How efficient is your cruncher?

[hat]

It might be worth it in the long run, though. If you're going to be running WCG for a while, that's a fair bit of savings you can earn over time with undervolting. Less power use also means less heat generated, too. When you're doing it with, what, 4 rigs? Might be worth the time investment. Even if you make 0 points for an entire week, after that it's smooth sailing. When you want your rigs to be efficient, it's worth looking into.

 

[mstenholm]

Since I'm planing to replace my daily rig with something from this century I began to test/measure my old gear so I had some numbers to compare the new PC with. To my surprise one of my old X58s (W3670 = i7 970) on Linux only took 205 W running 3.875 MHz. 12 threads is 12k on a good day. My similar but water cooled Windows "powered" does 6k on 10 threads. That takes 215 W. Again a small OC is involved. My two similar 2700Xs takes 190 and 200 W at stock which in my case is 3950 MHz all cores. They will do 25k on a pure Zika diet, 12k on MIP.

My surprise was not so much the points but the small difference in power consumption. Sure the new stuff running the right type of WCG is way more efficient but for basic computation the AMDs doesn't blow the old Intels away.

Let's hope that the promised better energy efficiency and higher IPC holds true for 3900/3950X since that's were I'm heading.

 

[phill]

I'll be updating the Pie this evening everyone, I've been off of work for a few days and busy at home over the weekend... Normal service should resume soon I hope everyone is alright

 

[Deleted member 178884]

i9-7980xe @ 3.5ghz 0.9v (in bios)
190-200W CPU only, no wall meter.
X299 Omega
Seasonic snow silent 750W

This is roughly pulling 18-19k points per hour.

 

[Simplex0]

Xx Tek Tip xX That's impressive, almost twice as much as my Threadripper 1950X@3.9 GHz. Are you running Windows or Ubuntu?

 

[Deleted member 178884]

Xx Tek Tip xX That's impressive, almost twice as much as my Threadripper 1950X@3.9 GHz. Are you running Windows or Ubuntu?

Windows 10 pro + cut down myself.

 

[phill]

i9-7980xe @ 3.5ghz 0.9v (in bios)
190-200W CPU only, no wall meter.
X299 Omega
Seasonic snow silent 750W
View attachment 130581

This is roughly pulling 18-19k points per hour.

Is this with HT on now?? I'm thinking you might mean 18k to 19k WCG points not Boinc? Impressive stuff mind

 

[Deleted member 178884]

Is this with HT on now?? I'm thinking you might mean 18k to 19k WCG points not Boinc? Impressive stuff mind

Yeah WCG points, HT is on and it's undervolted : P

 

[phill]

Yeah WCG points, HT is on and it's undervolted : P

Awesome So the AIO isn't quite at melting point just yet??

Was testing something last night that surprised me, but sadly that was not quite as fast as I had hoped... A bit more than a Ryzen 6 core but not masses sadly

 

[Deleted member 178884]

Was testing something last night that surprised me, but sadly that was not quite as fast as I had hoped... A bit more than a Ryzen 6 core but not masses sadly

Yeah Ryzen is the best choice perf/watt, this CPU is a beast in any and every other workload I've ran with it : )

Awesome So the AIO isn't quite at melting point just yet??

Not yet, stock is doable, but I felt as if an undervolt would work better and provide better perf/watt
Maybe running Linux would increase output by a large amount.

 

[mstenholm]

Yeah Ryzen is the best choice perf/watt, this CPU is a beast in any and every other workload I've ran with it : )

Not yet, stock is doable, but I felt as if an undervolt would work better and provide better perf/watt
Maybe running Linux would increase output by a large amount.

Not maybe, it will increase a lot if you choose the right project (OpenZika).


Edit: I'm down to zero watt per rig these days. My power is off and I'm 15 hours away from the HFI relay / RCCB - Residual Current Circuit Breaker or what it is called in English, lightning took it all. OK I only had two laptops and my 28k PPD Linux/2700X running.

 

[phill]

Yeah Ryzen is the best choice perf/watt, this CPU is a beast in any and every other workload I've ran with it : )

Not yet, stock is doable, but I felt as if an undervolt would work better and provide better perf/watt
Maybe running Linux would increase output by a large amount.

Linux and AMD CPUs are definitely a hard combo to beat.. I'm very much looking forward to the day I can replace my 2600km E3-1245 V3 and my 6700k with either Ryzen or some more Xeon's Just seems daft not to since I have things here I can make use of, why not make use of them?

