Computer Build #5

Right, this will be the final instalment until one of 2 things happen:

1) I decide not to get water cooling (in a month or 2) and make the outside pretty.

2) I decide to get water cooling and replace the drive bay covers.

If I decide to pretty it up now and get water cooling, I will not need that big fan at the front - or I *may* not need it, so I would quite like the case to go back the way it was, looking all normal at the front. Though I might decide to keep that fan. Actually I probably will, but I don't want to make a rash decision, so there is no rush to pretty this up, so I can leave it as it is until I make a final decision.

The advantage of the water cooling - 4.5GHz. Need I say more?

The purpose of this post is to show off the lighting. I couldn't resist putting a cold cathode lighting setup in this - I figure that if the chip catches fire, I may as well have the case glowing red so that I don't notice at first thus lessening the blow of my i5 processor going down the toilet, hahaha! I am usually not one for "speed stripes" or "things" but I just wanted to do this, after all I did build this with an angle grinder, so here goes.

2 RED CCTs, mounted to the side panel - the tubes cannot easily be seen here, so aesthetically this is the best choice.

This is how they look when running:

This is the wee transformer. Note the black switch (out of focus due to wide aperture for light) - it is huge. Where exactly did they think it would fit? Was I going to drill a 12 or 13mm moutnting hole in my case? Seriously guys, thanks for supplying it as a quick fit, pre-wired job, but that switch is a real huge ugly brute.

All set and whistlin' dixie!

Here is a picture showing a little... unconventional wiring. Well it is conventional in the sense that + to + and - to -, fingers in the mains socket standing in a pool of water barefoot next to a lightning conductor on a stormy day, but since I ran out of fan jumpers, I pushed single core wire into the back of one of those 3 pin plugs to create the connection. People are too precious about the power drain through these. News flash: the fans are about 2 watts each... I'm not loosing sleep. I know, I know - I can't control the fan speed by jury-rigging them, but my motherboard only supports speed control on the CPU fan and *maybe* 1 peripheral fan. I have speed control disabled - I want these working hard :D

That is all for now, computer wise. Here are some things that I plan to do in the next month:

Collect a bandsaw (all being well, tomorrow! Amazing to have found one, what a God-send)

Chamber ensemble and Indian (Haydn, Handel and Bhuna... Friday night sorted!)

Design some more instruments

Continue preparations for "Sweeney Todd" which will be performed in the Mill Side Theatre as the NI Première (Theatre at the Mill). This is running from 2/2/10 to 13/2/10. Again, on my birthday I will be playing trumpet, and I wouldn't change that for the world :)

Write more music - I plan to write maybe a head-chart a week and post it to my blogger as inspiration, I might start with one I have already written to give me some buffer time!

I will have to clear a workspace in the workshop. I have been preparing for this day... I am still delaying! Man alive. Could it be more messy? The answer is no, and I will post a picture soon, no-one will expect the level of mess. Only people who have seen it will believe it.

Keep to a better sleeping routine: 8-9am bedtimes and 1-3pm getting-ups is horrible, I feel like such a waster when I do this and I hate waking in the dark evening! Last night I managed a 12am bed, sleep by 2am, up with no alarm clock at 9.30am. More like it!

That is all for now, I won't continue an endlessly boring list for no reason, besides there are things which I don't want to be accountable for, like clearing the back drain - but I did that already. Horrible day that was!

Here's hoping! (hopping)

Mike

Computer Build #4

This is an update concerning the cooling solution on my computer. By messing with the order of fans and the case layout a little I have managed to drop the overall temps by 6-10*C. The main reason for this is airflow, or the lack thereof. The intake vent on the front panel of the case is abominable! It was providing no throughput, so I decided to take out the 5.25" bay windows, cut the spacers away and install a fan there.

Now I did every iteration of fan layout possible, measured my temp tests under all the circumstances and what I now have is probably the best I will get short of replacing all the 120mm fans with that Akasa Apache which kicks out 60CFM. Also, the next logical step is water cooling, but I will only venture there if the summer's heat drives me to (haha, NI heat) or if I want to bump up to 4.4GHz.

Onwards. I am currently running the PC in a room which is getting warm (thanks fire) and I will be able to stress test as-if the summer was here, and actually warm. More on that later (at a cursory glance the temps have gone up a few degrees, but not much.

What I did:

CPU cooler back to normal (ie 92mm fan).

5.25" drive bays opened, Akasa Apache 120mm fan installed sucking in.

Rear 120mm fan extracting.
120mm fan (perpendicular to the rear case fan, sucking air through the CPU cooler and out the PSU fan).

