Power consumption G530 vs i7-3770T

[dZeus]

In case you're curious, I've ordered an Intel Core i3-4130T with 35W TDP (building a system for a relative), and can do some power measurements when everything has arrived and when I've assembled the thing. I have a simple wall-plug power consumption meter that gives semi-accurate readings...

I'm measuring the power consumption 'at the wall' using the following monitor:
Technoline 'Cost Control'
According to this link, it has a less than 5% measurement error at the loads we're looking at.

Components monitored:
Asus Q87T motherboard
Intel i3-4130T dual core cpu
8GB SO-DIMM 1.5v memory
Samsung EVO 120GB SSD
Intel 7260 AC wireless adapter
Lite On RP-090-SNAA power adapter

The system is running Windows 7 with the latest drivers installed. The screen is hooked up through HDMI, Gigabit LAN is connected to the switch, keyboard + mouse are connected through a powered USB hub that is not included in the monitored power consumption.

Idle: 16 Watt
Normal use (installing service packs, libreoffice, etc.): 22-28 Watt
Prime95 CPU torture test: 41 Watt
Prime95 + Furmark (GPU Torture Test): 51 Watt

To me this suggests that this particular CPU does indeed reach its max. TDP under full load.

[Umfriend]

Wow, assuming it uses 0W idle it appears to be on the mark with a 35W TDP indeed. Many thanks for this. I've ordered the i7-3770 yesterday and of course the mail messes it up but somewhere next week my server will be running on a way way to powerfull 3770. Fortunately, it's got Rosetta to keep it busy and I'll shut it off at 1100 and 2300 hours for two hours (backuptimes of the server) or so. Pretty sure I should be crunching more and using less power.

Many thanks dZeus.

[dZeus]

Wow, assuming it uses 0W idle it appears to be on the mark with a 35W TDP indeed. Many thanks for this. I've ordered the i7-3770 yesterday and of course the mail messes it up but somewhere next week my server will be running on a way way to powerfull 3770. Fortunately, it's got Rosetta to keep it busy and I'll shut it off at 1100 and 2300 hours for two hours (backuptimes of the server) or so. Pretty sure I should be crunching more and using less power.

Many thanks dZeus.

ah if it only were as simple as that

The power adapter (Lite On RP-090-SNAA) is rated 'up to 90% efficiency'. I don't have its efficiency curve, but usually maximum efficiency is reached somewhere between 40 and 80% of the maximum load (90W).

If we take, for example, 88% efficiency for the power supply at 50W AC and 90% for the voltage conversion circuity / voltage regulators on the motherboard at 50W AC, and 84% at 16W AC for each, we'd arrive at the following DC levels:
0.88 * 0.90 * 51W = 40 W
0.84 * 0.84 * 16W = 11 W
That would result in a delta of 29W.

in practice it's very difficult to accurately measure the real DC consumption without knowing all the conversion efficiency curves involved, or without special measurement circuity directly installed onto the motherboard.

oh and the CPU most decidedly uses more than 0W at idle. More likely to be in the 5 - 10 Watt range.

I can also post some figures for my current (Northwood P4 2.8C) and soon to arrive 'new' server system (i5 750) too, just to show the TDP differences between different Intel CPUs.

[Umfriend]

That's a 90 PSU then? I should check my server, it'll have a way overpowered PSU.

[Umfriend]

OK, TDPs from Intel suck. Just put in the i7-3770 with a TDP of 77W and out went the G530, TDP 65W. At least one of those make no sense (and I think it is the G530). The i7 goes over 70C under full CPU load and the stock cooler is spinning like a jet. Why I went the stock cooler way? Because it is a server that is put in a place where sound won't really matter but mostly because the G530 was doing so well with its stock cooler (which, from the looks of it is the exact same cooler). The G530 was not te be gotten over 50C with quite lower fan speeds.

Switching turbo-boost and HT off gets me to 55C with the cooler running 2100+ RPM.

So, I'll be getting me a lower price range cooler. Any suggestions? I'm primarily looking for way better cooling, sound levels is secondary, and easy, real easy installation.

