Hi all - I attempted to search the forums but didn’t find an answer to what I was looking for. My apologies if there is already an answer somewhere in the forums.
I’m interested in building a linux box for test & dev, one with a GPU so that I can start learning about CUDA. Funny thing is that a while ago I got a PowerPC G4 off craiglist and on which I’ve installed Yellow Dog Linux.
Have people here had any experience installing a GPU card on a PowerPC G4 and done CUDA development on it using YDL (or any other Linux variant)? If it was a PC, I would have slapped the GPU a long time ago, but I’m completely clueless about the mac architecture, what it could take to do what I’d like to do (and whether it is possible.)
If I can save some bucks by reusing this PowerPC G4, as opposed to building it from scratch or buying it, that’d be great.
Hi all - I attempted to search the forums but didn’t find an answer to what I was looking for. My apologies if there is already an answer somewhere in the forums.
I’m interested in building a linux box for test & dev, one with a GPU so that I can start learning about CUDA. Funny thing is that a while ago I got a PowerPC G4 off craiglist and on which I’ve installed Yellow Dog Linux.
Have people here had any experience installing a GPU card on a PowerPC G4 and done CUDA development on it using YDL (or any other Linux variant)? If it was a PC, I would have slapped the GPU a long time ago, but I’m completely clueless about the mac architecture, what it could take to do what I’d like to do (and whether it is possible.)
If I can save some bucks by reusing this PowerPC G4, as opposed to building it from scratch or buying it, that’d be great.
To run CUDA code, you need the closed source Nvidia driver, which is only available for Intel CPUs. So unfortunately, you will not be able to use CUDA on a PowerPC for now.
To run CUDA code, you need the closed source Nvidia driver, which is only available for Intel CPUs. So unfortunately, you will not be able to use CUDA on a PowerPC for now.
I am pretty sure that PowerPC G4 computers only have a PCI bus inside. It is very hard to find a CUDA capable GPU on a PCI graphics card. The bus is so much slower than PCI-Express, the overhead of moving data kills the benefit of CUDA. The few GeForce 8 PCI cards that do exist also have pretty slow GPUs.
NVIDIA only ships a precompiled CUDA driver and compiler for x86 32 and 64-bit hardware. There is no source code (especially for the driver) that could be used to port CUDA to PowerPC. With appropriate wrappers, I could image the FreeBSD folks getting CUDA to work on x86 FreeBSD, but I think PowerPC Linux is completely impossible in any practical way.
If you can find pretty much any old Intel/AMD-based computer with a PCI-Express bus, that should be sufficient to pair up with a cheap CUDA card for learning purposes.
I am pretty sure that PowerPC G4 computers only have a PCI bus inside. It is very hard to find a CUDA capable GPU on a PCI graphics card. The bus is so much slower than PCI-Express, the overhead of moving data kills the benefit of CUDA. The few GeForce 8 PCI cards that do exist also have pretty slow GPUs.
NVIDIA only ships a precompiled CUDA driver and compiler for x86 32 and 64-bit hardware. There is no source code (especially for the driver) that could be used to port CUDA to PowerPC. With appropriate wrappers, I could image the FreeBSD folks getting CUDA to work on x86 FreeBSD, but I think PowerPC Linux is completely impossible in any practical way.
If you can find pretty much any old Intel/AMD-based computer with a PCI-Express bus, that should be sufficient to pair up with a cheap CUDA card for learning purposes.
Thanks tera and seibert for your replies. I guess then the PowerPC box is not a viable option.
Funny thing is that I have a HP G60-123CL laptop running Vista+cygwin (and several Linux vms) for development. I never really cared about the specific details other than RAM (3gigs) and that is a dual-core AMD Turion.
But out of curiosity I checked the specs, and this puppy has a NVIDIA GeForce 8200M G (which according to the NVIDIA site is a GPU card). With these particular specs, being a AMD Turion laptop, any concerns, caveats or tips in using my laptop for learning CUDA programming?
Thanks tera and seibert for your replies. I guess then the PowerPC box is not a viable option.
Funny thing is that I have a HP G60-123CL laptop running Vista+cygwin (and several Linux vms) for development. I never really cared about the specific details other than RAM (3gigs) and that is a dual-core AMD Turion.
But out of curiosity I checked the specs, and this puppy has a NVIDIA GeForce 8200M G (which according to the NVIDIA site is a GPU card). With these particular specs, being a AMD Turion laptop, any concerns, caveats or tips in using my laptop for learning CUDA programming?
Don’t get disappointed by the speed - programs are going to run faster on the CPU than on the GPU. But for learning CUDA it should be ok. Is there any particular thing you want to use CUDA for?
Don’t get disappointed by the speed - programs are going to run faster on the CPU than on the GPU. But for learning CUDA it should be ok. Is there any particular thing you want to use CUDA for?
Nothing really in particular, just pure unadulterated curiosity :) I’ve been doing application level programming (in Java) for the last 10 years, and I’ve been very disconnected with developments outside of that realm. I’ll be getting a gig to work on system level programming within a few months (I hope); I’m using my spare time to play with a lot of things; and I’m pursuing a MS in EE and CE. My main interest down the road is HPC and I wanted to see what CUDA is and how it is used for HPC.
Nothing really in particular, just pure unadulterated curiosity :) I’ve been doing application level programming (in Java) for the last 10 years, and I’ve been very disconnected with developments outside of that realm. I’ll be getting a gig to work on system level programming within a few months (I hope); I’m using my spare time to play with a lot of things; and I’m pursuing a MS in EE and CE. My main interest down the road is HPC and I wanted to see what CUDA is and how it is used for HPC.