New rig processor quandry...

[Albuquerque]

Woke up this morning to Intel BurnTest v2 maximum load setting having completed 10 passes without issue. Processor is doing 4.5Ghz / 1.25x strap with all threads/cores active, set to 1.300v and "mid" vdroop correction; 74*c under a Zalman CNPS 9900 Max with a single 120mm exhaust fan and the case-side 250mm intake fan. Lots of possibilities for better cooling in this case, but I'm ridiculously happy with this result and noise level.

Task today after work is to go beat up the eight sticks of ram. They're all running VERY relaxed settings for now, something like 9-9-9-27 at 1333Mhz. I'm pretty sure I can get 8-8-8-8 timing at 1600 or better.

Power draw from the wall with the processor at full whack was ~335W. Looks like I'm not going to have any problems with my existing powersupply.

 

[Blazkowicz]

damn these times I'm doing browsing only with firefox eating all of my 2GB memory.
I'm planning a great upgrade, get an asrock 970 extreme 3 mobo or something, and 8GB ddr3 while keeping the same athlon X2 but this time I can overclock it. (upgrade from AM2 to AM3)

btw you could probably have gone with a 990FX mobo, only an FX is quite less sexy unless you wanted a space heater.
it would have been funny to run an o/c AMD FX and GPGPU folding during the harsh cold wave (but it's gone)

 

[Albuquerque]

Ye ol' 990FX isn't much competition for the X79 except for the space-heater aspect

 

[Grall]

Woke up this morning to Intel BurnTest v2 maximum load setting having completed 10 passes without issue.

Sweet. Grats on the new rig!

I'm pretty sure I can get 8-8-8-8 timing at 1600 or better.

It's weird with the timing on RAM sticks; I've got a set of 6 OCZ Platinum 2GB DIMMs that I bought years ago now and are specced at 7-7-7-24@1600, and I run them 7-7-7-20@1660 no issues. All sticks I see these days are higher timings, even for a relativley paltry 1600MHz...

Especially sticks made for low-voltage have worse timings, so there's some sort of physical correlation going on there.

Power draw from the wall with the processor at full whack was ~335W.

Whoah. That's pretty rough just for the CPU! I remember when PC processors first started getting heatsinks on them instead of just a bare (ceramic) chip carrier; PASSIVE ones initially. Then those dinky little 40mm fans that gummed up in a couple months' time, and from there things just grew and grew.

Crazy now that I'm thinking of it... Of course, performance has grown a LOT more than power useage, so it's not a very straight comparison really.

 

[DuckThor Evil]

Whoah. That's pretty rough just for the CPU!

It wasn't just for the CPU, but total system with CPU at full load. How would you measure just the CPU power usage from the wall?

 

[Albuquerque]

Yeah, keep in mind that IBT soaks up pretty much all available ram (when placed into "Maximum" stress mode) so it's pretty much maximum power draw excluding your video subsystem. I just played with that last night, and my 5850 overvolted and overclocked running some FurMark Xtreme Burn + the CPU doing IBT at max yielded a whopping 500W power draw at the wall!

The ram timings are limited by having all eight channels populated. If I pare it down to four sticks, I can hit 7-9-8-24 at 1T (the XMP profile for these Mushkin sticks) on their rated 1.65v. However, with all eight sticks in, I can't maintain 1T at any speed My current stable point is 8-9-8-22 at 2T running 1666Mhz while still maintaining the 1.65v on memory voltage and stock volts to the CPU IO (VCCIO) voltage.

Later tonight I'm gonna poke around with the IO voltage, looks like I have a lot of room to move before it starts becoming "too much." According to most tweak guides, that IO voltage is what normally holds the keys to allowing me to squeeze a bit more on the memory speed and timings.

 

[Grall]

It wasn't just for the CPU, but total system with CPU at full load.

You're mincing words. "The full system" (minus GPUs) is just a bunch of support systems for the CPU. Chipset, RAM and so on are all CPU sub-systems without which the CPU could not do its work.

It's easier to just shovel all this crap in under CPU power useage rather than write a small essay describing and defining exactly what I mean.

How would you measure just the CPU power usage from the wall?

I don't even think that's interesting. Actual CPU-only power use will vary (often quite a lot actually) from board to board even with the same CPU, as seen in Anandtech motherboard reviews for example, depending on VRM and board design and such factors.

IMO, it's the whole package that's interesting, how much the CPU (and related systems) draw as a cohesive, whole unit. "A CPU is not an island", to paraphrase a common saying.

 

[Grall]

However, with all eight sticks in, I can't maintain 1T at any speed

Hmm... I can run 1T on my rig with all 6 sticks on my now soon ancient Nehalem plug... I'm really pleased with that myself actually (although actual performance benefit may be nigh unmeasurable in normal use of course....) Is it common for SB-D chips to not manage 1T rate with 8 sticks, or is it as much a function of the mobo/BIOS as of the sticks themselves?

 

[DuckThor Evil]

You're mincing words. "The full system" (minus GPUs) is just a bunch of support systems for the CPU. Chipset, RAM and so on are all CPU sub-systems without which the CPU could not do its work.

It's easier to just shovel all this crap in under CPU power useage rather than write a small essay describing and defining exactly what I mean.

Well I'm quessing his other components used at least 100w in that scenario, so it's not insignificant amount. At stock 5850 alone already idles at 30W. Eight RAM sticks and six SSDs and what else does he have in there. The CPU don't need that much sub-systems

 

[Grall]

Right, the SSD RAID array... I forgot about that. Still, the vast majority of power should be consumed by either the CPU or devices directly related to the same.

 

[Albuquerque]

Well, to be quite fair though, the SSD's weren't under any load. At idle, they probably use less power all together than a single spinning disk would. The RAID card needs a few watts I'm sure, but it too probably isn't doing much during the CPU stress portion of the test. At full torture-test load under AS-SSD or something, I do wonder how much power they might be drawing. TBH, I bet it's the RAID card that sucks more power than the SSD's do -- I think they're all rated for like 3 or 4 watts at peak.

I'll have to keep an eye on the kill-a-watt when I start doing disk benches, just to see what the result is. Hmm...

As for the 1T? It was pretty common even on the Nehalem products; I've got two more super-geeks at my office with i7-920 and i7-965 rigs that swore off the full six-sticks of ram config because they couldn't keep 1T. I am more inline with what you said though; the performance delta is likely negligible to non-existent. It's not like the quad-channel bus is starving the CPU of bandwidth, really...

I'll attempt to get some figures for the SSD array and post back with my findings, as now I'm curious...

 

[Albuquerque]

Got some starter benchmark numbers; ATTO saw at least two hits at the 2.5GB/sec mark and Anvil Storage bench saw a ~6,400 score. The rig is all at default speeds for these runs as I'm having problems maintaining my overclock on my existing power supply when the whole rig is under load. The replacement Kingwin Lazer Platinum 1KW unit is arriving tomorrow to solve that problem, and I'll be able to start tweaking a bit more and provide some pictures, better benches (hopefully), and power draw figures.

I had never heard of that particular power supply builder before, but it got phenomenal reviews from JonnyGuru (linked) whom I trust to make the proper assessment of PSU quality.

 

[Albuquerque]

For those who are curious, the six SSD's and raid card peak out at approximately 60W under full load while doing these benches (comparing idle power draw to maximum that I saw indicated.)