[Evildead666]

The Pentium would never have gotten to its TDP rating, Dual core, no turbo, it may just have that TDP to allow for some really bad bins of chips.
The i7-3770 though, with 4 cores, eight threads, and turbo, is going to hit that 77W TDP as soon, and as often as it can, temperatures permitting.
Especially so if it is working at 100% load

If you're going to stay aircooling, maybe get the biggest heatsink that will fit in your case, and use the case fans to control the airflow.

Also, have you tried undervolting ? Many of the CPU's are set very conservatively, and you may be able to drop the CPU voltage a bit. It'll take some trial and error testing, but it might be worth it.

[dZeus]

The Pentium would never have gotten to its TDP rating, Dual core, no turbo, it may just have that TDP to allow for some really bad bins of chips.
The i7-3770 though, with 4 cores, eight threads, and turbo, is going to hit that 77W TDP as soon, and as often as it can, temperatures permitting.
Especially so if it is working at 100% load

If you're going to stay aircooling, maybe get the biggest heatsink that will fit in your case, and use the case fans to control the airflow.

Also, have you tried undervolting ? Many of the CPU's are set very conservatively, and you may be able to drop the CPU voltage a bit. It'll take some trial and error testing, but it might be worth it.

I second the undervolting, provided that your motherboard allows for it. Also, test undervolting only when running non-critical applications! When undervolting too far, you might get some really nasty side effects such as system instability or worse, data corruption.

One method which I used for finding the minimum voltage at each multiplier of my Pentium Mobile:
- Run Prime95 and find the minimum voltage at which it would run for an hour without errors.
- In order to guarantee system stability, I'd then increase the voltage for each multiplier by 3 notches above the list of minimum voltages found.

[Umfriend]

Now they tell me.....

I got a Scythe Samurai on my i5-2500 with a TDP of 95W and it stays at about 56C with the fan running at 2300RPM, audible but for that RPM rather quiet. It's just a bugger to install as the heatsink is larger than the area between the heatsink-fixation thingies. I think I should re-install because it used to do way better, perhaps to much paste this time.

The server is a larger case I have I think, at least more room, so I might try that one.

I'm not that inclined to experiment with the server...

[Umfriend]

Well, the Scythe wasn't that big of a succes the first time. Thought I had perhaps messed up the paste, it is something I am somewhat insecure on, how much to use? How to check if it is done properly? Can I reseat cooler without cleaning and applying new paste?

Anyway, CPU went to 100C so something was wrong. Decided to clean everything and redo. Luckilly I only had to clean the CPU of paste. The coolers' paste was easily removed by removing the plastic cover THAT SHOULD HAVE BEEN REMOVED IN THE FIRST PLACE! LOL.

Anyway, I am undervolting and underclocking. iGPU and CPU at -0.04V (offset), turbo boost disabled and speed at 3.2Ghz (instead of 3.4Ghz stock). It seems to be running at slightly below 45C with a shitload of RPM in reserve, just as I like it. That's with a full CPU load. My server is now done for at least 10 years I think expect for potentially a migration to some Server Essentials edition, which is not likely to occur soon if at all (it would be nice though if WHS2011 would (a) allow for more than 8GB internal memory and (b) support vhdx, but all that ain't gonna happen).

Zokes: you reading this? You'll never catch up with me in Rosetta! HA! ;-)

[Umfriend]

Ah, no end to my grief....

Crashes, even if everything is at default and no load. I do have an USB issue, the external drive on an USB 3 port would just not have any IO. Also, switching my keyboard+mouse back and forth from Server (temporarily at my desk) and PC (thtough a KOnig manual switch) causes the server to crash.. it took a while to figure that out. But it simply is not stable anymore. I've had the EUFI Bios crash on me at times. I wonder whether the PSU or MB might be toast. Darn...

[Evildead666]

Have you got the intel drivers for the USB3, or just the ones Windows installed ?
Is your KVM a PS/2 one or a USB one for Keyboard/Mouse ?

