HD2900XT benching

[1Day]

Over the weekend I had a opportunity to do some benching (Child minder on overtime :lol: ) but since the summer is certainly here on the High Veldt with ambient midday temperatures where I live in the mid thirties benching at a hectic CPU clock speed on air is pretty hard to do. So having to accept that 4.1 GHz was the highest stable limit I could 3D bench on my Dfi Lanparty UT P35 T2R and my Tuniq Tower 120 I needed to look elsewhere for a source of benchmark improvement. Now those that know me are aware of my love for the HD2900XT, and while I have a card that is supposedly a better card I get far more pleasure out of benching the Ati graphic card than the nVida offering.

But again temperature is an issue; it is all very well for Marcus to calmly state he can clock his HD2900XT at 1000 MHz on air – when your ambient is low. Try doing that here in Pretoria and you will have a cooked card. The only WC stuff I have is a Maze4 block and a triple rad and pump that was given to me by a friend ages ago. So as they say in South Africa “n’ Boer maak a plan” [Literally - a farmer makes a plan]. Within hours the HD was in pieces and a Maze4 block affixed and the odds and sods of tubing that James had given me attached and ready to go. But not with fans attached because what would the point be of blowing 36’C air through a rad? Nope I stuck it in a plastic crate of ice water. Cranked the volts up of the graphic card to 1.4 volts and played with AtiTool 2.7b.

http://img490.imageshack.us/img490/2720/dsc04897jx0.jpg

This is my attempt at “extreme” cooling of my graphics card – compared to stubby it is quite tame. I had planned to use Dry Ice 3mm pellets and a mix of acetone and water but found that in testing the materials the tubing became quite useless and jello like. Not good so in the end it was just ice water and Dry Ice

http://img301.imageshack.us/img301/9713/dsc04898bf2.jpg

The TT120 dwarfs pretty much all on the Dfi Lanparty – which is a both a joy and a nightmare to tweak; a joy because it is so stable voltage wise and you really do not need to do a thing to it to overclock the [insert appropriate word here] out of your kit and a nightmare because there are just so many options to fine tune.

http://img301.imageshack.us/img301/8408/dsc04899hn9.jpg

My test bench and bits of wood and the plastic crate with the cold stuff inside and of course the single most important benching aid, the ever present cable tie.

With the current lack of any utility that would allow me to edit the bios of my Radon card, thanks for nothing Ati, I had to rely on third party applications to make voltage adjustments and to overclock the card for each run. W1zzard has provided enthusiasts with a means of adjusting voltage but unfortunately it is a beta release and does not allow making multitude of adjustments to the clock speed once set as it resets to default if you subsequently adjust clock speed or memory frequency. So you basically have to set the voltage at any random setting and then load your clocks and get back to the voltage setting page as quickly as you can and adjust voltage to the desired actual settings. In this instance I set my Vcore to 1.45 volts. If you are not smart about it your card will freeze up as the volts set do not stay fixed. After much testing of the best – read workable – clock to memory ration I began to bench.

In no particular order here are the results of an hour’s benching – all personal bests for me with the HD2900XT in all of these benchmarks and three PB’s for all graphic cards.

http://img301.imageshack.us/img301/5541/3dmark0347178nv1.png

The 3Dmark03 benchmark is the least friendly towards the Ati Radon card and prior to today I have never been able to get past 45K with the 2900XT, so I was very happy at this result.

http://img146.imageshack.us/img146/4256/3dmark0525409tq8.png

This benchmark on the other hand is very HD2900XT friendly and I was very happy to set a PB.
http://img528.imageshack.us/img528/8369/3dmark0614179pc1.png

These last two benchmarks really surprised me, I had expected the Ultra to be much stronger here and was stunned when I first set a PB in 06 and then improved my PCmark05 score by quite a bit.

http://img507.imageshack.us/img507/2548/pcmark059719lt4.png

All in all I had a good afternoon but would love to get to grips with this graphic card with a stubby of my own. So if anyone has a detailed drawing of a simple Dry Ice graphics pot could you please share it with me (thanks Jay I had a look and understand the principle).