So in that burn test I have it from a peak of 82*C down to a peak of 74*C... much happier now!

In any case there is excellent air throughput now! Some temps for you (Idle):

Before (at new 4.1GHz clock):

Case: 21
CPU: 26
C0: 36
C1: 32
C2: 33
C3: 29
GPU: 35

After (also at 4.1GHz clock):

Case: 19
CPU: 20
C0: 30
C1: 26
C2: 27
C3: 21
GPU: 30

In the following test the 5 temperatures are the temperatures of the CPU die for each iteration of the test of which 5 were run.

Before (Improved Vcore, sub-optimal fan setup!)

Burn test (CPU temp):
53, 63, 66, 67, 67 with a core peak of 78

After (Improved Vcore and fan setup)

Burn test (CPU temp):
59, 62, 63, 63, 64 with a core peak of 74

Now some nerdy things. (turn away if numbers give you dizzyness!) Bear in mind that measured voltages don't often 100% match set BIOS voltages, so U try to give an account of both where applicable.

Vcore: 1.31v measured by CPUID (BIOS set @ 1.318750 I think)
Vin1: 1.62 - measured by CPUID (BIOS was set to 1.64)

Bclock: 216MHz
Multiplier: 19x
DRAM multiplier: 6x
Core Speed: 4104MHz
DRAM Speed: 1296MHz
DRAM Timings: 9, 9, 9, 24

All extraneous power BIOS power functions, power save, turbos etc etc... turned off.

My previous settings were something like (for the 4.1GHz clock):

Vcore: 1.36v measured by CPUID (BIOS set @ 1.3618 I think)
Vin1: 1.62 - measured by CPUID (BIOS was set to 1.62)

So initially I got the Vcore as low as possible by lowering by a step, testing, repeat. When it failed to boot and was unstable I bumped it up a setting or 2 and stress tested. Thermal implication is that lower voltage is obviously going to heat the chip less, and even though we are changing millivolts here, it really matters! I got ~2*C off the chip temp by getting the Vcore slimmed down! What I did find was that with the Vcore at its most efficient setting, I had to bump the Vin1 for the RAM from 1.62 to 1.64V. Note that the ram will work fine at 1.65V. (my Gigabyte m/board adds in .02 at a time, so 1.64/1.66/1.68 etc... but 1.64 is working well.

To break this down into manageable data:

CPU speed is a by product of the FSB x Bclock multiplier. Vcore supplies the voltage/power for the chip. This is nominally AUTO, but when clocking you want control of it.

On the i5 1.35-1.4v is about a high as they recommend pushing it.

RAM speed is a by product of FSB x DRAM multiplier. Vin1 supplies the voltage/power for the RAM chips. Again, this is AUTO, but in clocking you need the control to give the RAM more power when needs be.

You change the FSB to accommodate a higher clocking and since the CPU and DRAM are affected by this you may have to settle for a lower RAM speed.

It is recommended that the Vcore and Vin1 are kept within 0.5v of each other.

At 4.1GHz my CPU benchmark was 7115.5 which I am dead pleased about! I ran a bunch of stress tests. OCCT for 4 hours, Prime95 - 2 hours on large FFTs, 10 mins on small FFTs (I read that 10 mins should show whether the system would cope or not), hours on blend. Intel Burn Test - I ran the default 5 cycles, and it pushed the chip hard, but it was fine. I also ran SuperPi and LinX.

I am monitoring heat with:

SpeedFan
Core Temp
CPUID

Real Temp

Stress testing with:

OCCT

Intel Burn Test

Prime95

LinX

SuperPi

Benchmarking with Performance Test 7 on evaluation.

The settings which REALLY push the processor are the small-FFTs. I assume that stands for Fast Fourier Transform, and there endeth my knowledge on that matter! Also LinX gets her warm!

Now in the future I can see me running a water-cooled unit to get that 4.4GHz, but until then I may as well do some work on this instead of endlessly testing it. Currently a faster computer has led to less work done than ever (except computer faffing and searching for bandsaws...).

Finally, I have the black plastic 5.25" drive bay overs to cut up and make the front fans look much neater - everything is a bit "temporary" right now, and it will eventually look neat! That black tape on the side fan will be cut into a circle soon.

That ought to do it for now, I must process some photos and try to find a 14" bandsaw somewhere. I forget that we are in 2010... and I was born *after* imperial was the norm, so I should really be talking MMs more! 350mm would do :) I blame dad for my use of both systems (not totally illogical - lots of the guitar gear is american, and sometimes inches just work neater!).

God bless all,

Mike