The question that interests me the most about such projects is whether you can get more computational power from such a cluster than from one or two PC/server like devices which cost the same amount of money or have the same power consumption.
PC/Servers will have much more power consumption. The advantage of these setups is when you can run things in parallel which is the case of simulation.
I think the point here is that getting access to a supercomputer for research purposes can be expensive, if you aren't affiliated with a university. Having a cluster-environment to test things on before running your simulations on a large scale supercomputer can be very helpful. Of course, there is also the issue of scalability. Vertical scaling get very expensive (eg. buying a more powerful computational device), so most people choose to scale horizontally (buying more of the same type of device, and coordinate them).
The performance/cost ratio is not very clear from the presentation, although it feels like little bang for buck. It is an interesting project though. Apologies for my pessimism - academia projects simply don't play my note.
+Jaroslav Malec Haven't been able to work out all the numbers, but it looks like the amount of computational power for the power input is at least a factor of 10 better than with a much faster multi-core PC. The real power of clusters is due to the amount of parallelims, and that is down to the number of nodes more than the outright speed.
So what's constitute a "supercomputer"? I feel that this term is being used inappropriately in this presentation. I know it sounds much catchier, but it deflects from the real point that she build a computing cluster from a series inexpensive SBC, which in itself nothing to be ashamed of, in fact, she should really proud of
It depends on the tasks you are trying to process. Some tasks can be optimised on a GPU, some tasks are more optimised for CPUs. Some new FPGAs are the best you can get for optimising the FPGA for the software. FPGAs are what I use - I recommend using those in computational tasks.
The question that interests me the most about such projects is whether you can get more computational power from such a cluster than from one or two PC/server like devices which cost the same amount of money or have the same power consumption.
PC/Servers will have much more power consumption. The advantage of these setups is when you can run things in parallel which is the case of simulation.
I think the point here is that getting access to a supercomputer for research purposes can be expensive, if you aren't affiliated with a university. Having a cluster-environment to test things on before running your simulations on a large scale supercomputer can be very helpful. Of course, there is also the issue of scalability. Vertical scaling get very expensive (eg. buying a more powerful computational device), so most people choose to scale horizontally (buying more of the same type of device, and coordinate them).
The performance/cost ratio is not very clear from the presentation, although it feels like little bang for buck. It is an interesting project though.
Apologies for my pessimism - academia projects simply don't play my note.
+Jaroslav Malec It's really not a bad attempt however..given her original monetary constraints.
+Jaroslav Malec Haven't been able to work out all the numbers, but it looks like the amount of computational power for the power input is at least a factor of 10 better than with a much faster multi-core PC. The real power of clusters is due to the amount of parallelims, and that is down to the number of nodes more than the outright speed.
Very good! Just couldn't find her in linkedin or Facebook!
Or I can say it's a low-cost cluster? just curious, what script she used to dispatch each board?
How do they communicate with each other if not for Ethernet ?
I wonder how she interfaced with it. I would imagine you could use Putty and a serial connection.
I guess that she installed an ssh server on each os
Браво Кристина . Много интересно. Аз пък търся как да си направя cluster с AMD и Nvidia карти и да използвам CUDA и Vulkan в едно :D
So what's constitute a "supercomputer"? I feel that this term is being used inappropriately in this presentation. I know it sounds much catchier, but it deflects from the real point that she build a computing cluster from a series inexpensive SBC, which in itself nothing to be ashamed of, in fact, she should really proud of
“High performance “
Wouldn't a modern GPU have better computational capabilities? I thought that was where it was at
It depends on the tasks you are trying to process. Some tasks can be optimised on a GPU, some tasks are more optimised for CPUs. Some new FPGAs are the best you can get for optimising the FPGA for the software. FPGAs are what I use - I recommend using those in computational tasks.
The problems are the software developer, Nvidia, AMD and Microsoft with the fucking directx.
yeah but a cluster of GPU is too much $$$
Did you not hear her when she said she was on a tight budget
GPUs are horrible for algorithms that have many branch statements, which CPUs are better at.
4:40 I'm from Bulgaria raspberry pis have GPUs, if you're gonna spend more money then you would on them you would get thinkpads I'm just saying!
awesome =0)
yeah but can you make bitcoins on it?
+Dylan T You can make bitcoins on almost anything, it's just not a good idea to do so because ASICs are far more energy efficient.
heh?