[boshuter]

Very nice :thumbs: I've been telling harley and shim that the 2900 is going to be my next card for benching for quite a while ;) If they ever get the oc'ing and crossfire straightened out, they will be hard to beat (not to mention the fact that you could by 2 of them and have enough cash left for a decent motherboard compared to the Ultra) ;)

If you have a Maze 4 gpu block, you could make a very good dice pot easily. I use the Maze 4 base for evaporaters in phase change systems, and the design would lend itself very well for dice also. The base on them has rings in it that a 1 1/2" copper cap just fits into, making it very easy to braze the cap to the maze 4 base; you could do the same, using a copper elbow instead of a cap, then just braze a copper pipe to the elbow and ..... Viola.... a gpu dice pot :yup:

If you need any help with this... just lmk :)

[1Day]

Boss consider yourself lmk for the help thing. I understand most of your words but could you be quite specific about sizes and all that. Because if I am going to kill a Maze4 I want to do it right - first time up. Simple reason being I have no margin of error as we have no access to WC products in South Africa and a Maze4 goes for nearly twice its dollar value here.

[boshuter]

I use a 1 1/2" cap for my evaps, it just fits in one of the rings in the maze4 base. There is a larger ring (the one for the sealing O-ring, that would be perfect for dice tube). Your best bet is to take the block apart and take the base with you to the hardware store, just try elbows until you find one that fits the largest ring on the maze base. Then get a section of copper pipe that will fit the elbow for the tube section.

I can't give you an exact measurment, I don't have any more bases laying here. But the o-ring out of one is approximately 1 3/4". Just take the base and try them until you find one that fits that ring.

[harleybro]

Great stuff and congrats on your personal bests. :thumb: Yup boshuter is definatly sticking with the 2900 no way to change his mind. As for the dice tube here's pics of one that I built using the maze4. The little extension piece could be shorter but the extension just needs to be a diameter that fits inside where the O-ring would be. ;) So all it is: 2" copper pipe, 2"cap brazed to it, short piece of 1 3/4"? pipe used for extension. This is the size that needs to be checked. :thumb:

http://i63.photobucket.com/albums/h140/harleybro/IMG_0082.jpg
http://i63.photobucket.com/albums/h140/harleybro/IMG_0080.jpg

http://i63.photobucket.com/albums/h140/harleybro/IMG_0079.jpg

[1Day]

Tried again tonight to bench with chilled water - this time I took the rad off and filled a Colman cooler with ice and some water. Waited about 30 mins to let the temp drop and then primed up the pump and let it run for a while - loop is on a separate PSU to the benching rig.

Sadly it was not a success - base temps were about 10'c higher than the rad dumped in the liquid method. Not sure if that is because of the different mixture used or because humidity is in the 90 percent level where I live. So unless the bench rig was running at max there was way to much condensation. Not going to kill my Lanparty just yet thank you very much.

Not happy at all - I know have 10 bags of ice sitting around waiting to melt:o :o

Win some lose others - there will be a next time when it works I am sure

[harleybro]

Wow really strange about being 10c higher. :scratch: Clear laquer spray paint is a great idea. It works for mobo and graphice cards. Just cover anything that needs good contact (mem chips, gpu, pci-e slots) and spray away. When dried it will appear as nothing is on there. Other then that armaflex 1/2" id for insultaing your tubing and foam tape for the gpu block. That should keep all condensation away. A cpu guide on insulating here (http://www.bleedinedge.com/forum/showthread.php?t=30908&highlight=condensation)

[1Day]

I never quite got round to getting that clear lacquer spray harleybro but I did manage to use up some of that ice I had bought. :o Tip to anyone who is about to use chilled water for the first time - you do NOT need :hide: 15 bags of ice for an evenings benching. Three or four will last the whole evening and still have floating icebergs the next morning. :bonk:

I had packed the bags of ice on-top of each other and just left them to melt when I went to work this morning, and was very pleased and surprised to see that there was quite a large volume of ice left over on my return home. So that was an omen (a sign as they say) to try some chilled water benching with the HD again. :thumbs:

http://img525.imageshack.us/img525/5844/dsc04903rd5.jpg

Went a bit better this time as I had modded the HD to make it a bit more receptive to the whole over voltage thing. Nothing drastic just putting back the two metal shrouds and the fan which is located over the row of voltage thing'ies. Those tiny puppies get really hot according to Rivituner.

http://img215.imageshack.us/img215/3936/dsc04905gj7.jpg

Humidity is a huge problem where I live - and condensation is not a good thing so as you can see lots of tissue paper and a dish drying mat used to prevent any spillage onto my toys.

http://img146.imageshack.us/img146/4879/dsc04908iz6.jpg

One other problem with this Water Cooling set up I am using is the NB cooler. It is just in the way and I had to modify one of the screws to get it to fit. You can see how the top fastener extends way over the NB cooler.

