Message boards : Number crunching : First Skylake CPUs hit the streets
Previous · 1 · 2 · 3 · 4 · Next
Author | Message |
---|---|
dcdc Send message Joined: 3 Nov 05 Posts: 1831 Credit: 119,523,428 RAC: 9,566 |
Here's my first Skylake: https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=2403229 That's a g4500 (£60/$100) at stock speed (it doesn't allow overclocking). I'll measure the power consumption tonight. The CPU/mobo/8G RAM combo was around £175 because I needed a motherboard with displayport and a few other things, so could be done cheaper. I expect a quad-core i5 would give better value in overall crunching terms. It looks like it will top out around 2500 credits per day from the first two results, so that's higher than the quad-core Q6600 (~2000/day) that it will replace, and hopefully while using less power. |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
Here's my first Skylake: pretty impressive per-core speeds: your skylake cpu CPU type GenuineIntel Intel(R) Pentium(R) CPU G4500 @ 3.50GHz [Family 6 Model 94 Stepping 3] Measured floating point speed 4265.01 million ops/sec my haswell cpu CPU type GenuineIntel Intel(R) Core(TM) i7-4771 CPU @ 3.50GHz [Family 6 Model 60 Stepping 3] Measured floating point speed 3546.6 million ops/sec one thing though i'm running 'turbo off' hence speeds is capped at 3.5 ghz but if i leave turbo on it'd probably be closer to 3.7 gflops & it is still a margin lower per core compared to skylake the only 'saving grace' for now is that i7 is hyper threaded, hence i'm able to run 8 r@h jobs concurrently apparently at around 3.5 gflops (roughly 1 flops per cycle) do let us know about the temperatures average temperatures running 8 concurrent r@h threads keeps temp around 60 deg C ambient at around 28-30 degs C. using an aftermarket cpu cooler / heat sink http://www.coolermaster.com/cooling/cpu-air-cooler/hyper-212x/ |
dcdc Send message Joined: 3 Nov 05 Posts: 1831 Credit: 119,523,428 RAC: 9,566 |
It's looking like it will top out at 2700 which is more than I expected. My Haswell g3258 (3.2GHz, not currently overclocked) gets 2100 when not heavily used, which scales to 2300 when the clock speeds are matched at 3.5GHz. The g3258 has 2xDDR3 1600, which is quite quick and in dual channel mode. The g4500 has 1XDDR4 2666, so not in dual channel mode. I might have a play with the bclk clock next week and see how it overclocks- didn't realise I could do that until reading up on it yesterday... |
dcdc Send message Joined: 3 Nov 05 Posts: 1831 Credit: 119,523,428 RAC: 9,566 |
It's pulling 45W at the mains. The PSU isn't particularly efficient - it's currently on an old but decent quality FSP PSU, a laptop hard drive and stock (tiny) heatsink/fan. I expect that will drop when I move it to the Seasonic SII 380W PSU and an SSD. CPU temp is hovering around 36°C, although it's in the cellar and so is quite cold in here - probably around 16°C I would guess. I'll hopefully be able to use my big Thermaltake heatsink on it when I swap everything over to it, but not sure if I've got the right adapter... |
Timo Send message Joined: 9 Jan 12 Posts: 185 Credit: 45,649,459 RAC: 0 |
I've added the Dell Optiplex 7040 (Skylake i7 6700T) to my super compact BOINC farm. I learned that the cooling solution on the lenovo boxes is way quieter than the one used on the Dell box. Unfortunately that means I'll need to purposely down-clock the dell box so the noise profile is quiet enough for approval by the wife. (I've mentioned in other threads, I'm allowed to continue BOINC as long as it's out of site and inaudible to the wife - hence these are stacked on a chair under a table in the back room.. Never thought I could fit 24 threads of CPU power into such a small space without spending some major coin, not to mention the profile is small enough that it can remain out of sight of the wife XD) Below: My stack of BOINC boxes, the Skylake box is on top of the two other boxes. (The other two boxes are Haswell i7-T series). ** I know someone noted putting them vertically like books, but it seems to make little difference to be honest, the heat transfer vertically is minimal between these cases, and they all blow hot air out of the same side (what people in the data center world call the hot aisle, I gather) And another shot just to show how compact these things are: Everyone likes pictures, so I am happy to share :) Will be interesting to see if this can push my productivity past 15K/day even after down-clocking... |
dcdc Send message Joined: 3 Nov 05 Posts: 1831 Credit: 119,523,428 RAC: 9,566 |
Rather than underclocking, if your motherboard supports it, you could try undervolting. Dropping from 1.2 to 1.1V will reduce CPU heat production by up to 16% (it might be less than that as some parts might not be undervolted). Also, Tthrottle http://efmer.com/b/ might be of interest as it'll dynamically adjust the load based on the CPU temp. |
[VENETO] boboviz Send message Joined: 1 Dec 05 Posts: 1994 Credit: 9,524,889 RAC: 7,500 |
