• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Ampere coming in hot.....

Orta

Banned
No wonder it's fucking huge

NVIDIA-GeForce-RTX-3090-Ampere-Flagship-Gaming-Graphics-Card_1.jpg

Bph-Mep2-Ig-AAXHw1.jpg
 

The Skull

Member
Can some expert explain to me in leyman words why series 3000 is so big and hot and power hungry?

Isn't the power jump the usual 30% of every new gpu series? Is it caused by the new improved rtx?

The rumors are that big Navi is around 30-40% ( a bit optimistic for my liking ) faster than a 2080ti and that Nvidia refuses to lose the performance crowd, so they have just busted out the Titan (3090?) at launch and pushed it to the absolute max. I don't think the 3080 and lower should be this big or power hungry but there are rumors that Nvidia are using Samsung's 8nm process, which isn't particularly great, so who knows at this point. I believe the performance increase this gen should be quite a bit bigger than previous ones but potentially at an increase in power draw.
 

Ellery

Member
Yooo.

You have way way too much faith in Big Navi; 20 percent performance gain over the 2080TI.?
Nada, my prediction it only just beats the 2080ti and doesnt even get within spitting distance of the RTX 3080.

AMD best chip will be fighting with the RTX 3070, hopefully they price it right so the 3070 drops in price.

I think there is a world in which we are both on the right track. Maybe I am underestimating the power of the 30 series and it is actually a much bigger jump than I expect it to be. Would maybe even make sense given the rumors of the gigantic coolers and expected power draw (much much higher than turing. Like 350W or something).
Maybe the 3070 is actually better than the 2080 Ti, the 3080 much better and the 3090 completely obliterates the 2080 Ti by like 80% or so. (heavily speculation and I doubt it is going to be that big. I think RTX performance is getting a gigantic jump though)

Though AMD has to have something. The only way I could see them having troubles with the RTX 2080 Ti is if they heavily overstated the performance gains from Navi to Navi 2.

Like the 5700 XT is not even a flagship product and it is pretty good. It is a 251mm² die or something. It is a smaller chip than the GTX 1660 Ti (though that is not really a fair comparison since it is a different manufacturing node).

AMD is going to give us something. The difference between 5700 XT and 2080 Ti is smaller than people think. It is like 35-50% ~ and AMD could easily make a Frankenstein Big Navi card that is at least 500mm² (or bigger) and has like double the amount of transistors compared to the 5700 XT. Things don't scale linearly, but that alone should put it up against the / or even above the 2080 Ti. And then we add architectural improvements on top of it and we have a card that would be above the 2080 Ti.

ylKmXis.jpg


If AMD doesn't manage to do that then I am heavily disappointed.
 
Can some expert explain to me in leyman words why series 3000 is so big and hot and power hungry?

Isn't the power jump the usual 30% of every new gpu series? Is it caused by the new improved rtx?

A few reasons:

- It's made on Samsung 8nm process node, not TSMC 7nm
- 24GB of memory drinks a lot of power and produces heat
- They've had to push clocks, presumably to compete with Big Navi and/or hit the generational uplift Nvidia fans expect (+40% flagship to flagship).
- RT performance is impressive and a big leap from Turing, but this adds more power consumption/heat.
 

FireFly

Member
Yooo.

You have way way too much faith in Big Navi; 20 percent performance gain over the 2080TI.?
Nada, my prediction it only just beats the 2080ti and doesnt even get within spitting distance of the RTX 3080.

AMD best chip will be fighting with the RTX 3070, hopefully they price it right so the 3070 drops in price.
If they're doubling up the number of CUs, as rumoured, then a 50% performance increase would be very disappointing. We also have this early VR benchmark showing *something* substantially beating an (IIRC) overclocked 2080 Ti

 

sendit

Member
Will wait for the TI/Super with 20GB vram in another 6 months.

I mean, what games now needs a 3080?

Incorrect reasoning. There are graphic features available in games today that would eat up massive amounts of bandwidth. For example, Im sure FS2020 can cripple the 3080 or the 3090 given the right graphic settings.
 