I love making the most of the hardware I have here, since I have free electric when the suns out, why the heck not I think most CPUs are also best at stock or very nearly too it.. I only run my 5960X a bit above stock when it comes to volts, but the clock speed is an extra 1.20Ghz per thread, that can't be bad can it?

 

[melk]

I'm starting to retire my Sandy/Ivy era rigs and try to stick to anything Haswell or newer.

Easier to just sell off the old cpu's and buy some Ryzen instead.

 

[thebluebumblebee]

I'm starting to retire my Sandy/Ivy era rigs and try to stick to anything Haswell or newer.

Easier to just sell off the old cpu's and buy some Ryzen instead.

Welcome to TPU!

 

[melk]

Welcome to TPU!

Thanks How do I add my WCG stats like you have there at the bottom?

 

[thebluebumblebee]

If you're talking about the badges, https://www.techpowerup.com/forums/account/preferences

 

[phill]

I'm starting to retire my Sandy/Ivy era rigs and try to stick to anything Haswell or newer.

Easier to just sell off the old cpu's and buy some Ryzen instead.

Welcome to TPU

I'll be doing the same with my few quad cores, why have quad cores using 75% of the power of octo cores?? I'll be retiring mine whenever I can eventually afford it

 

[melk]

Yep, it's also a pain to manage multiple boxes. Especially not worth it for only dual cores. I need to try and sell off the Dell server I have while I can still get a few bucks for it. Running a big tower server just for 4-cores really isn't worth it.

edit: Also I set my WCG username in my forum account prefs but that doesn't seem to have done anything. Maybe because I'm not a member of TPU team there?

 

[phill]

Yep, it's also a pain to manage multiple boxes. Especially not worth it for only dual cores. I need to try and sell off the Dell server I have while I can still get a few bucks for it. Running a big tower server just for 4-cores really isn't worth it.

edit: Also I set my WCG username in my forum account prefs but that doesn't seem to have done anything. Maybe because I'm not a member of TPU team there?

Definitely not, sadly anything other than some newer CPUs such as a 6700k or the Ryzen's do tend to really bring any decent points without big power usages. I have some X58 kit that I use when the solar is working nicely and let that chew 400w+ for 24 threads (dual Xeon's) but even then that will produce half the points that my Ryzen 1700X does, whilst using less than 180w of power.. Times have definitely changed now

It might take a little while to sync, you can change your team you crunch for to TPU if you wish? (Unless you have already?)

 

[mstenholm]

3900X stock Linux, 23 instances (one left for folding), on a diet of 95% SCC, 4% MCM and 1% ARP/HST it will do +40K (301k for the past week = 43K). AX850, water cooled. 230 W at the wall. Beat that @phill

 

[hat]

How efficient is your cruncher... a valid question I've been thinking about lately. I'd like to get some better cooling and clock the snot out of my 2600k for now, because I'm not really interested in new processors for my main rig just yet, and the cooler (and new case to go along with it) will be handy in the future. In a way, though, it doesn't matter how good my cooling is... a chip that pulls 200w is gonna pull 200w and that heat still gets dumped into the room.

For my secondary machine, Ryzen 1700/2700 are apparently 65w chips. That's pretty good. I'm sure they'll sip less power than my old Athlon II x4 630. It's 95w out of the box at 2.8GHz. I have it overclocked to 3.5GHz with +.150 volts I think. I'm sure it's close to 125w or more, double what a chip with twice the cores and probably more than 4x the performance would take.

 

[phill]

3900X stock Linux, 23 instances (one left for folding), on a diet of 95% SCC, 4% MCM and 1% ARP/HST it will do +40K (301k for the past week = 43K). AX850, water cooled. 230 W at the wall. Beat that @phill

I'll never get anywhere near to that I don't run my rigs very often 24/7 but I'm averaging I guess about 10k to 14k ish a day with the 10 hours ish they are on and they are doing a diet of everything My machine has no bounds

I've been doing a few tests with the hardware I'm currently using so when I get 5 minutes from work and Sophia, I'll try and post up my results I do know however that the rigs are certainly not energy efficient as I've not done anything very much with them at all. I do however have to set my 3900X to 3.15GHz and then set the vcore to 0.875v and away it goes. I'm still hitting close to 80C at times so I definitely know I need water to tame the beast. However that said, I do wonder if a 3950X would be any better given the fact there's more cores and threads in the CPU.... If I can find one I will grab it hopefully!!

