/5.html
Yeah, that's a lot less stuff than I'm used to seeing in that area. Guess that solves that problem, even if fitting the power delivery and other circuitry for the VirtualLink would still be an issue I guess that could be squeezed in below the VRAM.
Posted on Reply
#65
cueman
erhh, when ampere are tested no1 took big navi, or any amd gpus,rtx 3060 is enough and when and if radeon viii aka big navi coming, and if it coming, same thing ,belive it...or not.

fact are, amd use still 7nm process tech without any new arch or updates and top on that hill is ,still 7nm radeon vii gpu, rx 5700 xt little brother is same gpu core with less tmus and rops but clocks maxed what is possible when hardware gadgets taken off.

so only way to get more power radeon vii(i)/big navi is put it inside more tmus and rops...thats all...but if so, except tdp 400W and watercool what i say that if happends ,its 7th wonder for 7nm gpu.
finally, lisa Su says 'we have big navi,sure, nice to ask it,yes its coming..' this she says only few month ago...i mean time,capice. so its it is radeon vii(i).


ampere leaps are process tech from 12nm to 7nm and new core arch and new cudas innovation cuda 3... any1 need more facts?

stop dream and stop took big navi.. its coming or not,but 7nm ampere coming 100%....just waiting and see.

my opinion is that desktop gpus max tdp should be 200W. if over 100-200 $ tax on it,different is science and other industrial kind using gpus.


naah,, lets joy summer and wait interesting autum with 7nm nvidia ampere (1st 7nm!), 10nm intel cpu (1st under 14nm cpu!), intel 7nm gpu (1st!) and amd erhh, ok, big navi.
Posted on Reply
#66
Dante Uchiha
It seems that the cooling system is optimized for SLI ? Anyway, I think it's pretty ugly compared to the 2XXX FE model. :p
Posted on Reply
#67
Mats
Dante Uchiha
the 3XXX FE model
Do you have a pic?
Posted on Reply
#68
Dante Uchiha
Mats
Do you have a pic?
Sorry. I just noticed the confusion, I meant 2XXX. XD



Posted on Reply
#69
dyonoctis
But now I wonder how the 3060 cooler is going to look like. The pcb is already really short on that 3080, I don't see how they can shave 4 cm of that...
Posted on Reply
#70
ARF
nguyen
Easy answer, yield for 7nm was terrible (5600XT-5700-5700XT all use navi 10 die), it's still terrible now for bigger die, plus their CPU designs are winning against Intel, why waste precious 7nm fab when they know their high end GPU would never sell as well as Nvidia.
7nm yields have been great www.overclock3d.net/news/cpu_mainboard/amd_is_reportedly_achieving_great_yields_on_their_zen_2_cpu_dies_but_not_as_good_as_last-gen/1?fbclid=IwAR31VUNiuZOF5LuVrfanB-mziRJSJIldcTrB7PwAXx0u2oXyxo9K8BhtICY

5nm yields are even greater www.anandtech.com/show/15219/early-tsmc-5nm-test-chip-yields-80-hvm-coming-in-h1-2020
wccftech.com/amd-zen-4-5-nm-launching-2021/


Big Navi wouldn't have used too much capacity, it would have been enthusiast level with lower quantity of sales.
Posted on Reply
#71
nguyen
ARF
7nm yields have been great www.overclock3d.net/news/cpu_mainboard/amd_is_reportedly_achieving_great_yields_on_their_zen_2_cpu_dies_but_not_as_good_as_last-gen/1?fbclid=IwAR31VUNiuZOF5LuVrfanB-mziRJSJIldcTrB7PwAXx0u2oXyxo9K8BhtICY

5nm yields are even greater www.anandtech.com/show/15219/early-tsmc-5nm-test-chip-yields-80-hvm-coming-in-h1-2020
wccftech.com/amd-zen-4-5-nm-launching-2021/


Big Navi wouldn't have used too much capacity, it would have been enthusiast level with lower quantity of sales.
Zen 2 die are only 70mm2 vs Navi 10 251mm2. If you only get 70% yield with a 70mm2 chip, imagine the yield at 500mm2.
Posted on Reply
#72
ARF
nguyen
Zen 2 die are only 70mm2 vs Navi 10 251mm2. If you only get 70% yield with a 70mm2 chip, imagine the yield at 500mm2.
The costs are normally transferred to the consumers, anyways.
Halo products like Big Navi are very important in order to show to the consumers that the company can in reality make cutting-edge products, not just entry-level and mid-range. Halo is important to keep a recognisable brand reputation and awareness.

Also, it'd have very positive effect on the overall competitiveness and pricing structure. RX 5700 and RX 5700 XT were launched too expensive.

AMD should have launched the Navi 10 with proper voltage at 150-watt, and then build the product stack up and down as the wish.
Instead, there is just super duper expensive Navi 10 cards with crazy overclocks reaching staggering 250-300-watt TDP.


Competition but real one in order to keep the prices in check.
Posted on Reply
#74
Mats
Seems like the front exhaust will feed the rear fan with hot air. If the fans were on the same side then I guess the heat from the front exhaust would go directly to the CPU area..
Posted on Reply
#75
Vayra86
ARF
The costs are normally transferred to the consumers, anyways.
Halo products like Big Navi are very important in order to show to the consumers that the company can in reality make cutting-edge products, not just entry-level and mid-range. Halo is important to keep a recognisable brand reputation and awareness.

Also, it'd have very positive effect on the overall competitiveness and pricing structure. RX 5700 and RX 5700 XT were launched too expensive.

AMD should have launched the Navi 10 with proper voltage at 150-watt, and then build the product stack up and down as the wish.
Instead, there is just super duper expensive Navi 10 cards with crazy overclocks reaching staggering 250-300-watt TDP.


Competition but real one in order to keep the prices in check.
Seriously man just stop talking because every post you try to talk yourself out of your own BS is diggin a deeper hole. It happened with browsers, and its happening here. Why, I wonder... You can also just admit you don't know jack squat and start learning. This forum is full of people ready to educate. But when they do so, you stick to your ideas - with no basis.

Its not like the facade is working either. ;)

At its core you're not wrong about halo product... but the learning point here is about yields and fabs, how they work and why they work like they do. Its great to understand that because chip size and all that are very strong bits of info to gauge what'll happen next, and why companies do what they do. Simple rule of thumb, no matter what news you read: the larger the die, the lower the yield. This in turn automatically, always means that smaller chips are far easier to make cost effectively. Big chips are extremely difficult to make without losing money, or pricing them out of this world. Examples everywhere. 2080ti; Turing pre Super even was expensive across the whole stack. Even the Supers are a result of better yields. That wasnt just lack of competition. It was a series of big dies with little to show for it. This is also why Navi 1st gen was pretty good for AMD. The die wasn't too big. And its why GCN was long overdue for a shrink or overhaul: dies got too large and competitor could do the same with smaller ones. The gap was even large enough that Nvidia could pull a Turing on 12nm and still come out winning.
Mats
Seems like the front exhaust will feed the rear fan with hot air. If the fans were on the same side then I guess the heat from the front exhaust would go directly to the CPU area..

As I see it, this card will exhaust much like a blower, likely having the usual ventilation holes at the outputs- end of the card, and possibly the opposite end? Like a double radial blower, kind of idea. That makes for two very short paths for air to go through and avoids any sort of turbulence from multiple directions of airflow. Just two separate chambers, really. Given the location of PCB, one of those fans is just there to cool heatsink and the other is blasting away over the die.

Both fans are definitely intakes.
Posted on Reply
Add your own comment