Complete system testing: Socket A or 939?

[pHaestus]

OK I am trying to get cracking on another promised Procooling feature: Complete system cooling tests. I have a full tower GTower en route from Cooltechnica. This case should be ideal for water cooling kit tests because it has 2 120mm blowholes precut on the top and it has plenty of space. I have to decide now whether Socket A is too long in the tooth for testing or not. My gut is that our readers are still something like 50/50 SocketA and everything else, but that number's completely out of my ass.

The advantage to using SocketA is I have a mobile Barton, ram, and can get a cheap mobo for the box. I also have lots of Socket A waterblocks and have completed many tests with those blocks as a baseline. This is potentially a pretty big plus I guess.

The advantage of going with 939 system is that it's more relevant and modern over the next year or so. I can tap into the internal diode with the 939 chip no problem so that's not an issue.

I have the money to upgrade the box in either case. An NF2 mobo will cost me ~ $90CAD after taxes and a 939 mobo + 3000+ CPU will run me something like $350CAD. If you have strong feelings for one or the other let me know. If not then I may go with Socket A just because I'm a cheap ass :)

[TbirdX]

Socket A for me :)

[bigben2k]

Gotta go 939; it's what's going on, sorry.

[jaydee]

Tough call. I will be using Socket A for a long time to come but many people interested in water cooling use the newer stuff... I would think Socket A will continue to dwindle and 939 will take off.

Edit: wonder how the IHS will effect temp monitoring over time...

[black_dante]

could the 2 hole mounting system of socket 939 boards be a factor?

[Blackeagle]

I voted 939, like Ben & JD note 939 systems are increasingly dominant and socket A is fading.

One of the few criticisms of your work I see sometimes at other sites is the use of socket A which is increasingly seen as yesterdays news. The 70w heat load is well below what many of todays overclocked rigs put out. A few come close to double this, and many have 50% greater loads. This leads to some questioning of how valid results with a 70w chip, which is also of very differant surface area/shape, really are.

For system testing work these concerns will increase.

I relize that a M-XP Barton well clocked will provide a greater heat load & larger surface. But it will still be seen as a outdated system by many.

I think that so long as you have to upgrade anyway it would be best to go with the more up to date system. And you'll still be able to use the same RAM & vid card.

[pHaestus]

I really wonder about those reported loads that new AMD systems supposedly dish out. My 939 system with retail hs/fan rarely gets above 50C at 2500MHz and 1.7V. My DFI mobo PWMs the CPU fan down below 50C and unless I run Prime95 at this speed it doesnt really spin up. Case is around 28C or so and so it's hard for me to imagine that this chip is generating the 100+W I hear bandied about.

[jaydee]

I really wonder about those reported loads that new AMD systems supposedly dish out. My 939 system with retail hs/fan rarely gets above 50C at 2500MHz and 1.7V. My DFI mobo PWMs the CPU fan down below 50C and unless I run Prime95 at this speed it doesnt really spin up. Case is around 28C or so and so it's hard for me to imagine that this chip is generating the 100+W I hear bandied about.
Does it even matter? With the type of testing your doing the results shouldn't matter what the heat load is?

[RoboTech]

939 gets my vote because it is more representative of current users. IMHO the relatively small, exposed die of the older socket A chips is not as representative for testing as the A64 with IHS. It's a good ave compromise (again, representative for Intel and AMD) of the majority of users. As you are well aware, the actual contact surface area (be it CPU core, IHS or thermal die) can have a large affect on the resultant test temps and may affect performance differently from one platform/system to the next.

I've been using an A64 3200+ for about 9 mos. for all my HSF testing... :)

Good luck with your new rig!!!

[pHaestus]

I REALLY like my A64 box: DFI NF4 ultra-d, 6600GT, and 3000+@ 2.45GHz or so. I am going to get a second 6600GT one of these days just for bragging rights.

[bigben2k]

Strangely, there's a question that remains unanswered here; what's the difference?

I mean yeah, there's more clamping on a 939, but that's because the IHS allows more.

