Photo of a NVIDIA GPU Engineering Board fitted with GDDR6 Spotted

Published by

Click here to post a comment for Photo of a NVIDIA GPU Engineering Board fitted with GDDR6 Spotted on our message forum
https://forums.guru3d.com/data/avatars/m/260/260048.jpg
Very interesting stuff, thank you Hilbert. In conjunction with DDR6 production ramping up, perhaps there is some degree of truth to all the recent hype.
https://forums.guru3d.com/data/avatars/m/258/258664.jpg
My thinking too. Might work with the time frame of a Nvidia launch later this year, BUT I could imagine them to even do refreshes with GDDR6 and slightly higher clocks, instead of something actually new.
https://forums.guru3d.com/data/avatars/m/266/266726.jpg
Thats a big one, 384 bit bus.. hopefully christmas will come early this year.
https://forums.guru3d.com/data/avatars/m/63/63170.jpg
user1:

Thats a big one, 384 bit bus.. hopefully christmas will come early this year.
Yes, I noticed the seemingly 384-bit bus also. This looks like one of those "Lightning" Cards or one of the Asus very high end models for Extreme Overclockers. So probably no HBM for any of the Nvidia High end consumer cards, not that it will be a problem 🙂 edit : Also, That top connector looks like a single NVLink connector, where Volta GV100 has two. ANother thing, is to look at the bigger picture, and the Box to the right. That looks like a picture of the Metal protector plate that protects the GPU. The only problem with that, is usually, the protector has mounting holes all around the GPU socket, not just at the corners. Could it be a mount for a cooling plate, maybe to cool the Memory ? I'm 99% certain this is a Pro card, or Prosumer, which probably has another 12 chips on the back for 24GB.
https://forums.guru3d.com/data/avatars/m/63/63170.jpg
W1zzard over at TPU has speculated on the Die size, and is tentatively calling this GV102. From his die size speculation, it would seem this could be either a smaller version of the full Volta spec (ie. a cheaper Volta Pro card), or a version with the same number of shadres, and just all of the Pro stuff removed (ie. The Next top range consumer Ti card) It could go either way. I don't think it would be a problem to bring NVLink into the consumer space, especially if they limit it physically to two cards in SLi max. edit : W1zzard sepculates it would be about 676mm², whereas the current GV100 is 815mm², so I would expect it to be 12nm still.
https://forums.guru3d.com/data/avatars/m/63/63215.jpg
I'm going to stroke my invisible beard.
https://forums.guru3d.com/data/avatars/m/250/250418.jpg
Evildead666:

edit : W1zzard sepculates it would be about 676mm², whereas the current GV100 is 815mm², so I would expect it to be 12nm still.
For such a big silicon, I expect an equivalent Nvidia price tag.
https://forums.guru3d.com/data/avatars/m/156/156133.jpg
Moderator
It could be because it's an engineering board, but I'm thinking this is a Quadro card looking at the NVLink connector?
https://forums.guru3d.com/data/avatars/m/63/63170.jpg
vbetts:

It could be because it's an engineering board, but I'm thinking this is a Quadro card looking at the NVLink connector?
Yes, but why would they limit Pro users to only two cards, with only one NVLink connector ? Thats what makes me think this could be a Consumer card, or maybe a Prosumer card ?
https://forums.guru3d.com/data/avatars/m/156/156133.jpg
Moderator
Evildead666:

Yes, but why would they limit Pro users to only two cards, with only one NVLink connector ? Thats what makes me think this could be a Consumer card, or maybe a Prosumer card ?
They're getting more and more away from multi gpu setups, so possibly with say...Base Quadro cards they're going to be limited to one? Who knows, but honestly I'm hoping for something new to come from both AMD and Nvidia.
https://forums.guru3d.com/data/avatars/m/63/63170.jpg
vbetts:

They're getting more and more away from multi gpu setups, so possibly with say...Base Quadro cards they're going to be limited to one? Who knows, but honestly I'm hoping for something new to come from both AMD and Nvidia.
For the Pro crowd, MultiGPU is working quite well afaik. I wouldn't think Nvidia would like to hinder them in any way...
https://forums.guru3d.com/data/avatars/m/246/246171.jpg
Evildead666:

For the Pro crowd, MultiGPU is working quite well afaik. I wouldn't think Nvidia would like to hinder them in any way...
I agree with both you and vbetts - Nvidia seems so committed to mGPU that they continue to invest in NVLink, and the GPU shown in the photos do use that interface. Getting GPUs to synchronize their data has pretty much been the bane for mGPU gamers, but the thing about workstations and servers is they don't tend to care that much about realtime processing, they just want to crunch as many numbers as they can. So, mGPU is definitely here to stay for high-end non-consumer systems. However, I think it is safe to assume that lower-end Quadros won't have NVLink. The low-end models still have plenty to offer.