News Nvidia has another RTX 4070 variant brewing — this one uses a down-binned AD103 GPU from the RTX 4080 Super

[Admin]

Nvidia has a new RTX 4070 variant coming, this one using AD103 silicon rather than the usual AD104 GPU. The RTX 4070 (AD103) will apparently have over 40% of the potential GPU cores disabled, along with one memory controller.

Nvidia has another RTX 4070 variant brewing — this one uses a down-binned AD103 GPU from the RTX 4080 Super : Read more

 

[PEnns]

And the name of this new creation would be?? RTX 4080 Super Mini Me??

 

[JamesJones44]

And the name of this new creation would be?? RTX 4080 Sper Mini Me??

I doubt there be a new name. The flawed 4080 chips will have whatever extra working parts beyond what the 4070 requires disabled in firmware most likely.

 

[35below0]

And the name of this new creation would be?? RTX 4080 Sper Mini Me??

RTX 408050. Just to mess with everyone.

 

[Eximo]

They did do different names occasionally. Usually when they left part of the bigger chip enabled. Like the 2060 KO GPUs which had 2080 GPUs. It was a slightly better workstation card because of the way the chips were built internally. Extra ROPs or something, been a while.

It is possible it will retain the increased cache of the larger cards. Hopefully someone has a GPU-Z shot of it somewhere.

 

[Pierce2623]

They did do different names occasionally. Usually when they left part of the bigger chip enabled. Like the 2060 KO GPUs which had 2080 GPUs. It was a slightly better workstation card because of the way the chips were built internally. Extra ROPs or something, been a while.

It is possible it will retain the increased cache of the larger cards. Hopefully someone has a GPU-Z shot of it somewhere.

I’ve been trying for years to figure out what was left activated in the 2060 ko that raised it performance above the 2070 in some productivity apps. It wouldn’t have been ROPs, as they’re number is fixed per SM active.

 

[subspruce]

And the name of this new creation would be?? RTX 4080 Super Mini Me??

4070 KO

 

[Eximo]

I’ve been trying for years to figure out what was left activated in the 2060 ko that raised it performance above the 2070 in some productivity apps. It wouldn’t have been ROPs, as they’re number is fixed per SM active.

My vague recollection was something like that the bigger chip had some level of math precision enabled that the lower cards didn't. I couldn't find anything on it that didn't require watching some old videos.

 

[FunSurfer]

So what if the 4070 only needs three channels? Why can't it use the whole four channels so we could finally get the RTX 4070 16GB and the RTX 4070 Super 16GB?

 

[Eximo]

This is likely the non-functional part of the chip they had to disable and the reason for binning them down to this point in the first place.

The chip has 10240 potential CUDA cores, stripping it down to 5888 is pretty drastic. Next smallest GPU using that die is the 4070 Ti Super with 8448 cores working. Mobile 4090, 4080, and 4080 Super are the only other cards using it.

So those chips were really bad rejects to make this worthwhile.

 

[Pierce2623]

So what if the 4070 only needs three channels? Why can't it use the whole four channels so we could finally get the RTX 4070 16GB and the RTX 4070 Super 16GB?

Probably because that memory controller is in fused off area of the chip

 

[Pierce2623]

My vague recollection was something like that the bigger chip had some level of math precision enabled that the lower cards didn't. I couldn't find anything on it that didn't require watching some old videos.

This is likely the non-functional part of the chip they had to disable and the reason for binning them down to this point in the first place.

The chip has 10240 potential CUDA cores, stripping it down to 5888 is pretty drastic. Next smallest GPU using that die is the 4070 Ti Super with 8448 cores working. Mobile 4090, 4080, and 4080 Super are the only other cards using it.

So those chips were really bad rejects to make this worthwhile.

Yeah any chip with 40% of its SMs disabled was indeed likely to be in pretty bad shape to start with.

 

[Joseph_138]

This change will likely be done without any fanfare. You won't even know which GPU die you have, until you run GPU-Z. There will be no difference in performance, though I do wish they could have kept the memory management units intact, to give a boost to 16gb VRAM, on a 256-bit bus. 12gb is already becoming a bottleneck in some games, with all the settings maxed out.

 

[35below0]

This change will likely be done without any fanfare. You won't even know which GPU die you have, until you run GPU-Z. There will be no difference in performance, though I do wish they could have kept the memory management units intact, to give a boost to 16gb VRAM, on a 256-bit bus. 12gb is already becoming a bottleneck in some games, with all the settings maxed out.

Which games are these? There can't be many.

 

[Eximo]

Been on 12GB for a while, haven't run into anything I can't get playable rates out of at 1440p. I'm sure there are games at 4K and modded games out there that can push it over the edge. But that is what the bigger GPUs exist for.

 

[35below0]

Been on 12GB for a while, haven't run into anything I can't get playable rates out of at 1440p. I'm sure there are games at 4K and modded games out there that can push it over the edge. But that is what the bigger GPUs exist for.

Yeah, modded games for sure. Loads of modders have no restraint or common sense.

 

[subspruce]

Yeah, modded games for sure. Loads of modders have no restraint or common sense.

its like half of all modders either run 4090s or 720x480 CRTs