Heatsink Deathmatch!!

[The Boston Dangler]

Hi!

I'm putting the word out that I'm conducting analysis of the elusive Xigmatek HDT-SD964 CPU cooler, and comparison to the Zalman CNPS9500 LED. I hope to have the project written and posted in this forum by the middle of next week. There?s a good chance this will be the first published review of the Xigmatek in North America! I received the new cooler yesterday, and time is kinda tight right now. Early testing so far has been very promising.

The testing is done on my only computer, with an every-day installation of Vista Ultimate 32. I am re-running my initial tests, and hope to explore every possible configuration of available cooling hardware. I?m using synthetic tools to optimize a real-world personal computer, but I'd like to include some normal usage results as well as worst-case scenario.

Hardware:
AMD Opteron 165 (overclocked and undervolted)
Asus A8N32-SLI in stock form, BIOS 1303
Corsair 2x 1GB DDR400
XFX 7950GT (factory OC at 570 MHz, passive cooling)
Silverstone SST ? ST60F power supply with Noctua 120mm fan
Silverstone LC-20MB HTPC case with 2 Nexus 80mm fans.

Software:
Windows Vista Ultimate 32, all updates, some tweaks
Latest official Nvidia video and motherboard drivers
Prime95 v25.5
Rthdribl v1.2
Speedfan v4.33
Core Temp v0.96.1
Nvidia ntune 5.05.54.00

There are 4 mounting options for the Zalman, and 2 for the Xigmatek on S939/AM2. AMD mounting hardware is poor to begin with, but the Xigmatek hits a new low. I expect no such difficulties if the cooler was installed on LGA775. For each heatsink, tests will be run with the case lid on and off, for normal and maximum results. Video temps are taken (poorly) with ntune, while running rthdribl at 1280x720, and Prime95 loading the CPU. Not only does this reflect video game usage, but I?m able to generate a slightly higher temp with the card being blasted with warmer air. CPU and northbridge temps are taken with Speedfan, while running only Prime95. I have a Fluke DMM with a type-K and an Extech IR thermometer I can throw into the mix. There?s an Extech SPL meter around here somewhere, but I don?t think I?ll get much use out of it. I already know the Zalman is loud, the Xigmatek is better, and 99% of the time, my PC is well below the SPL meter?s 40dB threshold.

For the review itself, I?m including links and pictures, but no audio, video, pie charts, flow charts, nor flash cutaways. We?re focusing on meat and potatoes here, folks. If all goes well, I might slap together a couple more articles. I plan on hosting pics at pics.bbzzdd.com, but if someone has a suggestion for that, or anything else, please let me know.

Happy Valentine?s Day. Who loves ya, baby? That?s right,
TBD

 

[aigomorla]

Looking forward to this.

Wish you had a Tuniq Or Ultra120Extreme to base results off.

Just make sure you load up the processors for at least 15 min so the sink has enough time to reach max Delta Air / Sink.

 

[The Boston Dangler]

Good news, everyone. I can easily mount the Xigmatek cooler with a bracket from the Zalman. The retention bracket from Xigmatek is absolutely horrible, I nearly broke the black frame attaching the cooler to the clips. Even unlocked, the bracket was tight enough to secure the heatsink. The Zalman bracket fits perfectly with Asus' stock frame and backplate, including the stock screws.

Disaster struck last year when attempting to mount the AMD version, 9500AT. The black tabs broke off, and in desparation, I returned it for the CNPS version. It was great luck that it worked, and now at least doubly so.

 

[The Boston Dangler]

Good news, everyone. I'm making a couple changes which will speed things up and lessen the chance of breaking something. First, I'm cancelling one of the Zalman tests. The config has the Zalman blowing back-to-front, with the rear case fans blowing in. I tried this north to south setup before, and it DOES NOT WORK. The second change involves leaving out the 500GB hard drive and it's cage. This is totally a hassle during teardown and rebuilding, and isn't needed to run the tests. I expect no impact on temperatures from the cage's omission, as there is only a small, obstructed, almost useless vent that is affected. Smoke shows roughly the same draw at the vent as before.

I forgot to mention earlier I am using Tuniq TX-2 TIM. This stuff is great - good conductivity and no curing time.

