NVIDIA GeForce Ampere Architecture, Board Design, Gaming Tech & Software

[ZoneDymo]

keep in mind that nearly all claims of "2x performance" etc come from RTX scaling, most games still are resterized for the most part and I doubt the RTX3070 beats a 2080Ti there

 

[Jism]

when are the actual reviews coming out and these fluff articles end, want to see it in action.

You need to wait because of "NDA". That does'nt mean sites like TPU can write about the architecture.

It now depends on what AMD brings to the table, and how well RDNA2 can deliver.

 

[Dyatlov A]

keep in mind that nearly all claims of "2x performance" etc come from RTX scaling, most games still are resterized for the most part and I doubt the RTX3070 beats a 2080Ti there

Nvidia told by himself 3070 beats 2080ti. Also watching the TDP numbers, 3070 almost eats as much Power as 2080ti, considering the new more efficient Ampere architecture 3070 certainly beats 2080ti.

 

[ebivan]

keep in mind that nearly all claims of "2x performance" etc come from RTX scaling, most games still are resterized for the most part and I doubt the RTX3070 beats a 2080Ti there

Well, its only from an nvidia employee and not from an objective review by a third party, but in that reddit q+a the employe answered that the claim is valid for non RTX mesurements too:

https://www.reddit.com/r/nvidia/comments/iko4u7/_/g3qkrzk
But we will know when the NDA is lifted. Is there a different NDA for each 70, 80 and 90 or will we get the full coverage when the 3080 is released in two weeks?

 

[wolf32v]

I agree no need to upgrade the CPU, at 4k its not the CPU but the Video card that make the difference. I am waiting for the dust to settle with all the hardware launches. AMD's new chip is just on the horizon could be a game changer that we all been looking forward too. Plus big Navi is on the way want to see what its does before I do anything. I am no fan boy just want the best for my money. My 1080TI is doing fine still, so I can wait and see what they both have to offer before jumping the gun. Plus when the games start being programed for multi core CPU's then the Intel's will be left in the dust. With both gaming consoles in AMD's corner it may be time for change. I will wait and see, its only a month or two till we see all the cards all three companies have to play. Then I will make my choices and with Nvidia making us wait and not giving us the real world test upfront, got me thinking maybe a little to much hype going on. We will see in the next few weeks I can wait till then. Anyway Nivida's would never use a AMD as a test system with them being their main competition.

 

[kiriakost]

The industry of 3D animation... you mean like the industry of entertainment, music and video production... modding... coding for the lulz

Professional work is being done by amateurs across the globe. The only difference with professionals is they make a good paycheck on it.

We saw with the emergence of streaming and Shadowplay that suddenly the "Let's Play" was the new guaranteed source of Youtube income. This has only gotten bigger, and movies built in game engines aren't new either and are really the modder' and 'tubers' territory. And let's face it... there are some pretty big crowds going for that content. What else is entertainment other than something that manages to draw attention of a lot of people? 3D animation as it is done now, perhaps is far more static than having the crowd get creative with it. Whole new formats can emerge.

Even just screenshots - we saw a new form of photography emerge with Nvidia Ansel, as well. People will elevate it to an art form.

My next question will be, if the in game 3D content this is copyright protected?
Because if it is ? then no one can use in game 3D content for production of material which this could be freely distributed on-line.

It is much different of some one to taking footage ( a video ) of him playing the game (=demonstration).

 

[steen]

You need to wait because of "NDA". That does'nt mean sites like TPU can write about the architecture.

Marketing slide deck embargo is up. Ryan Smith should have a tech article on the gaming arch white paper & tech brief soon. Interesting stuff regarding concurrent alu/rtx/tensor operation.

It now depends on what AMD brings to the table, and how well RDNA2 can deliver.

The new GDDR6X memory narrows the gap between GDDR and HBM standards for the first time, without insane increases in bus width.

We've got a fair idea...

 

[kiriakost]

Listen to this guy. He's related to those that invented democracy

And I love how Nvidia drew that airflow diagram as if the card is to be used in a CPU-less computer

All things considered, Ampere landed about where I predicted it would. Which makes me wish I predicted it was cheaper. It's still good value for the money (fuggidabout 3090), but 3080 at $500 would have been dreamy.

We did invent democracy among with schools and universities.
We did that so any one called as doctor him to be a certified too.

But because this board it is read by all ages, it must be repeatably clarified for the ones under the age of 18, that all new high priced hardware this is mostly a tool for 3D development, and it can become also a toy exclusively for a few that they can afford it.
In other words, NVIDIA this will make it first income from those cards by the money of the developers.
Then the developers they will need time so them to debug their own work, then NVIDIA will also need time so to debug their own drivers.
Eventually in eight months from now the gamers will start to enjoy the first fruits of this new tree.