I was basically testing settings and trying to figure out how to get higher clocks with my HD2900XT using AtiTool 2.7b and was not really trying to do any serious benching so I was pleasantly surprised when I pulled in a really good run on 3Dmark03. It is not the most complimentary benchmark for the Radon graphics card but a benchmark that I really enjoy doing. And one I use as a stability test for my different setups when overclocking.

http://img146.imageshack.us/img146/1345/3dmark0347948clockssz6.jpg

While it will at first seen like I am posting the same image twice I am doing this for a reason, the top image has the clocks and the bottom one should have the voltage used. :coocoo: Unless I have got them mixed up. Nope seems I have not. I was able to push the clock-speed up the next notch and run PCmark05 and Nature in 2001SE as a test of the clock speed of the card but for some reason 3Dmark06 would not start at all. I took that to mean the condensation problem was getting a bit serious and stopped for the night.

http://img215.imageshack.us/img215/1936/3dmark0347948uf8.jpg

[Just Learnin']

Careful 1Day, if you have any pets or other fuzzies in the air, Harley was getting some goofy temps once running from an ice filled cooler like that. When he opened his waterblock up he found what appeared to be matted fur stuck inside it!!! :yikes: :fire:

Link (http://www.bleedinedge.com/forum/showthread.php?t=30707&highlight=fur)

[FB]

ICE+WATER+SALT = colder temps and faster than regular ice+water

[1Day]

Thanks FB gave my left over ice a quick half pint of salt and a brisk stir and started benching before dashing to work :irk: - barely made the office on time. :thumb:

PB in both 3Dmark06, 05 and PC Mark05 before work so maybe I should get up an hour earlier each day and bench as it seems to work for me. No hassles setting up and was able to increase core speed to 982/999 of the HD2900XT :bow: :bow: I seriously love this card.

3Dmark05
25553

http://img216.imageshack.us/img216/9398/3dmark0525553oq6.png

and a 3Dmark06 of 14248 while a PB is way off the pace when compared to most of the hwbot results but at present I will take any gain that is on offer.

Did some 2001SE too but just as a hardware test 72K and some change for that benchmark. The HD2900XT seems stronger for me in PCmark05 and there I was able to set a PB.

PCmark05

9749

http://img216.imageshack.us/img216/1055/pcmark059749xy4.png

Included a 3Dmark03 run with the last of the chilled water and was able to clock a respectable score of 49464 - respectable only for HD2900XT mind you :chit: this benchmark is pwnd by nVidia cards.:bow:

http://img215.imageshack.us/img215/6712/3dmark0349464wb4.png

Thats all for now - and all for a while I think. My next project is to try and get both a GPU and CPU pot made locally.

[1Day]

Careful 1Day, if you have any pets or other fuzzies in the air, Harley was getting some goofy temps once running from an ice filled cooler like that. When he opened his waterblock up he found what appeared to be matted fur stuck inside it!!! :yikes: :fire:

Link (http://www.bleedinedge.com/forum/showthread.php?t=30707&highlight=fur)

The only fuzzies I have are my two kids :rotflmao: and based on your recommendation I have shaved both girls heads bald. Good point there mate :thumbs: do not want anything to get in the way of the benching now do we.:bonk: :bonk:

Seriously now - I check both the pump and the maze4 after each benching session. What I have killed is a PSU, some cheap thing which I bought just for the fans and the water pump, but that budget import could not even keep up with a few fans and a water pump pulling 18volts.:scratch: :scratch:

[rra]

Having fun with that x2900 ,nice benching 1Day good clocks and very nice 03 score well done .What was your nature score in 3d01 by the way? :thumbs:

[harleybro]

Definately nice clocks on that 2900. Looks like you are learning the card well. I also would love to see an 01 link or even just the fps screen. Curious how the card makes out in nature too. :mischeif:

[boshuter]

ICE+WATER+SALT = colder temps and faster than regular ice+water

While this is true; I'd think twice about trying it without the rad, not sure how good an idea pumping salt water through water loop is. If you have the rad submerged in salt water, you only have to worry about corrosion on the outside of the rad.

