Message boards :
Number crunching :
NVIDIA Jetson TK1
Message board moderation
Author | Message |
---|---|
W-K 666 Send message Joined: 18 May 99 Posts: 19012 Credit: 40,757,560 RAC: 67 |
I've just been informed that the Zotac NVIDIA Jetson TK1 is now available from Maplins for UK customers, price is £199.99 Zotac NVIDIA Jetson TK1 Developer Kit (previous thread on low power computers has been closed) |
Dr Grey Send message Joined: 27 May 99 Posts: 154 Credit: 104,147,344 RAC: 21 |
14 W for 300 GFLOP/s sounds amazingly efficient. Would this work? A review of the board is here: http://www.itpro.co.uk/desktop-hardware/22731/nvidia-jetson-tk1-review |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
14 W for 300 GFLOP/s sounds amazingly efficient. Would this work? A review of the board is here: Here is one actually running. http://setiathome.berkeley.edu/show_host_detail.php?hostid=7354060 As I recall they had to compile the app for the hardware & there was quite a bit of effort to get it producing valid results. For the money I am sticking with Bay Trail-D. The ASRock Q1900-ITX I bought recently only cost around $80 US, or about £50 GB (with VAT I imagine more), & looks to be running around 3200-3500 for RAC. My system total is around 25w, but once optimized I am hoping to get closer to 15w at full load. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Dr Grey Send message Joined: 27 May 99 Posts: 154 Credit: 104,147,344 RAC: 21 |
I'm not sure whether to trust the GFLOPS figures quoted on our rigs but just looking at 17 tasks for the TK1 today compares pretty favourably with the 37 done by my i7 if you compare 14 W vs 70 W. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I'm not sure whether to trust the GFLOPS figures quoted on our rigs but just looking at 17 tasks for the TK1 today compares pretty favourably with the 37 done by my i7 if you compare 14 W vs 70 W. The NVIDIA GK20A on the TK1 lists Average processing rate 15.82 GFLOPS An old NVIDIA 8500 GT I have overclocked lists Average processing rate 15.12 GFLOPS The GK20A is a bit faster than my 8500. Given that I think the app GFLOPS listed seem to work for comparison sake at least. I think the 14W listed is for the whole SOC. So just the GPU running should be a lower value. Maybe 7-8W?. Looking at the tasks on your i7 they take ~2 hours for a normal AR vs 2.5-2.75 hours on that GPU. However you are probably running 4-7 in parallel on your 70w CPU? So the PPW could be somewhat close. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Dr Grey Send message Joined: 27 May 99 Posts: 154 Credit: 104,147,344 RAC: 21 |
Yes, 7 on the i7. So it looks like the TK1 isn't quite the supercomputer they're saying it is - but still a neat unit though. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Yes, 7 on the i7. So it looks like the TK1 isn't quite the supercomputer they're saying it is - but still a neat unit though. It has a really good GPU, for an arm processor. The Nvidia Shield Tablet, which uses this same SOC, is not that much more than a Google Nexus 7 & has a lot better GPU. So that is another option for those looking at this hardware. http://shield.nvidia.com/gaming-tablet/ or http://shield.nvidia.co.uk/gaming-tablet/ SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
ivan Send message Joined: 5 Mar 01 Posts: 783 Credit: 348,560,338 RAC: 223 |
Yeah, that's me. I had to add some ifdefs to the machine-code routines (someone else worked that out...) to get the floating-point working right. Note that the results it's producing are purely from the GPU. I've compiled the CPU code and it seems to run fine standalone but under BOINC itself it throws an exception. I haven't had time to try to debug this.14 W for 300 GFLOP/s sounds amazingly efficient. Would this work? A review of the board is here:Here is one actually running. Also note that some CUDA code I have for reconstructing digital holograms compiled and ran without any code changes on the system -- I just had to change the location of the includes and libraries in the Makefile. It runs about 1/10th the speed of my GTX 750 Ti, but takes perhaps 1/15th the power. I've got my wattmeter at work now so I'll try to get some power figures soon (got the chance to grab it when I had to shut down my home machine last night to remove the GTX 640 which had developed a faulty fan -- the third Gigabyte GPU that's done that to me now! :-( ). For the money I am sticking with Bay Trail-D. TheASRock Q1900-ITX I bought recently only cost around $80 US, or about £50 GB (with VAT I imagine more), & looks to be running around 3200-3500 for RAC. My system total is around 25w, but once optimized I am hoping to get closer to 15w at full load.Mine (Acer Aspire XC-603) is running around 1250, but that's on Linux and without using the GPU. Cost £130. |
ivan Send message Joined: 5 Mar 01 Posts: 783 Credit: 348,560,338 RAC: 223 |
Yeah, that's me. ... It runs about 1/10th the speed of my GTX 750 Ti, but takes perhaps 1/15th the power. I've got my wattmeter at work now so I'll try to get some power figures soon (got the chance to grab it when I had to shut down my home machine last night to remove the GTX 640 which had developed a faulty fan -- the third Gigabyte GPU that's done that to me now! :-( ). OK, some figures. At idle, the wattmeter was showing 9 W at the socket (it was 4 W with the machine shut down, so that's about the inefficiency of the power supply). Running two CUDA jobs simultaneously shows 16 W, with occasional drops to 14 W and sometimes 11 W. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Any patches or recommendations for X-branch ? It's on idle while I address some complex Boinc issues, but I don't stop taking issues for the ToDo list :) "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
ivan Send message Joined: 5 Mar 01 Posts: 783 Credit: 348,560,338 RAC: 223 |
Any patches or recommendations for X-branch ? It's on idle while I address some complex Boinc issues, but I don't stop taking issues for the ToDo list :) No, IIRC it just compiled as usual on the Jetson. No real problems with the s@h, just the missing include I reported last January, and I had to edit the config file to remove the old compute capabilities that nvcc didn't like and put in 3.2 for the Tegra.I had to add |
ivan Send message Joined: 5 Mar 01 Posts: 783 Credit: 348,560,338 RAC: 223 |
GT 640 which had developed a faulty fan -- the third Gigabyte GPU that's done that to me now! :-( )..Just ordered a replacement fan from China on eBay for about £1.50, so might be back in action by the end of the month. Otherwise I'll get the Asus GT640-1GD5-L as a replacement -- it has to fit in a single-width slot with only about 5 mm extra clearance. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.