The outstanding question still is: what is the effective clamping force on the core itself?

I mean, pHaestus, this might not even be an issue at all; for all we know, the clamping force may be exactly the same, or it may be closer to the max from Socket A specs, since an IHS would allow one to reach the max clamping force more safely.

Am I making sense? The core material is the same, right? So why would the clamping force change significantly?

[pHaestus]

In MY mind, a lot of the reasoning for the switch is marketing and perception ben

[jaydee]

If you did get the 939 it would be very interesting to see how much different the results would be from your Socket A results.

[leejsmith]

I am curently looking at what systems people are overclocking at oc.com xs and legit to see if it's worth upgrading my p4 northwood system.
A lot of people are overclocking a64 winchesters , fx's and the new venice chips on the nf4 systems and one thing i have found is a lot of them are saying that they cant overclock any higher with water over a good hsf and if they want to go higher then they need sub zero temps to do so.

if you use a s939 system to test i am sure a lot of people would be intersted in the results.

[j813]

I second Jaydee's comment, might as well do it since you can. Will be very interesting.

[Hoot]

939 all the way. You've already put together an excellent Socket A collection of data. Time to move on. Most 939s OC 25% with an OV of 15%. They are indeed very tolerant to temperature, which is a nice way of saying that it's hard to gain leaps and bounds by reducing the core a mere 10C. The exposed core and at least my motherboard will take the same pressure as recommended with the IHS in place. I love Swiftech springs with their internal limiting standoff. Easy to hit the same amount of screwdown every time. That's the ones that come with the MCW6002-64 I'm referring to. Not sure which other blocks they come with. Don't be surprised if you core temps go up from 10C to >20C differential to H2O intake though. Those 90nm cores are a whole new ballgame compared to Tbreds and yes wattage does matter. There are some blocks that do a Yeomans job at 70W and perform miserably on a 939 with 120+W. Just my opinions though.

Hoot

[dacooltech]

I agree with Hoot, 939 is the way to go man :)

[TerraMex]

939.

10char

[pHaestus]

OK I'm closing the poll now. Mostly because I just ordered a 3000+ Winchester CPU and Soltek KT800Pro mobo shipped to me overnight. When it gets here tomorrow or Wednesday I'll post some more of my system testing plans :)

[TerraMex]

why not nf3/4 ?
(something i missed?)

[pHaestus]

KT800 Pro and NF3 250 are really similar in performance and features. NF4 is nicer, but PCI-E only. I have settled on 9600XT as the video card to use for GPU block testing, and I have 2 of those lying around for the bench. So I needed an AGP board and the Soltek KT800Pro was about $30-40CAD cheaper than any of the alternatives.

Stuff arrived this am 8:30 unf.

[Hoot]

Make sure you float out those diode leads before the socket. Even if you can locate the dedicated components on the motherboard and remove them, as well as lifting the input pins to whatever onboard monitoring chip they use (Winbond?), the stray capacitance the traces add, if when combined with the 2200 or 2700pf input to the external reader circuit can go above the recommended maximum of 3300pf and make the readings a few degrees lower than they really are.

Hoot

[pHaestus]

I was going to bend the pins out from the chip (they are on the outside row) and cut a little gap in the socket so that they fit when the CPU is locked in place. I figured I could properly insulate and solder the wires to the pins then.

[Hoot]

That's what I started out to do, but wound up desoldering the two pins and soldering two short lengths of 30ga wirewrap wire to the dots. Just enough to clear the socket, then transition to miniature shielded two conductor. That way, you don't need to grind away too much of the socket. I did this several times before on Tbreds and believe it or not, the pins solder back on if you ever want to sell it. Just store them in a place where you won't forget.

Hoot

[pHaestus]

I've accidentally pulled pins off and had a helluva time trying to put them back on. What's the trick there? Maybe I'll play around a little with a dead TBred I have in the basement...

[DryFire]

gl with the 939 setup, I think a lot more people will be using 939, but I think the 775 would be good soley because of the heat it puts out.

