- Jan 28, 2019
- 11
- 0
- 10
Couple preliminary points:
1) We use NAMD -- GPUs have got to be CUDA-capable Nvidia.
2) Our budget is around 5k USD.
3) By the end of the year we may have additional 5k to spend on the workstation, so it MUST have the potential to scale.
Questions:
1) Should we go for a single node workstation and try to make it as powerful as possible or a cluster of two sub-optimal nodes which together will outperform the optimal single node workstation? In case several-node systems are mostly needed to run several distinct simulations simultaneously, this is not our case -- assume we'll be running one task at a time.
2) Do we go with a single CPU system or a dual? Naturally, there will be more than one GPU. Do we need to have the second CPU to handle the second GPU? Or are modern CPUs capable of efficiently handling two GPUs at once? We won't be buying the 2nd CPU just now, but we need to plan for it, motherboard-wise.
3) Which CPU/CPUs? Should it necessarily be from Xeon family? Or would, say, i7-9800x be up to the task? Which particular models of both Xeon and non-Xeon series would you recommend?
4) What do we cool them with?
5) GPU. If we are able to afford Nvidia Tesla (K40, for example), should we go for it? Or are Teslas basically for supercomputers so they make sense only when there are thousands of them when even slight computational gains really add up, and we'd be no noticeably worse if we spent this money on a pair of decent GeForces?
Then which ones? Looks like a pair of 1070 buys 1080. Same goes for a pair of 2070 vs 2080. Do we go with the 10xx family or 20xx? Do we buy a single xx80 now and add another one at the end of the year? Or do we buy two xx70 now and add another two later? Will we be able to handle 4 GPUs? Are there even any benefits to having 4 GPUs instead of 2 or do the effects of having additional GPUs plateau at this point?
Also, which video card producers are the most reliable these days? Asus, EVGA, etc.?
6) Motherboard?
7) RAM. I guess no less than 64GB, but what other specs would be the best? Latency, etc.
8) PSU. How much watts will we need? Which PSUs are the most reliable?
9) And, perhaps, which tower. Well-ventilated, with enough space and everything.
Thanks.
1) We use NAMD -- GPUs have got to be CUDA-capable Nvidia.
2) Our budget is around 5k USD.
3) By the end of the year we may have additional 5k to spend on the workstation, so it MUST have the potential to scale.
Questions:
1) Should we go for a single node workstation and try to make it as powerful as possible or a cluster of two sub-optimal nodes which together will outperform the optimal single node workstation? In case several-node systems are mostly needed to run several distinct simulations simultaneously, this is not our case -- assume we'll be running one task at a time.
2) Do we go with a single CPU system or a dual? Naturally, there will be more than one GPU. Do we need to have the second CPU to handle the second GPU? Or are modern CPUs capable of efficiently handling two GPUs at once? We won't be buying the 2nd CPU just now, but we need to plan for it, motherboard-wise.
3) Which CPU/CPUs? Should it necessarily be from Xeon family? Or would, say, i7-9800x be up to the task? Which particular models of both Xeon and non-Xeon series would you recommend?
4) What do we cool them with?
5) GPU. If we are able to afford Nvidia Tesla (K40, for example), should we go for it? Or are Teslas basically for supercomputers so they make sense only when there are thousands of them when even slight computational gains really add up, and we'd be no noticeably worse if we spent this money on a pair of decent GeForces?
Then which ones? Looks like a pair of 1070 buys 1080. Same goes for a pair of 2070 vs 2080. Do we go with the 10xx family or 20xx? Do we buy a single xx80 now and add another one at the end of the year? Or do we buy two xx70 now and add another two later? Will we be able to handle 4 GPUs? Are there even any benefits to having 4 GPUs instead of 2 or do the effects of having additional GPUs plateau at this point?
Also, which video card producers are the most reliable these days? Asus, EVGA, etc.?
6) Motherboard?
7) RAM. I guess no less than 64GB, but what other specs would be the best? Latency, etc.
8) PSU. How much watts will we need? Which PSUs are the most reliable?
9) And, perhaps, which tower. Well-ventilated, with enough space and everything.
Thanks.
Facebook
Twitter
Instagram