Could there be Hyper-Threaded GPUs?

[DanielMDA]

Hey, so I was wondering, GPUs are made up of MANY little less powerful CPUs (according to Linus`s Techquickie of GPUs). So, since there are CPUs with Hypertreading so that your CPU wastes less or no time waiting for an instruction and therefore boosting it by around 30-40%, could these be applied to future GPUs so that they are better?

[TheCheese]

Theoretically yes although practical problems may arise when integrating the technology into the GPUs and would probably require new drivers for it

[Oshino Shinobu]

I doubt it. GPUs have many, many times more processors than CPUs. Hyperthreading basically tries to make up for a lack of cores on CPUs, but when you have hundreds and even thousands like on a GPU, it seems kind of pointless to me. 

 

It could work, but I feel like it would have already been done if it would be useful. 

[Guest]

Since that there are numerous different technologies implemented in GPUs such as CUDA exceleration for boosting performance when rendering and other features such as boost clocks to improve fps, it seems pointless to me, because there are all of these technologies integrated into the GPU to excelerate performance in games/rendering.

[DanielMDA]

Just now, TheBeastPC said:

Since that there are numerous different technologies implemented in GPUs such as CUDA exceleration for boosting performance when rendering and other features such as boost clocks to improve fps, it seems pointless to me, because there are all of these technologies integrated into the GPU to excelerate performance in games/rendering.

And if CUDA implemented this? Shouldn`t it be interesting?

[ivryk]

8 hours ago, DanielMDA said:

Hey, so I was wondering, GPUs are made up of MANY little less powerful CPUs (according to Linus`s Techquickie of GPUs). So, since there are CPUs with Hypertreading so that your CPU wastes less or no time waiting for an instruction and therefore boosting it by around 30-40%, could these be applied to future GPUs so that they are better?

Isn't Async compute pretty much like hyper-threading?, in which each queue can submit commands without waiting for other task to complete.

[Jumper118]

5 hours ago, ivryk said:

Isn't Async compute pretty much like hyper-threading?, in which each queue can submit commands without waiting for other task to complete.

No. It's not same at all. 

[razaldo]

Don't know about hyper, but there is always some room for HYPE-threaded GPUS 

[ivryk]

4 hours ago, Jumper118 said:

No. It's not same at all. 

It isn't the same however they both in their own way keep the cores busy whenever they can and try to eliminate idle time.

[Jumper118]

5 minutes ago, ivryk said:

It isn't the same however they both in their own way keep the cores busy whenever they can and try to eliminate idle time.

Not really. 

[ivryk]

Just now, Jumper118 said:

Not really. 

Sure, then explain how they both (in their own way) dont do that.

[Jumper118]

5 minutes ago, ivryk said:

Sure, then explain how they both (in their own way) dont do that.

They are totally different. One is serial the other is parallel the cpu isn't losing time idling without ht. It's still going flat out. The gpu cores are sittings there idle waiting for the slower ones to finish. They can't really be compared in anyway. 

 

It is like saying more ghz is the same as having more cores.

[ivryk]

Just now, Jumper118 said:

They are totally different. One is serial the other is parallel the cpu isn't losing time idling without ht. It's still going flat out. The gpu cores are sittings there idle waiting for the slower ones to finish. They can't really be compared in anyway. 

I respect your opinion however, hyper-threading does help with idling by scheduling the task, heck even watch Linus's video about it. Again they're not the same however they both make their respective environment more efficient.