Running a piece of code on hi-end PC
-
Guys, who can tell me whether there is some site/service allowing coders to run their code on really fast PCs, testdrive-like. Maybe I am too naive but it would be nice if Intel/AMD and why not CP to have one such PC (with, say, 32 threads and 64GB) dedicated to users' benchmarking purposes, yes?
Get down get down get down get it on show love and give it up What are you waiting on?
-
Guys, who can tell me whether there is some site/service allowing coders to run their code on really fast PCs, testdrive-like. Maybe I am too naive but it would be nice if Intel/AMD and why not CP to have one such PC (with, say, 32 threads and 64GB) dedicated to users' benchmarking purposes, yes?
Get down get down get down get it on show love and give it up What are you waiting on?
You may find some answers here: http://aws.amazon.com/ec2/faqs/[^]
I'm not questioning your powers of observation; I'm merely remarking upon the paradox of asking a masked man who he is. (V)
-
Guys, who can tell me whether there is some site/service allowing coders to run their code on really fast PCs, testdrive-like. Maybe I am too naive but it would be nice if Intel/AMD and why not CP to have one such PC (with, say, 32 threads and 64GB) dedicated to users' benchmarking purposes, yes?
Get down get down get down get it on show love and give it up What are you waiting on?
Usually people test against lower-end CPU/IO, not high-end. What are you wanting to see or prove by running your code on a higher end CPU ?
I'd rather be phishing!
-
You may find some answers here: http://aws.amazon.com/ec2/faqs/[^]
I'm not questioning your powers of observation; I'm merely remarking upon the paradox of asking a masked man who he is. (V)
Thanks, I will have it in mind, however it is not exactly what I need: Your application will execute on a virtual computer that we call an instance. AFAIK, a Virtual Machine is not the proper choice for benchmarking.
Get down get down get down get it on show love and give it up What are you waiting on?
-
Thanks, I will have it in mind, however it is not exactly what I need: Your application will execute on a virtual computer that we call an instance. AFAIK, a Virtual Machine is not the proper choice for benchmarking.
Get down get down get down get it on show love and give it up What are you waiting on?
Sanmayce wrote:
for benchmarking
For that you have to use an average computer you customers will use to. Those fast (GPU powered) VMs are most for prove of concept...
I'm not questioning your powers of observation; I'm merely remarking upon the paradox of asking a masked man who he is. (V)
-
Usually people test against lower-end CPU/IO, not high-end. What are you wanting to see or prove by running your code on a higher end CPU ?
I'd rather be phishing!
The key thing for me is to calibrate a C snippet for modern CPU, e.g. Haswell. I want to see what boost e.g. the Haswell's 4 ALUs can deliver in internal loops where every shift/rotate/multiplication matters. My old Core 2 cannot give me the real picture.
Get down get down get down get it on show love and give it up What are you waiting on?
-
Guys, who can tell me whether there is some site/service allowing coders to run their code on really fast PCs, testdrive-like. Maybe I am too naive but it would be nice if Intel/AMD and why not CP to have one such PC (with, say, 32 threads and 64GB) dedicated to users' benchmarking purposes, yes?
Get down get down get down get it on show love and give it up What are you waiting on?
Now stop and think for a moment about throughput on that PC. If you want an accurate benchmark, then you need to be the only active application on the machine. So once you are finished, the machine has to be cleared completely and reset back to a clean install condition before the next guy can start installing the stuff he (and he alone) needs to run his app. Remotely. Which means he has to upload everything needed to the computer, slowing things more, and adding the fun of potential illegal software installation to the operators risk list... It doesn't make a lot of sense economically for the PC owners, unless they charge a considerable amount for the service - which means they aren't going to get many users, because a small number of runs will equal the cost of purchase of the equipment in the first place!
Those who fail to learn history are doomed to repeat it. --- George Santayana (December 16, 1863 – September 26, 1952) Those who fail to clear history are doomed to explain it. --- OriginalGriff (February 24, 1959 – ∞)
-
Now stop and think for a moment about throughput on that PC. If you want an accurate benchmark, then you need to be the only active application on the machine. So once you are finished, the machine has to be cleared completely and reset back to a clean install condition before the next guy can start installing the stuff he (and he alone) needs to run his app. Remotely. Which means he has to upload everything needed to the computer, slowing things more, and adding the fun of potential illegal software installation to the operators risk list... It doesn't make a lot of sense economically for the PC owners, unless they charge a considerable amount for the service - which means they aren't going to get many users, because a small number of runs will equal the cost of purchase of the equipment in the first place!
Those who fail to learn history are doomed to repeat it. --- George Santayana (December 16, 1863 – September 26, 1952) Those who fail to clear history are doomed to explain it. --- OriginalGriff (February 24, 1959 – ∞)
You discouraged me ... a bit. Speaking of profits I admit my dumminess. >If you want an accurate benchmark, then you need to be the only active application on the machine. So once you are finished, the machine has to be cleared completely and reset back to a clean install condition before the next guy can start installing the stuff he (and he alone) needs to run his app. That's right. >which means they aren't going to get many users No problema, there is no race for amounting users, few of them would suffice as long as they achieve their goals. As for the fee, if a serious coder needs feedback he willingly would pay the tax instead of buying such PC, but it is not about money, rather for the idea.
Get down get down get down get it on show love and give it up What are you waiting on?
-
The key thing for me is to calibrate a C snippet for modern CPU, e.g. Haswell. I want to see what boost e.g. the Haswell's 4 ALUs can deliver in internal loops where every shift/rotate/multiplication matters. My old Core 2 cannot give me the real picture.
Get down get down get down get it on show love and give it up What are you waiting on?
Well I've got a 4770K right here. Maybe we can work something out.
Sanmayce wrote:
4 ALUs can deliver in internal loops where every shift/rotate/multiplication matters.
Shift/rot still have a throughput of 2/cycle at best. Shifts by cl have become worse than they used to be, consider using "shlx" and friends.
-
Well I've got a 4770K right here. Maybe we can work something out.
Sanmayce wrote:
4 ALUs can deliver in internal loops where every shift/rotate/multiplication matters.
Shift/rot still have a throughput of 2/cycle at best. Shifts by cl have become worse than they used to be, consider using "shlx" and friends.
Many thanks for your readiness to help. One of the things that interests me is how FNV1A_farolito fares on i7 4th generation: http://www.sanmayce.com/Fastest_Hash/index.html#farolito[^] It has been a while (6 months) since my last play with FNV1A_farolito and I didn't put it in a bench package as other functions. Grrr, I'm not ready at the moment. The other is my MEMMEM function, however I won't bother you with either, again I am not ready. I will be happy if you are still willing to run my MEMMEM after I write Railgun_Swampshine package. I am not in a hurry at all, I have been asking for half a year for such benchmarking and nobody cared, thanks again.
Get down get down get down get it on show love and give it up What are you waiting on?
-
Many thanks for your readiness to help. One of the things that interests me is how FNV1A_farolito fares on i7 4th generation: http://www.sanmayce.com/Fastest_Hash/index.html#farolito[^] It has been a while (6 months) since my last play with FNV1A_farolito and I didn't put it in a bench package as other functions. Grrr, I'm not ready at the moment. The other is my MEMMEM function, however I won't bother you with either, again I am not ready. I will be happy if you are still willing to run my MEMMEM after I write Railgun_Swampshine package. I am not in a hurry at all, I have been asking for half a year for such benchmarking and nobody cared, thanks again.
Get down get down get down get it on show love and give it up What are you waiting on?
