Jump to content

Question about CUDA cores.

PepeRTXLover

Can someone explain me, how the CUDA Cores are implemented into GPU die ? 

 

Because I was wondering, if Nvidia is able to active some extra cores ? 

RTX 4070, have 5888 Cores, meanwhile 4070 Super, over 7000, but with same architecture AD104 as 4070.

 

And my question is, is there any way to enable cores ? Or the Die is just made with 5888 cores on 4070, and there's no way to change it ? 

Link to comment
Share on other sites

Link to post
Share on other sites

Usually they make one die with a set number of cores, and activate/deactivate cores for certain SKUs.

 

I'm not sure if the cores are enabled/disabled in hardware or firmware, though. If they're activated in firmware, theoretically you could flash a different vBIOS and have them activated, but that's a pretty sketchy venture with no guarantees of working and a higher chance of turning your GPU into a paperweight.

Quote or tag me( @Crunchy Dragon) if you want me to see your reply

If a post solved your problem/answered your question, please consider marking it as "solved"

Community Standards // Join Floatplane!

Link to comment
Share on other sites

Link to post
Share on other sites

7 minutes ago, Crunchy Dragon said:

Usually they make one die with a set number of cores, and activate/deactivate cores for certain SKUs.

 

I'm not sure if the cores are enabled/disabled in hardware or firmware, though. If they're activated in firmware, theoretically you could flash a different vBIOS and have them activated, but that's a pretty sketchy venture with no guarantees of working and a higher chance of turning your GPU into a paperweight.

Maybe I could ask Nvidia, maybe they can release some bios update (even paid one) to active more CUDA cores 🫣

Link to comment
Share on other sites

Link to post
Share on other sites

Welcome to the forums!
These days they laser off the other cores, often because they are defective. They tend not to down spec perfectly good product unless the market is being *really* weird. 

5950X/3080Ti primary rig  |  1920X/1070Ti Unraid for dockers  |  200TB TrueNAS w/ 1:1 backup

Link to comment
Share on other sites

Link to post
Share on other sites

28 minutes ago, Crunchy Dragon said:

I'm not sure if the cores are enabled/disabled in hardware or firmware, though. If they're activated in firmware, theoretically you could flash a different vBIOS and have them activated, but that's a pretty sketchy venture with no guarantees of working and a higher chance of turning your GPU into a paperweight.

That actually is a thing with some older cards. It's possible to flash an RX 460 and turn it into an RX 560, for example, if it turns out that the cores were good. The most recent example is the RX 5700, which could be flashed into an RX 5700 XT with a decent success rate.

 

But yes, it can come with the potential risk of bricking the card. And obviously, if the card was binned that way because it had defective SPs, then it can become wildly unstable. Power delivery can also become a concern, if the board just barely made spec for the lower tier part, the VRMs can overheat when it becomes the higher tier one.

 

21 minutes ago, PepeRTXLover said:

Maybe I could ask Nvidia, maybe they can release some bios update (even paid one) to active more CUDA cores 🫣

Nvidia has things very locked down these days. A recent leak has opened up BIOS modding to the RTX 20 series cards, but the RTX 30 series has resisted efforts from even the most determined enthusiasts. I highly doubt Nvidia would ever release something like that, for fear that it could be reverse engineered and given away for free.

Link to comment
Share on other sites

Link to post
Share on other sites

1 hour ago, PepeRTXLover said:

Maybe I could ask Nvidia, maybe they can release some bios update (even paid one) to active more CUDA cores 🫣

Only if they're not fused off. 

NVIDIA has numerous incentives to deactivate parts of the GPU, although it ultimately comes down to profit. 

 

Some of those cores may not have been fit for a higher end GPU - depending on the chip of course. Some only max out to a certain amount of set components - so enabling them would potentially cause you issues, both from a stability perspective of the GPU itself, as well as a power perspective from the VRM support on the PCB itself. 

 

Besides, NVIDIA is unlikely to do that if it was currently possible because they would rather you just buy another card instead. 

"It pays to keep an open mind, but not so open your brain falls out." - Carl Sagan.

"I can explain it to you, but I can't understand it for you" - Edward I. Koch

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×