I've added the Dell Optiplex 7040 (Skylake i7 6700T) to my super compact BOINC farm. On Dell's site, i find only 7040 with i5 cpu.... |
Timo Send message Joined: 9 Jan 12 Posts: 185 Credit: 45,649,459 RAC: 0 |
I've added the Dell Optiplex 7040 (Skylake i7 6700T) to my super compact BOINC farm. I bought this one, which is an i7-6700T Mind you, I'm in Canada so the product lineup may differ if you're in Europe (ships from the US). |
Chilean Send message Joined: 16 Oct 05 Posts: 711 Credit: 26,694,507 RAC: 0 |
|
Timo Send message Joined: 9 Jan 12 Posts: 185 Credit: 45,649,459 RAC: 0 |
This DELL bios has pretty damn limited options, and doesn't have anything available for touching the voltages, clocks, multipliers, etc - though it's impressive for a BIOS in that the mouse has full functionality and it looks to be in full blown 16k color or better... I've managed to find a nice balance that lets this thing run 24/7 100% CPU usage (Rosetta@Home) @ 2.5GHz or so while staying cool enough and barely making a whisper. Even with the slight down clock, it Looks like it will be the highest production of my cluster yet. Not to mention, when the wife's at work I can crank it up to full throttle (3.2GHz) for an extra push.. The virtues of working from home :) |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
It's pulling 45W at the mains. The PSU isn't particularly efficient - it's currently on an old but decent quality FSP PSU, a laptop hard drive and stock (tiny) heatsink/fan. I expect that will drop when I move it to the Seasonic SII 380W PSU and an SSD. it is impressive in energy efficiency, my i7 haswell box sits idle at 55w and bumps up to 100W full throttle running 8 threads at the wall. ambient temp 30 deg C idle temp 40 deg C load temp 60 deg C the higher idle consumption probably comes from the hard drives (i've more than one as some are simply 'old' lower capacity drives). the other consumption is the led monitor as it is normally turned on and off conrurrently, which could factor in part of that 55w at idle |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
Rather than underclocking, if your motherboard supports it, you could try undervolting. Dropping from 1.2 to 1.1V will reduce CPU heat production by up to 16% (it might be less than that as some parts might not be undervolted). i think back in the 90's there is Dennard scaling in which smaller transistors means thinner oxides and lower voltages can be used. Since around 2000, Dennard scaling starts to fail. smaller sizes and increasing number of transistors (e.g. multi core cpus) did not lead to correspondingly lower voltages. today at 14nm feature sizes, the oxides can only get so thin and the gates are probably only atoms thick. this could mean in a way that 14nm could be near the end of Moore's law and further smaller features beyond this point could one day become impossible. below certain voltages the cpu may simply become unstable. in fact i've come across some forums devoted to over-clocking and to achieve that with stability higher voltages is needed which means higher power consumption. With such extreme small feature sizes, it is not surprising that the heat intensity/density becomes much higher and as power consumed cannot be reduced which leads to rather notorious haswell cpus overheating problems. http://www.extremetech.com/computing/116561-the-death-of-cpu-scaling-from-one-core-to-many-and-why-were-still-stuck https://communities.intel.com/docs/DOC-23517 |
Jim1348 Send message Joined: 19 Jan 06 Posts: 881 Credit: 52,257,545 RAC: 0 |
i think back in the 90's there is Dennard scaling in which smaller transistors means thinner oxides and lower voltages can be used. Since around 2000, Dennard scaling starts to fail. smaller sizes and increasing number of transistors (e.g. multi core cpus) did not lead to correspondingly lower voltages. today at 14nm feature sizes, the oxides can only get so thin and the gates are probably only atoms thick. this could mean in a way that 14nm could be near the end of Moore's law and further smaller features beyond this point could one day become impossible. below certain voltages the cpu may simply become unstable. I attended an International Solid State Circuit Conference (ISSCC) meeting about 20 years ago where a professor(probably from Stanford) laid it all out for our current time. The end of the line comes at around 1 volt, below which the transistors don't operate. You can at most only add more transistors per chip, if you can figure out what to do with them, but you can't make them any faster. The End is Near. |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