See if Prime95 will run on all 8 Threads without crashing a thread. It should rash pretty quickly if its not stable.
Then try a memtest, or first of you prefer.

Put your nose up really close to the PSU and smell it. While its on or off. If it smells even a little like burning, its probably toast. If it isn't then its probably fine. PSU's are pretty stable, even if you push them nowadays. A short circuit is something else.

Leaving the plastic on the CPU or Heatsink is usually a very bad idea, and if you got very high temps, and then removed the plastic and got good temps but crashing, it could be the CPU.
You could also try downclocking it to 2GHz or something, and rying Prime95 at that, since cooling/Power won't be a problem, just the CPU/Memory.

[Umfriend]

Yeah, had the real drivers installed (not Intel, H67 chipset) and it is not a KVM, it is really a simple switch that only allows one of two computers access to a USB device, which happens to be 1 monitors' 4-port USB Hub (which goes to another).

Maybe my undervolting experiment borked the MB but I don;t think that is likely. Far more likely is perhaps that the USB 3.0 controller went belly up after processing two server backups a day for two years? It was really burning down, crashes becoming ever more frequent (as in, I could crash it easily without doing anything out-of-the-ordinary).

Anyway, got me an Asrock B75 PRO3-M today and so far so good. Installation, updates, BOINC, backups, restore of the drivepools, no crash at all sofar, fingers crossed. I really think the MB was worn out.

Two downsides to the MB (I did not have much time to find a better one, Server needs to run monday 8 am):
- Less OC/UV options, none on LLC and VCore only -50 and -100MV, I may try -50 later when I have a decent backup again.
- Only one case fan header

edit: Oh, and using the USB switch has not crashed it yet either.

[Evildead666]

I wouldn't see undervolting doing any physical damage, only data corruption.

If a new Motherboard has done the trick, then thats good.
I've never heard of USB usage killing a mobo, but I have overloaded a Motherboard with too many USB devices (6 or 8 multi-car readers).
it didn't damage the motherboard as I remember, but that was back in the P4 days.

[Umfriend]

Neither have I until I thought of it myself ;-). Ah well, that's what you get with sub-50 euro MBs I guess. Although otherwise, ASRock has kept me pretty happy with all other MBs I got. We'll see.

Doing well with Rosetta yours is.

[dZeus]

...
I can also post some figures for my current (Northwood P4 2.8C) and soon to arrive 'new' server system (i5 750) too, just to show the TDP differences between different Intel CPUs.

The second-hand i5 750 story didn't end well unfortunately. I've got a non-functional cpu+motherboard+ram, and I'm not sure which one of the components is not working. With no intention to sink even more money in this pit, I decided to buy the cheapest Intel socket 1150 motherboard and cpu combination available (less than 90€ for the both), as a stop-gap solution. And here I'm posting the promised power readings on old and new server platform.

Configurations
Old platform: Intel P4 2.8C Northwood (69.7 W TDP), Intel D875PBZ, 4 x 1GB ECC DDR SDRAM, PCI Video, 3c905C-TX PCI LAN.
New platform: Intel Celeron G1820 2.7 GHz (53 W TDP), MSI H81M-P33, 1 x 8GB 1.5v DDR3 SDRAM

Testing methodology
The tests were performed with a Nexus 430W PSU, running Debian Wheezy booted off a USB Flash drive. Idle is tested with no cpu intensive tasks running, the load test is run with a program called burnP6 (one instance per physical CPU core).

Power readings
Pentium 4 2.8C idle: 45 W
Pentium 4 2.8C load: 106 W

Celeron G1820 idle: 19 W
Celeron G1820 load: 38 W

Notes
- In Windows, stressing the GPU by running Prime95+Furmark would probably add another 5-10 Watt to the load readings I got in Linux.
- In Linux, I haven't activated frequency scaling (available through Speedstep functionality in the CPU). Once activated, this should push down idle consumption considerably when running at 800 Mhz instead of the 2.7 GHz it's constantly running at for the moment.
- Need to double check sleep states support in Linux (C1, C2, etc.)