BTW: did you know that it doesn't have to be salt, any substance added to a water ice mix will have the same reaction, IE; sugar, alcohol, etc. The coldest you can get in a salt water mix is -21.1c, other substances will result in different temps. For every mole of foreign substance dissolved in a kilogram of ice water the temp will go down 1.8c.

OT: 1day..... PLEASE resize your pics :beat: Aside from the fact that I'm on dialup, I hate having to scroll sideways to see the whole SS..... a larger pic won't result in a larger score ;)

[ol'norton]

Looking Good 1Day. :thumbs:

[1Day]

As per request small sized details page of a 2001 run. Untweaked in anyway at all. Ram settings were on auto if I recall correctly, selected the game tests and pressed run. Do not think there was even a generic LOD setting used.

http://img215.imageshack.us/img215/2096/2001naturelr0.png

[harleybro]

Thanks 1Day. :thumb: I will be playing with all water in the near future and I am curious how the GTX will make out against the 2900. :yup:

[rra]

Yip those are impressive figures 759 in Nature is strong with no tweaks and or LOD so you well on your way to great set up for even stronger bench once ram tweaked and tweaks implemented,nice well done.:beer:

[1Day]

After the weekends fun some 3Dmark03 with the HD2900XT was the order on the day - it does takes some getting used to the different fps markers that I use to monitor the benchmark run. I felt that it would be interesting to do the runs using the same systems set up that I had used for my nVida 3Dmark03 benching efforts with some surprising results.

3Dmark03 - 49673

http://img404.imageshack.us/img404/8596/3dmark0349673up4.th.jpg (http://img404.imageshack.us/my.php?image=3dmark0349673up4.jpg)

Did some testing and a quick 05 bench-run as well. Of possible interest is that this run was second run - without any reboot between - and a LOD setting of 20 (mistake) so maybe that is worth looking at. Of other interest to me is that my results were better with 5-4-4-11 ram timings as apposed to 4-4-4-9 or 4-4-4-11, which were used when 3Dmark03 benching with the nVidia card. In other words the set up for 05 and 03 appear to be different; or the set up of the system is different between graphic cards with one card working better with tighter timings and the other with more relaxed timings.

3Dmark05 - 25896

http://img479.imageshack.us/img479/4850/3dmark0525896iu7.th.jpg (http://img479.imageshack.us/my.php?image=3dmark0525896iu7.jpg)

[rra]

A little late but bravo those are super scores but was not expecting less from you well done,sometimes in 3d benching loose timmings eliminate bottlenecks :thumb: :bow:

[Just Learnin']

My 8800 loved 4-4-4-8 and 4-4-4-12 with the P5W DH, hated anything else. 1Day has better scores posted to the bot already than that and probably a bunch in reserve! :yup:

[rra]

Just had a peep at Bot scores nice .Like to see some Aqua from x2900 1Day as you only have your default run there ;) room for pts lol even if its not Ati favorite.I have always stated on the Systemshock forums where we met that 1Day will be a strong force to be reckoned with ,and has proved me right.:bow:

[Just Learnin']

She's not shy!

[1Day]

I am a shy retiring rose - and you better not forget it buster. LOL

As to AM3 - currently can not get AM3 to work with the HD2900XT - application runs but does not post result. I had this issue with the initial HD set up I did when the card cam out but do not recall how I fixed it. But with a limited time budget an issue at present I would rather focus on stuff that works.

[rra]

I am a shy retiring rose - and you better not forget it buster. LOL

As to AM3 - currently can not get AM3 to work with the HD2900XT - application runs but does not post result. I had this issue with the initial HD set up I did when the card cam out but do not recall how I fixed it. But with a limited time budget an issue at present I would rather focus on stuff that works.
Ha remember I did same thing you just need to do the rename the direcdll file and she should work.;)