a buzzword that goes around seemed to be 'dark silicon', i.e. it is too hot to turn on all the transistors at the same time. liquid nitrogen cooling is probably needed. so chip designers 'spent' transistors to do specialized functions that may or may not be used frequently. e.g. Intel chips has mp4 encoding/decoding acceleration, on chip AES, TSX - transaction sync extensions. for all you know as crunching such molecular dynamics projects are getting very popular, Intel & other chip factories may start introducing instructions which accelerates such calculations. that'd probably make the camp who wish to make rosetta@home top the credit charts happy :) http://darksilicon.org/ |
Jim1348 Send message Joined: 19 Jan 06 Posts: 881 Credit: 52,257,545 RAC: 0 |
Yes, they can implement specialized functions, or instructions such as SSE, AVX, etc. But then the software engineers have to use them. |
Chilean Send message Joined: 16 Oct 05 Posts: 711 Credit: 26,694,507 RAC: 0 |
The era of silicon is near its end. But, that doesn't mean there aren't other alternatives to it. I'm sure Intel/AMD are full aware of this and are spending lots of brainpower and money to figure this out. According to Intel, they should at least reach 7nm with relatively little problem. Cannonlake, which should have the 10nm die shrink is coming around 2017... so there's still time until the bitter end. Many have stated that CPU performance is increasing at a slower rate than it did before due to the focus on "more-cores per core" approach rather than "faster cores" approach (because of the whole power/heat-limit thing). Good thing that BOINC utilices all cores perfectly fine. So... more cores "might" not be of much benefit for web browsing or gamin, BUT, it is very useful for Rosetta :) |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
ok, this seemed to be an 'old' thread by now. but i came across something interesting and useful Intel power gadget https://software.intel.com/en-us/articles/intel-power-gadget-20 the good thing is that it lets you estimate and monitor the power information in watts without needing any additional instrumentation. it depends on your CPU's model specific register. Hence, not all Intel CPUs would support that. It is apparently pretty useful if it is supported. when my haswell i7 4771 is crunching Rosetta@hhome, crunching 8 threads concurrently Average Processor Power(Watt)=40.1253 that would be the processor alone consumption. It won't include the energy/power consumed by RAM, harddrives etc. i capped the cpu frequency at 3.5 Ghz, i used an aftermarket heatsink/cpu cooler http://www.coolermaster.com/cooling/cpu-air-cooler/hyper-212x/ that gives a cpu temperature of 60 deg C while the ambient temperature is 30 dec C while all 8 r@h threads are concurrently running. when idle this processor consume a 'mere' 2-4 watts. if you happen to run Intel power gadget for your skylake processors, do share the statistics as well |
rjs5 Send message Joined: 22 Nov 10 Posts: 273 Credit: 22,994,271 RAC: 9,725 |
ok, this seemed to be an 'old' thread by now. Some of these numbers still don't seem to make sense. Power popped up when I increased from 3.9ghz to 4.10ghz but did not drop after lowering it back to 3.9ghz. I am using ( http://www.gigabyte.com/microsite/369/images/system-tweaking.html ) to change the frequency but I declined the app request to reboot after changing frequency. Windows 10 Haswell 5930x running 12 Rosetta threads base/turbo: 3.5ghz/3.7ghz PkgPwrLimit0: 3408.3W Base Frq0: 3.50GHz Max Temp0: 95 overclocking at 3.9ghz Package0: 22.54W Package Temp0 61 Cooler fan reading 1253rpm overclocking at 4.1ghz Package0: 26.42W Package Temp0 66 Cooler fan reading 1253rpm SkyLake has removed the voltage regulator from the CPU package so it should show either show higher frequency or lower power without the VR measurement included. |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
Some of these numbers still don't seem to make sense. Power popped up when I increased from 3.9ghz to 4.10ghz but did not drop after lowering it back to 3.9ghz. 12 threads! it is a monster of a cpu! i'd guess 22.54W is a per core power? otherwise running 12 threads & consuming a mere 22.54W is truly extremely power efficient and if 22.54 is per core power, that translates to 132W - 158W and keeping temps at 61 deg, u'd have a huge cooler? :D lol |
rjs5 Send message Joined: 22 Nov 10 Posts: 273 Credit: 22,994,271 RAC: 9,725 |
Some of these numbers still don't seem to make sense. Power popped up when I increased from 3.9ghz to 4.10ghz but did not drop after lowering it back to 3.9ghz. Liquid Cooling Package Specifications •Large Copper Waterblock •Sealed System / Maintenance Free •High Air-Flow 120MM Radiator Cooling Fan •Arctic Silver 5 Thermal Compound •2 x 120mm Case Fan I don't worry about overheating when overclocking from 3.7GHz turbo to 3.9GHz but at 4.1GHz, it hung once when running PrimeGrid AVX tasks. My SkyLake 6700k system shipped today and is expected to arrive on 12/17. I put a similar liquid cooler on it too. It cost about $120 more over the Intel type CPU fan. |
Message boards :
Number crunching :
First Skylake CPUs hit the streets
©2024 University of Washington
https://www.bakerlab.org