 

[toadeater]

Ah, that's the 4-pipe Xigmatek. Should be interesting.

 

[DerwenArtos12]

Where'd you manage to find one? The 3 pipes are all over the place but, can't find the 4 pipes in stock anywhere.

 

[The Boston Dangler]

http://www.gezekian.com/index....h/previd//prevstart/0/

seems to be the only shop in north america carrying the cooler. why? i don't know.

 

[DerwenArtos12]

Only thing I don't like about their 92mm series is they're only using 6mm heatpipes, the 120mm series uses 8mm heatpipes. Here's to hoping they come out with a HDT-S1284!

 

[The Boston Dangler]

Good news, everyone. Testing for today is over.

screenshot

The Speedfan chart shows the completion of open lid cpu testing and the beginning of the closed lid test. The config is Zalman/west to east. Air is drawn in from around and through the video card and exhausted by the 2 80mm's. I had expected more from this setup. The video temp of 61C is the best so far, but the cpu hitting 66C is obviously unacceptable.

Load......Closed Lid......Open Lid......Ambient: 20C

CPU.......66! (+46).......59 (+39)
NB.........44 (+24)........49 (+29)
VID.......61 (+41).........73 (+53)

 

[DerwenArtos12]

What TIM did you use and what cure time did you give it?

 

[The Boston Dangler]

Originally posted by: DerwenArtos12
What TIM did you use and what cure time did you give it?

Originally posted by: The Boston Dangler

I forgot to mention earlier I am using Tuniq TX-2 TIM. This stuff is great - good conductivity and no curing time.

for the zalman, i lay a small "grain of rice" bead down the center of the chip, a little wiggling, then gently tighten the screws to positive stop.

for the xigmatek, i'm having success with laying a thin-as-possible bead down the width of each pipe. i tighten one end (ends up beneath the fan) of the zalman bracket all the way, while holding the cooler off of the chip. once the first screw is tight, i slide the cooler into position, make contact with the chip, and thread the second screw. after a little adjustment, the second screw is then tightened to positive stop. using the grain of rice or dollop or whatever results in a thumbprint-looking smudge in the center of the joint, missing one pipe entirely.

i'm using 99.953 isopropyl and white paper towels to clean the surfaces. once contact is made between joint surfaces, i tolerate some sliding and twisting motion, but if contact is broken i clean the surfaces and start over.

 

[DerwenArtos12]

Originally posted by: The Boston Dangler

Originally posted by: DerwenArtos12
What TIM did you use and what cure time did you give it?

Originally posted by: The Boston Dangler

I forgot to mention earlier I am using Tuniq TX-2 TIM. This stuff is great - good conductivity and no curing time.

for the zalman, i lay a small "grain of rice" bead down the center of the chip, a little wiggling, then gently tighten the screws to positive stop.

for the xigmatek, i'm having success with laying a thin-as-possible bead down the width of each pipe. i tighten one end (ends up beneath the fan) of the zalman bracket all the way, while holding the cooler off of the chip. once the first screw is tight, i slide the cooler into position, make contact with the chip, and thread the second screw. after a little adjustment, the second screw is then tightened to positive stop. using the grain of rice or dollop or whatever results in a thumbprint-looking smudge in the center of the joint, missing one pipe entirely.

i'm using 99.953 isopropyl and white paper towels to clean the surfaces. once contact is made between joint surfaces, i tolerate some sliding and twisting motion, but if contact is broken i clean the surfaces and start over.

ah, yeah found that now, sorry. for the base of the xigmatek you should probably tint the base just to make sure you're getting adequate TIM around the heat-pipes and in teh tiny cracks between the base and heat-pipes.

 

[Rubycon]

AMD Opteron 165 (overclocked and undervolted)

Torture testing of heatsinks should provide a challenge to them. Using this for a load especially with heatpipes presents a challenge similar to playing Solitaire for a cpu stress test.