[Hoot]

I've accidentally pulled pins off and had a helluva time trying to put them back on. What's the trick there? Maybe I'll play around a little with a dead TBred I have in the basement...

Yeah, the heat of soldering to them, even with my 17W pen is enough to unsolder them. When I reinstalled them, I put a tiny dot of liquid rosin flux on the tinned dots on the underside of the CPU. I then tin the flat head of the pins ever so slightly with some fresh .015 solder. I then mate each head to the tinned dot on the underside, tap it with the 17W pen and voila. Needless to say, a steady hand is in order. I clamp the CPU in a Panavise and set something next to it to actually reast my hand upon. Then I don the Jewellers Magnifier with all three magnifications lumped in. If you have a B&L Stereo Microscope, that is the best approach. Which one I use depends upon the mood I'm in.

Hoot

[Brians256]

When soldering small things, I cannot say enough good things about
1) refraining from caffiene for 24 hours before the operation
2) using clamps and waldoes to hold things while you work
3) putting hand-rests at convenient spots so that you don't get large muscle spasticity ruining your delicate work
4) move in between heart beats and just after a deep breath and exhale.

Without doing those, I'm more likely to punch a hole in whatever I touch with my soldering iron, especially because of my caffiene addiction.

Thank God I'm not your surgeon.

[pHaestus]

1) would result in my punching holes in someone at work or at home. Not an option :)

[Brians256]

I am also thankful that pHaestus is not my surgeon!

[pHaestus]

eh I'm a doctor of sorts

How hard can it be right? There's a manual (Gray's Anatomy) after all :)

[jaydee]

Didn't need that artery connected anyway! :D

[pHaestus]

To get an appropriate hdd ready for the new test system, I have to go through the following steps:

1) Partition/format/reinstall Windows on this here new Seagate 200GB for main system.

2) Pull out the 120GB from that box, format, partition, and place in my HTPC

3) Format the 45GB boot drive in the HTPC and use it for the new test system.

Damn this is too much like an IT job for pH's taste :)

[leejsmith]

why not use ghost or something that will copy the 120gb to the 200gb and the 45gb to the 120gb then you only need to load an OS onto one.

[j813]

How about the NF4? That will be an official world testing result? Expensive toy, any sponsors? :shrug: DualGT H2O cooled? :drool:

[pHaestus]

I decided to just leave the 120GB drive as is and replace the 160GB 2mb with the 200GB 8mb drive last night. That means I can put Windows MCE 2005 on the 160GB tonight and put it in HTPC. Yay? I wanted to upgrade from my cobbled together WinXP/myhtpc install on that box anyway...

[leejsmith]

have you tried sagetv i dont know how i would manage without it.
it took a bit of work getting the program guide working in the uk but i am very happy with it.

[pHaestus]

Updates:
I just finished installing windows, patching, updating drivers, overclocking, and began OCing/Prime 95 run with the new system. The CPU is the same stepping and week as the chip in my main box: CBBID 0502. Looks like ~ 2.52GHz at 1.65V is about all it'll do with retail hs/fan. The Soltek KT800Pro mobo doesn't adjust fan speed with CPU temp so the system is a LOT noisier than my DFI NF4 box. My plan is to run Prime95 overnight and then assuming the system is running ok then I'll do the diode operation tomorrow sometime.

I installed MCE2005 on Thursday and went and got the Microsoft Remote and a TV Tuner (NVTV) Friday. Wow is it ever slick and user friendly compared to the last time I used a TV Tuner! The remote was truly plug and play, and the recording functions are really easy and good quality. Live TV looks worse than the digital cable box by a fair amount though. So I just split the analog signal and have my HTPC and my digital cable on 2 different inputs of the tv. Next up is to get MAME running on that box I think :)

[pHaestus]

OK it looks like 2.52GHz is stable enough: Ran Prime95 for 13 hours so far. I'll pull that CPU out in a little bit and solder it up (wish me luck!)