 

[ebivan]

We did invent democracy among with schools and universities.

Sorry, it a little OT, but I have to say it.
You mean good old Greek democracy, where free adult male citizens would decide stuff, while the majority of the population (women, slaves, kids) did not?

 

[bogami]

Much new or not, the Micron with the new RAM elements has basically done the most here. Power outlets will be a problem and customers will buy contract versions. . Maybe they will offer interfaces next to the cards.

 

[kiriakost]

Sorry, it a little OT, but I have to say it.
You mean good old Greek democracy, where free adult male citizens would decide stuff, while the majority of the population (women, slaves, kids) did not?

Yes this is how it should be, the vote it should be delivered only by them qualifying.
qualifying = spending some of their time to watch their parliament for an hour daily, even due a TV channel.

 

[BiggieShady]

Hey @W1zzard , on page 4 heading there's an error calling 2-nd generation RT core 'The new Turing RT Core and Tensor Core',

 

[Caring1]

We did invent democracy among with schools and universities.

Bullshit, you created a word for it, Democracy was around long before the Greeks.

 

[kiriakost]

Bullshit, you created a word for it, Democracy was around long before the Greeks.

Thanks for the strong laughs, I will simply agree with you my King, by this way my head will stay in place. Lol

 

[Dirt Chip]

NV has realy pull all the triggers on this one, seems like an "all in" move.
Very impressive lunch and feature set that surly need review conformastion (I believe they will hold).

AMD MUST lunch their utmost best line ever to cope with AMPERA.
But If we considering the past of AMD GPU lunch, we must lower your expectation.

 

[Vya Domus]

Some execution units are more equal than others it seems. So that's another way they managed to fit that many shaders, some can do integer math and some can't. I guess you don't really need the same throughput.

 

[W1zzard]

Hey @W1zzard , on page 4 heading there's an error calling 2-nd generation RT core 'The new Turing RT Core and Tensor Core',

Fixed, thanks!

 

[Manoa]

this PCB will be hot, the high heat amount is a problem of itself (300+) but becuase the PCB so small you also will have alote of density, I think will be a good idea for anyone air cooling to wait custom PCB designs

 

[Dyatlov A]

this PCB will be hot, the high heat amount is a problem of itself (300+) but becuase the PCB so small you also will have alote of density, I think will be a good idea for anyone air cooling to wait custom PCB designs

But Nvidia put lots of effort to make a good cooling for these GPUs and i don’t think the custom PCB designs will be better, if not worse.

 

[sergionography]

Yeah really interested to see what people can create with Omniverse Machinima, certainly a fun tool to play around with either way.

Anyway, nice write up, all those apparent leaks and the core counts were still wrong.

PC devs actually would like nothing more than "boring", as they don't need to learn how to optimize all over again.


On another note, the pricing is decent indeed, but the performance expectations people have seem unrealistic. We haven't see benchmarks and I doubt we will see the numbers that Nvidia touted. I feel like people aren't putting enough emphasis on the cuda cores being different. Each SM has twice the number of cuda cores as Turing, but not twice the performance. Because half of the cores in the SM are shared for FP32 and INT32. The charts show the 3070 a bit above the 2080ti, which is where I would expect it to be. The best way I would do the math compared to Turing is divide the core count by 2 and then add up to 50% over it or so. So roughly I'd expect a 5888 core 3070 to be equivalent to a Turing card with 4416 cuda cores (in terms of rasterization). The main benefit of Ampere over Turing is having less penalty when using ray tracing

 

[Anymal]

@W1zzard Without saying yes or no (thus breaking a NDA or contract), are you currently in possession of any samples of the RTX 3080 or 3090 from NVIDIA?

1, not 0!

 

[steen]

Anyone remember FCAT? Now we have PCAT, for competitive analysis...

Got one @W1zzard? Plugin for FrameView?

Edit: I'll answer my own question. Frameview 1.1 now supports PCAT & LDAT. Very useful. It's interesting that end-to-end latency is now a big marketing deal.

 

[Cheeseball]

1, not 0!

It's worth a shot asking LOL

 

[Steevo]

Seems like Nvidia have a winner.

 

[Dyatlov A]

Fixed, thanks!

Dear W1zzard, can you test also with smaller PSUs as what is recommended? I currently have Corsair SF450 Platinum Plus and afraid with 3080 will explode, also 3070 questionable. Or if undervolting possible?
And if you could also write the size, the thickness i do not find at Nvidia website, just 2 slot card. And a good old 3Dmark11 score would be awesome too.