For overclockers a 200W load should be used to weed the men from the boys. At the highest levels you will also start to see a difference in operating position as the heatpipes lose some effectiveness. This is controversial, however but has been my experience with the three piped Tuniq. The TRUE did not experience this runaway condition when mounted horizontally. QX6700 B3 3.6GHz at 1.485VCORE. Prime95 stable at 80°C with motherboard horizontal. With motherboard vertical the system shut off (TJ limit reached) after just 5 minutes of Prime95 small FFT. Core temp reached over 100°C all cores. Mounting pressure exceeded 75 psig and the heatsink was braced to avoid the possible effect of gravity pulling it away from the IHS.

Speaking of HS mount pressure, here's an interesting tidbit of information here.

This is of concern with TRUE owners as the stock mounting bracket may NOT provide sufficient hold down pressure particularly if the base has been lapped.

There's an ongoing discussion here about this very issue.

Sorry OP not to derail your thread - your efforts to test these products are always appreciated and should prove useful for those with "normal" systems. Heatpipe based heatsinks, however most certainly have a limit in which once approached will allow core temps to rise rapidly. If graphed the curve is nearly hockey stick shaped. To the novice this may suggest seeking help or remounting frequently with the same results or condemning the darn thing!

The Tuniq does so well because it has a huge amount of surface area on its fins. However when the pipes cannot effectively transfer the heat, deltas will increase. The base on the Tuniq, however rough it may look (insert comment about it appearing to have been set on a belt sander here!), it's very flat (true). My TRUE, OTOH, was quite convex and I had it milled to 0.00001". Did didly to temps but all cores are within 2 degrees max of each other whereas before there was a good 6 degree spread. I was too chicken to hand over my QX9650 to the machinist for truing, but I don't think it would be worth it. I understand even removing the IHS completely does little for temps. Perhaps for a watercooler it would work as the waterblock could be machined to fit over the chip touching its naked cores directly. It's getting to the point where the cores themselves need to be made more (heat) conductive! :Q (is that possible?)

Anyways...keep up with the good work!

 

[The Boston Dangler]

Originally posted by: Rubycon

AMD Opteron 165 (overclocked and undervolted)

Torture testing of heatsinks should provide a challenge to them. Using this for a load especially with heatpipes presents a challenge similar to playing Solitaire for a cpu stress test.

I'm just working with what I have (one computer I've spent a lot of money on, my living room, and some spare time). Since I'm able to reach dangerous temperatures with the existing setup, I don't really care if the heatpipes stall at 95C, as my pc will be dead by then. Currently, the chip is putting out approximately 100W. I think the max I could generate is about 140W. Vertical orientation isn't possible, since the PS intake would be flush agaiinst the table.

A note about readings:
I am recording what Speedfan calls the K8 sensor, not Temp 1. K8 is a sensor within the core(s), Temp 1 sits on top of the cores under the IHS. Core Temp (mostly) agrees with the reading on K8. On a very consistant basis, Temp 1 is roughly 5C hotter at idle, and 5C cooler at load. I think a lot of people are relying on the Temp 1 sensor, so if that's you, go ahead and subtract 5C from my CPU temperatures.

I'm operating under the assumption Temp 2 is the northbridge. I haven't found any documentation to support this. It could be the power regulation or anything else, but I'm sticking with northbridge.

Temp 3 seems to be useless. It reads -128C (thermistor), and 39C (diode). The reading doesn't seem to move.

For video temps, I would prefer NVTempLogger, but it doesn't work with Vista.

 

[aussiestilgar]

Originally posted by: The Boston Dangler
http://www.gezekian.com/index....h/previd//prevstart/0/

seems to be the only shop in north america carrying the cooler. why? i don't know.

frozencpu.com stocks it too.

 

[The Boston Dangler]

Originally posted by: aussiestilgar

Originally posted by: The Boston Dangler
http://www.gezekian.com/index....h/previd//prevstart/0/

seems to be the only shop in north america carrying the cooler. why? i don't know.

frozencpu.com stocks it too.

that's a different model, the xp-s964. i will touch upon the differences in my review.

 

[The Boston Dangler]

Bad news, everyone. It seems I have invalidated existing data by leaving in a PCI blank. This was choking either intake or exhaust, and depending on setup, had a large effect. Re-testing might push release back to next weekend. The good news is I've refined my testing technique to be as quick and accurate as possible.

edit:
is it worth performing each test twice? so far, i've performed most tests twice, with minimal variation in temps. i'm talking about 1C, if that. since i seem to have the mad mounting skillz, i think i can cut the testing time in half. if there is an aberrant result, i'll perform an autopsy and do-over.