@hat I'd look forward to seeing what sort of results you are getting with your crunchers as I like to see how others manage to set their machines up I do have a 2700 (non X model) and that thing is amazingly efficient I think it was pulling about 130w fully loaded whilst testing in R15/R20 etc... I need to track down another board for it...

 

[hat]

Oh, I don't have anything like that... at least not yet. I have the Athlon II x4 still. I've only got the Ryzen chips in my head...

 

[xvi]

I was wondering exactly that the last few days about the new Ryzen chips. I was a little confused at the TDP ratings between them.

Seems really odd to me why the Ryzen 5 3600X (6 core) has a higher TDP than the Ryzen 7 3700X (8 core). I'm assuming it comes down to target TDP and throttling to reach that.

Spoiler: Very simple estimates

Since IPC should be near the same between the two (I would imagine?), I've taken the lazy approach of multiplying base freq by the number of cores for a rough comparison of multi-threaded performance. I've then divided that by TDP to roughly estimate performance per watt. I assume these won't reflect real-world performance at all because I suspect the 3600X will spend a higher amount of time in boost clocks whereas the 3700X will be throttling down to base freq much more often to keep on the target TDP. Does anyone else have thoughts on this?

3600X - Base freq * cores = 22800 MHz total, divided by 95w TDP = 240 MHz/watt
3600X - Turbo freq * cores = 26400 MHz total, divided by 95w TDP = 278.9 MHz/watt
3700X - Base freq * cores = 28800 MHz total, divided by 65w TDP = 443 MHz/watt
3700X - Turbo freq * cores = 35200 MHz total, divided by 65w TDP = 541.5 MHz/watt

Seems like from a perf/watt standpoint, even if the 3600X could spend all of its time on all four cores at 4.4GHz, it would still be much less efficient than the 3700X in any scenario.

I assume given the same cooling abilities, the Ryzen 5 will spend much more time in boost clocks but it still seems like it wouldn't come close to the 3700X for efficiency even in its worst-case scenario. I suppose it boils down to how much time the CPUs will spend in boost clocks. The performance just seems hard to estimate.

I've nabbed this list of CPU performance from SETI@Home with hopes that it would shine some light on the real-world compute performance of the chips.