Ellery

Member
One thing I have to add in regards to Big Navi is that, whether or not it does easily beat the RTX 2080 Ti, I think it is going to be late to the market. I expect it in like December and by that time Nvidia will have sold more 3080/3090 and 3070 than Lisa Su would like.
 

kittoo

Cretinously credulous
Unless RDNA 2 (Navi2x) can match rasterization performance and be somewhat close enough Ray tracing performance, ill be getting a 3080.
Just don't like the idea of the power/heat this thing will draw and generate. (Not a deal breaker though).
This makes so sense. are you saying that if Navi is good, you'll get Ampere (3080)?
 

RoboFu

One of the green rats
Well that sucks. Unless rdna 2 something special I guess I won’t be updating gpus anytime soon.
 

Kenpachii

Member
I already assumed this, i bought a few nvidia cards before that where hot same with AMD, memory modules always shit the brick on those cards. Never again.
 
Last edited:

Papacheeks

Banned
Yooo.

You have way way too much faith in Big Navi; 20 percent performance gain over the 2080TI.?
Nada, my prediction it only just beats the 2080ti and doesnt even get within spitting distance of the RTX 3080.

AMD best chip will be fighting with the RTX 3070, hopefully they price it right so the 3070 drops in price.

What if I told you it was close to 20-30% better than a 2080ti with higher scalable clocks? Much better efficiency as well and a regular sized card?
 

Papacheeks

Banned
Samsung 8nm and Big Navi on TSMC's superior 7nm are to blame for Nvidia's woes.

This is Fermi 2.0

Nvidia is to blame if you know the story correctly. They fucked up. It's not TSMC's fault that Nvidia tried to muscle their way into a deal, TSMC already at the time was at capacity. Instead of just biting the bullet and delaying they rushed to samsung on a worse node. 100% their fault not NAVI's, not TSMC but straight nvidia pulling a intel.
 

longdi

Banned
What if I told you it was close to 20-30% better than a 2080ti with higher scalable clocks? Much better efficiency as well and a regular sized card?
What? Are you insider?
Damn if so, i took a big punt in Nvidia stocks, expecting ampere to crush all sales as PCMR are hungry to upgrade
 
Last edited:

Papacheeks

Banned
What? Are you insider?
Damn if so, i took a big punt in Nvidia stocks, expecting ampere to crush all sales as PCMR are hungry to upgrade

at those leaked prices they are not going to crush anything. It's going to push people to buy consoles and hold off on upgrading till later next year when more AIB's and possibly a refresh happen on TSMC'S new refined process.
 
  • Thoughtful
Reactions: GHG

Dirk Benedict

Gold Member
This is disappointing. Only because I thought I'd be able to use this till my other parts arrive. It's going to be collecting dust for a few weeks.
 

McCheese

Member
I never understood the size of graphic cards. You have the series X system-on-a-chip which packs a CPU, 12TF GPU and media encoders on one square of silicon the size of a stamp; so why are these so huge?!?
 
Last edited:

RoboFu

One of the green rats
I never understood the size of graphic cards. You have the series X system-on-a-chip which packs a CPU, 12TF GPU and media encoders on one square of silicon the size of a stamp; so why are these so huge?!?

Consoles don’t have the overhead a pc does. They are specifically made to play games. They can’t do more than the hardware allows but they are much more efficient at getting the most out of it.
 

Black_Stride

do not tempt fate do not contrain Wonder Woman's thighs do not do not
I don't see it being over $900.

We are expecting the 3070FE to be ~600

I never understood the size of graphic cards. You have the series X system-on-a-chip which packs a CPU, 12TF GPU and media encoders on one square of silicon the size of a stamp; so why are these so huge, why don't manufactures make 'system-on-a-chip' PCs?

You mean like laptops?
35eq4UQqx4QRRQU2.jpg



And small systems?
80




Intel-Ghost-Canyon-NUC-and-Element-Modular-PC-Review-6.jpg


The thing labeled as A, is the whole computer actually.
 