 

[The Boston Dangler]

my testing methodology:

0 min: power on, no lid
5 min: idle (quiet, approximately 7V) temp taken, go to 100%
10 min: idle (100%) temp taken, start prime95 and rthdribl
30 min: cpu, nb, vid (load) temps taken, kill rthdribl
50 min: cpu, nb (load) temps taken, compared to previous temp, keep higher number, put the lid on
60 min: cpu, nb (load) temps taken, restart rthdribl
80 min: cpu, nb, vid (load) temps taken, highest numbers kept, loads removed
90 min: idle (100%) temps taken
100 min: idle (quiet) temps taken

edit:
CoD4 will be added to every test

 

[The Boston Dangler]

Good news, everyone. In an effort to separate the men from the boys (something Rubycon is pretty good at ) I've decided to raise the OC from 2.25 GHz to 2.5. Voltage comes way up, too. The only problem is: as cpu speed increases, the integrated memory controller craps out at lower thresholds. 2.5 GHz and DDR360?

screenshot

 

[DerwenArtos12]

Originally posted by: The Boston Dangler
Good news, everyone. In an effort to separate the men from the boys (something Rubycon is pretty good at ) I've decided to raise the OC from 2.25 GHz to 2.5. Voltage comes way up, too. The only problem is: as cpu speed increases, the integrated memory controller craps out at lower thresholds. 2.5 GHz and DDR360?

screenshot

Hrm. What memory at what voltage. I don't see any reason that it should need to be that low, especially set to 2T timings and so loose at that. I can run my OCZ plat 2's to 528 w/ 3-3-3-8 1T with 2.7v.

 

[The Boston Dangler]

CORSAIR XMS 2GB (2 x 1GB) TWINX2048-3200C2PT

2.9V

at a 1:1 ratio, this memory is flawless at 225 MHz, 2-3-3-8 1T. i loosened the timings when going up to 2.5 GHz and haven't bothered to fine tune the memory.

besides the obvious, what does this mean and/or do for me?

AI Clock Skew for Channel A/B [Auto]
Current Clock Skew [Delay450ps]

 

[DerwenArtos12]

Originally posted by: The Boston Dangler
CORSAIR XMS 2GB (2 x 1GB) TWINX2048-3200C2PT

2.9V

at a 1:1 ratio, this memory is flawless at 225 MHz, 2-3-3-8 1T. i loosened the timings when going up to 2.5 GHz and haven't bothered to fine tune the memory.

besides the obvious, what does this mean and/or do for me?

AI Clock Skew for Channel A/B [Auto]
Current Clock Skew [Delay450ps]

not much but with a 450ps delay you should be stable at reasonable timings. You're pretty safe to leave those at auto unless you're pushing past DDR500. By DDR360 do you mean the 8:10 divider because if so you're actually running the ram at 252mhz, you should try dropping it down to the DDR333 setting at least for 233 or 300 for 210 on a 280fsb.

 

[BlueAcolyte]

I think the Xigmatek wins.

 

[n7]

Originally posted by: BlueAcolyte
I think the Xigmatek wins.

I think so too

But nice to see a vs. anyway

 

[The Boston Dangler]

Good news, everyone. The Zalman testing is complete. I'm very confident in the quality (and quantity) of the results. With luck, I might be able to finish the Xigmatek tests in a day or two. If I can get my box propped up on it's side without damage, I'll run a vertically-oriented test with each cooler in it's previous-best setup. I have decided against testing the Xigmatek with 2 Nexus 90mm, as these won't even come close to cooling my comp.

Sneak Preview (these are the Zalman's best results)

Zalman, south to north (blowing towards the back of the case), Ambient - 19C

Synthetic Load........Lid............No Lid
CPU........................62.............59
NB..........................43.............44
VID........................75..............92

CoD4.....................Lid............No Lid
CPU........................55..............50
NB..........................44..............43
VID.........................77..............100

Idle........................Lid..........................No Lid
CPU fan............60%....100%...............60%....100%
CPU..................40........38...................36........35
NB....................41........39...................40.......42
VID...................53........50..................69.........65