Nvidia Ampere Discussion [2020-05-14]

[DegustatoR]

Wouldn't there be a safety issue with regards to power draw? In theory the maximum for 2x8pins is 300w (150w each). It's presumed part of the reason here is that the draw will need to exceed that.

If they plan to use 8-pin to 12-pin converter then this 12-pin can't use more than 300W through this connector without breaking the spec. 3x8 to 12 makes no sense from electrical perspective I'd say. That being said most single rail PSUs will likely be able to carry more power than 300W over 2x8-pins and it remains to be seen how Ampere will use this power config.

Partner cards will probably use 3x8-pins because they will be factory OCed and consume more than FE by default.

So perhaps it'll need to be 4x8pins.

Even less likely as that would need 24 wires total to work.

 

[Malo]

I didn't know the 8-pin wasn't standardized on the PSU. I've never had issues with using those cables across different brands.

 

[SlmDnk]

Am I the only one who looked at this image and saw a city instead of a board?

 

[trinibwoy]

Surely any included adapter will connect to the end of the 8-pin cables and not directly to the PSU.

 

[Malo]

Surely any included adapter will connect to the end of the 8-pin cables and not directly to the PSU.

Yeah they'll just be short Y-splitters I assume.

 

[Benetanegia]

Then there's also the issue on the PSU end. Given the Seasonic example it seems like at least they feel you need x2 PSU end connectors.

Yes, of course on their end they feel that, because the spec for 12-pin connector is 600W, so a 4x 8-pins equivalent is needed to match that. That means nothing in regards to the power draw of the card or 2x8-pin to 12-pin being more than enough for the card.

 

[Kaotik]

Am I the only one worried about the mechanical stress users will put on that microfit connector which is standing on it's short end?

 

[BRiT]

Am I the only one worried about the mechanical stress users will put on that microfit connector which is standing on it's short end?

Wait a minute. You mean that's how it's actually going to be connected in reality? It won't be placed flat on the long side to provide more solid connection to the PCB? If it's using the short side.

 

[DegustatoR]

Am I the only one worried about the mechanical stress users will put on that microfit connector which is standing on it's short end?

It will probably be reinforced by the cooler around it or something. I'm more worried about all the dust which may collect over the open wires on its back. Hopefully these will be protected with some cover / isolation.

 

[SimBy]

Wait a minute. You mean that's how it's actually going to be connected in reality? It won't be placed flat on the long side to provide more solid connection to the PCB? If it's using the short side.

Wouldn't that defeat the purpose of saving every tiny bit of space?

 

[BRiT]

Wouldn't that defeat the purpose of saving every tiny bit of space?

Surely they saved enough space by not having the larger standard dual connections?

 

[Malo]

Am I the only one worried about the mechanical stress users will put on that microfit connector which is standing on it's short end?

My assumption is that it's reinforced on both ends by the top shroud. Likely something behind it to provide support.

 

[Leoneazzurro5]

Latest leak on Videocardz show 350W TGP for the 3090 (boost clock just short of 1,7GHz) and 320W for the 3080 (1710 Mhz)

 

[Krteq]

They are also reporting 7nm fabrication process

The data that we saw clearly mention the 7nm fabrication node. At this time we are unable to confirm if this is indeed true.

VCZ - NVIDIA GeForce RTX 3090 and GeForce RTX 3080 specifications leaked

 

[DavidGraham]

Anybody who thought NVIDIA will do all the work to produce A100 on 7nm TSMC, and then dump all that on another completely different node for Geforce Ampere is clearly thinking in the most illogical and absurd way possible.

 

[Rootax]

Anybody who thought NVIDIA will do all the work to produce A100 on 7nm TSMC, and then dump all that on another completely different node for Geforce Ampere is clearly thinking in the most illogical and absurd way possible.

Before we knew about A100 on 7nm TSMC, Yoo Eung-joon said that futur gpus would be produce with Samsung (http://www.koreaherald.com/view.php?ud=20190702000692&ACE_SEARCH=1 ). So my guess is at one point they make a design on a Samsung process. The samsung name did not fall from the sky. And if one company has the money to make a chip with two different process, it's nVidia.

 

[Leoneazzurro5]

The same article also states the 7nm EUV process for Samsung.

 

[SimBy]

If it's indeed on TSMC 7nm the power consumption and clock speeds are even less flattering. That would mean Nvidia basically regressed in perf/W despite the node switch. I find that hard to believe.

 

[Nebuchadnezzar]

Anybody who thought NVIDIA will do all the work to produce A100 on 7nm TSMC, and then dump all that on another completely different node for Geforce Ampere is clearly thinking in the most illogical and absurd way possible.

Samsung won Nvidia as a customer, that's a fact and there's plenty of proof around there. If this ends up being Samsung 7nm then your post here will look silly.

 

[Kaotik]

Anybody who thought NVIDIA will do all the work to produce A100 on 7nm TSMC, and then dump all that on another completely different node for Geforce Ampere is clearly thinking in the most illogical and absurd way possible.

How much does A100 really share with GA-series? SM units will definitely have major differences, like GA most likely missing DP and there will at least be less Tensor units, possibly different ones too, then there's many parts of GPU completely missing from A100 since it's not used for graphics and so on.
I mean, it's far more logical split than just performance split somewhere in the GPU lineup (personally not taking any stance on where the split between TSMC/Samsung will end up, just that it's not that illogical to think A would be TSMC and GA Samsung, definitely not most illogical and absurd possible way.)

Before we knew about A100 on 7nm TSMC, Yoo Eung-joon said that futur gpus would be produce with Samsung (http://www.koreaherald.com/view.php?ud=20190702000692&ACE_SEARCH=1 ). So my guess is at one point they make a design on a Samsung process. The samsung name did not fall from the sky. And if one company has the money to make a chip with two different process, it's nVidia.

Jensen Huang clarified later that most of their GPUs will still come from TSMC