Papacheeks

Banned
We are expecting the 3070FE to be ~600



You mean like laptops?
35eq4UQqx4QRRQU2.jpg



And small systems?
80




Intel-Ghost-Canyon-NUC-and-Element-Modular-PC-Review-6.jpg


The thing labeled as A, is the whole computer actually.

I was talking about Big NAVI. The biggest sku will be trying to aim to beat 2080ti performance. So pricing i am expecting it to be no more the $900 on the high side or watercooled sku.
 

FireFly

Member
I never understood the size of graphic cards. You have the series X system-on-a-chip which packs a CPU, 12TF GPU and media encoders on one square of silicon the size of a stamp; so why are these so huge?!?
They have to be able to dissipate 300W+ of heat, which is basically the output of an entire console, or more.
 

Three

Member
I never understood the size of graphic cards. You have the series X system-on-a-chip which packs a CPU, 12TF GPU and media encoders on one square of silicon the size of a stamp; so why are these so huge?!?
Well the SoC is just the SoC but you still have the VRAM and voltage regulators and things like that. On a graphics card you have to have a cooling block and fan. With a console you can split that up, specifically design for it or combine things like the RAM, cooling block and fans. On a graphics card you can't.
 
Last edited:

Celcius

°Temp. member
Cramming all these cores into cpu's and all this tech into gpu's is causing heat to start getting out of comfort for high-end parts on air these days...
 

Trimesh

Banned
Can some expert explain to me in leyman words why series 3000 is so big and hot and power hungry?

Isn't the power jump the usual 30% of every new gpu series? Is it caused by the new improved rtx?

This is a guess, but I suspect that nVidia are just pushing these chips to places they really don't want to go so they can have a high end SKU with a performance level that's sufficiently ahead of the top end AMD card - even if it means using 3x8-Pin aux power jacks and comedy sized coolers to do it.

If my theory is correct, I also suspect the high-end nVidia cards are going to have close to zero overclock headroom because they are already being pushed pretty much as far as they will go just to meet the published specs.
 

longdi

Banned
at those leaked prices they are not going to crush anything. It's going to push people to buy consoles and hold off on upgrading till later next year when more AIB's and possibly a refresh happen on TSMC'S new refined process.

phew at least you no inside info.
going to let my nvda run until sep1. :messenger_grinning_sweat:

but yes samsung 8nm seems worse, at least thats the results from their exynos soc.

also for $800, if the 3080 is as rumored 20-25% better than 2080ti, i think thats a win.
making it 50% faster than 1080ti. a lot of pcmr wants to upgrade from their 10 series gtx
 

Papacheeks

Banned
phew at least you no inside info.
going to let my nvda run until sep1. :messenger_grinning_sweat:

but yes samsung 8nm seems worse, at least thats the results from their exynos soc.

also for $800, if the 3080 is as rumored 20-25% better than 2080ti, i think thats a win.
making it 50% faster than 1080ti. a lot of pcmr wants to upgrade from their 10 series gtx

The info I have It's not worth sharing. Everyone will see how un-appealing these cards will be. And the will see AIB partners that may be absent with a nvidia variant cards in the coming months following the initial launch. Let's say right now some AIB's are not happy with nvidia right now. Similar to what happened to the 5700 launch.

AMD is being quiet a little as well, but is designing their board a little smarter for aib partners to make their cooling solutions much easier. Issue has been the PR people associated with Radeon previous releases, were pretty much gutted and have new group now.

What I can say is in terms of efficiency RDNA 2 will be better in terms of thermals and power usage, only on BIG NAVI with a frequancy boost will we see some high power usage. But that is their biggest card in the stack.

RDNA 2 is not worrying about nvidia in the sense they know in terms of value they will be the better choice.

If you have unlimited budget, and dont care about space or heat because your going to water cool or you have insane cooling solution in your case then yea nvidia will show performance crown for sure. But its going to come at large cost to consumer and it came at a terrible design cost.
 

Black_Stride

do not tempt fate do not contrain Wonder Woman's thighs do not do not
I was talking about Big NAVI. The biggest sku will be trying to aim to beat 2080ti performance. So pricing i am expecting it to be no more the $900 on the high side or watercooled sku.