Spoiler: Zen, Zen+, Zen2 SETI@Home GFLOPS

CPU model​	CoreSpeed(GHz)​	Family​	cores/CPU​	GFLOPS/core​	GFLOPs/CPU​
AMD Ryzen 5 1400 Quad-Core Processor​	​	Family 23 Model 1 Stepping 1​	7.84​	3.85​	30.21​
AMD Ryzen 5 1500X Quad-Core Processor​	​	Family 23 Model 1 Stepping 1​	8​	4.4​	35.21​
AMD Ryzen 5 1600 Six-Core Processor​	​	Family 23 Model 8 Stepping 2​	12​	4.54​	54.46​
AMD Ryzen 5 1600 Six-Core Processor​	​	Family 23 Model 1 Stepping 1​	11.96​	4.23​	50.52​
AMD Ryzen 5 1600X Six-Core Processor​	​	Family 23 Model 1 Stepping 1​	12​	4.58​	55.01​
AMD Ryzen 7 1700 Eight-Core Processor​	​	Family 23 Model 1 Stepping 1​	15.78​	4.05​	63.88​
AMD Ryzen 7 1700 Eight-Core Processor​	​	Family 23 Model 1 Stepping 1​	16​	3.57​	57.18​
AMD Ryzen 7 1700X Eight-Core Processor​	​	Family 23 Model 1 Stepping 1​	15.91​	4.38​	69.75​
AMD Ryzen 7 1800X Eight-Core Processor​	​	Family 23 Model 1 Stepping 1​	15.77​	4.62​	72.93​
AMD Ryzen Threadripper 1900X 8-Core Processor​	​	Family 23 Model 1 Stepping 1​	14.53​	4.6​	66.9​
AMD Ryzen Threadripper 1920X 12-Core Processor​	3.5​	Family 23 Model 1 Stepping 1​	23.61​	4.57​	107.89​
AMD Ryzen Threadripper 1950X 16-Core Processor​	3.4​	Family 23 Model 1 Stepping 1​	31.28​	4.53​	141.56​
AMD Ryzen 5 2400G with Radeon Vega Graphics​	​	Family 23 Model 17 Stepping 0​	7.97​	4.55​	36.23​
AMD Ryzen 5 2500U with Radeon Vega Mobile Gfx​	​	Family 23 Model 17 Stepping 0​	7.94​	3.37​	26.76​
AMD Ryzen 5 2600 Six-Core Processor​	​	Family 23 Model 8 Stepping 2​	11.95​	4.58​	54.68​
AMD Ryzen 5 2600X Six-Core Processor​	​	Family 23 Model 8 Stepping 2​	11.97​	4.85​	58​
AMD Ryzen 5 PRO 2400G with Radeon Vega Graphics​	​	Family 23 Model 17 Stepping 0​	8​	4.44​	35.53​
AMD Ryzen 5 PRO 2400GE w/ Radeon Vega Graphics​	​	Family 23 Model 17 Stepping 0​	8​	3.81​	30.49​
AMD Ryzen 7 2700 Eight-Core Processor​	​	Family 23 Model 8 Stepping 2​	15.88​	4.25​	67.46​
AMD Ryzen 7 2700U with Radeon Vega Mobile Gfx​	​	Family 23 Model 17 Stepping 0​	8​	3.73​	29.85​
AMD Ryzen 7 2700X Eight-Core Processor​	​	Family 23 Model 8 Stepping 2​	15.84​	4.91​	77.75​
AMD Ryzen Threadripper 2920X 12-Core Processor​	3.5​	Family 23 Model 8 Stepping 2​	23.6​	5.14​	121.26​
AMD Ryzen Threadripper 2950X 16-Core Processor​	3.5​	Family 23 Model 8 Stepping 2​	29.76​	4.83​	143.62​
AMD Ryzen Threadripper 2970WX 24-Core Processor​	3​	Family 23 Model 8 Stepping 2​	47​	4.19​	196.75​
AMD Ryzen Threadripper 2990WX 32-Core Processor​	3​	Family 23 Model 8 Stepping 2​	57.73​	4.34​	250.71​
AMD Ryzen 5 3400G with Radeon Vega Graphics​	​	Family 23 Model 24 Stepping 1​	7.83​	4.9​	38.33​
AMD Ryzen 5 3500U with Radeon Vega Mobile Gfx​	​	Family 23 Model 24 Stepping 1​	8​	3.68​	29.47​
AMD Ryzen 5 3550H with Radeon Vega Mobile Gfx​	​	Family 23 Model 24 Stepping 1​	8​	4.05​	32.41​
AMD Ryzen 5 3600 6-Core Processor​	​	Family 23 Model 113 Stepping 0​	11.92​	5.12​	61.02​
AMD Ryzen 5 3600X 6-Core Processor​	​	Family 23 Model 113 Stepping 0​	11.93​	5.2​	62.04​
AMD Ryzen 5 PRO 3500U w/ Radeon Vega Mobile Gfx​	​	Family 23 Model 24 Stepping 1​	8​	4.19​	33.54​
AMD Ryzen 7 3700U with Radeon Vega Mobile Gfx​	​	Family 23 Model 24 Stepping 1​	8​	3.87​	31​
AMD Ryzen 7 3700X 8-Core Processor​	​	Family 23 Model 113 Stepping 0​	16.46​	5.27​	86.77​
AMD Ryzen 7 3750H with Radeon Vega Mobile Gfx​	​	Family 23 Model 24 Stepping 1​	8​	4.25​	34.02​
AMD Ryzen 7 3800X 8-Core Processor​	​	Family 23 Model 113 Stepping 0​	15.92​	5.33​	84.87​
AMD Ryzen Threadripper 3960X 24-Core Processor​	3.8​	Family 23 Model 49 Stepping 0​	45​	5.59​	251.5​
AMD Ryzen Threadripper 3970X 32-Core Processor​	3.7​	Family 23 Model 49 Stepping 0​	60​	5.4​	323.82​

Some of the bits interesting to me is the meager difference between the 3600 and 3600X and the massive difference between 3600X (at supposedly 95W TDP?) and 3700X (at 65W TDP?). The 3700X has 140% of the performance and 68% of the heat output? That just doesn't seem right.

Edit: Now in Google Docs format with all CPUs. Sorting and filtering doesn't seem to work in read-only, but you can save a copy for yourself if you'd like.

 

[thebluebumblebee]

And then there's this:

For AMD, TDP does not equal power usage.