Unless you think Big NAVI is going to be a RTX 3080+ competitor.
It would have to be priced at or around the RTX 3070 right.
 

Rickyiez

Member
phew at least you no inside info.
going to let my nvda run until sep1. :messenger_grinning_sweat:

but yes samsung 8nm seems worse, at least thats the results from their exynos soc.

also for $800, if the 3080 is as rumored 20-25% better than 2080ti, i think thats a win.
making it 50% faster than 1080ti. a lot of pcmr wants to upgrade from their 10 series gtx

An $800 rtx 3080 that is 30% faster is really a win, I don't know what people expecting its just a reality nowadays. We should expect much better RT performance too
 

Papacheeks

Banned
Unless you think Big NAVI is going to be a RTX 3080+ competitor.
It would have to be priced at or around the RTX 3070 right.

They have multi card stack ready to go. But what they launch first or what they focus on price wise sounds like 3080, so expect $7-800 with a higher clocked water cooled sku being in the $900 range.
 

Siri

Banned
An $800 rtx 3080 that is 30% faster is really a win, I don't know what people expecting its just a reality nowadays. We should expect much better RT performance too

Right now, in this thread, I don’t think anyone‘s too concerned about the numbers - but about the dimensions of the card itself and the amount of heat it‘s going to release.

Three slots? Three 8-pin connectors (for the 3080 TI), much larger than an RTX 2080 TI. I was eager to buy the 3080 TI - but it won’t even fit into my pc (if all this is true).
 

GymWolf

Member
This is a guess, but I suspect that nVidia are just pushing these chips to places they really don't want to go so they can have a high end SKU with a performance level that's sufficiently ahead of the top end AMD card - even if it means using 3x8-Pin aux power jacks and comedy sized coolers to do it.

If my theory is correct, I also suspect the high-end nVidia cards are going to have close to zero overclock headroom because they are already being pushed pretty much as far as they will go just to meet the published specs.
Are they really that worried about AMD?

What is this phantomatic amd gpu who is gonna battle with the 3000 series??
 

Blond

Banned
Rumored prices:


3060 -399$
3070 - 599$
3080 - 799$
3090 - 1399$

The gap between 3080 and 3090 is to big IMO, if the prices are indeed true I think NV will eventually fill the gap with with 3080Ti/Super.

Jesus Nvidia is really going out of their way to prove some point that you can buy a console for the same price as a single PC component.
 

nemiroff

Gold Member
If that's the real deal I actually need to check if I can fit it in my case..
 
Last edited:

diffusionx

Gold Member
I know why Nvidia wants to drop a monster GPU, with the new consoles coming around and what not ,but this confuses me. With a die shrink to 7nm and all this time for architectural improvements, I would think they could have gotten very large gains while also keeping power/thermals reasonable and not having to resort to FX5900 levels of nonsense.

No wonder it's fucking huge

NVIDIA-GeForce-RTX-3090-Ampere-Flagship-Gaming-Graphics-Card_1.jpg


Top: 6'0
Bottom: 5'11"
 
Last edited:

LordOfChaos

Member
I know why Nvidia wants to drop a monster GPU, with the new consoles coming around and what not ,but this confuses me. With a die shrink to 7nm and all this time for architectural improvements, I would think they could have gotten very large gains while also keeping power/thermals reasonable and not having to resort to FX5900 levels of nonsense.


If it's true that it's built on Samsung 8nm instead, that would be part of why. It would be delicious if they got cocky and tried to make bigger margins by going to a worse but cheaper fab and that lost them the performance crown, but AMD would have to come up a ways for that to happen.
 
This card will absolutely have great performance. Unfortunately it's on Samsung's 8nm node. Power hungry bastard it will probably be. You just know they already have the Super series in their back pocket once TSMC has capacity open on 7nm EUV. If I were gunning for the 3000 series, I would wait for Super. I'm waiting to see what they'll do in the 4000 series of cards based on current speculation.
 
  • Like
Reactions: GHG
Top Bottom