Jump to content
HWBOT Community Forums

GTX580 Matrix Extreme OC: first session at Assembly


Massman

Recommended Posts

Heyo all,

 

Thanks to the boys at ASUS Benelux, Rogier and Milan, I was able to pick up a Matrix GTX580 from the ASUS office the day before I left for Assembly. Upon SF3D's request, I brought the card with me to test under liquid nitrogen. Oh, before I continue I would also like to thank the boys from Jimm's for making the event possible as well as Marcus from ASUS Nordic for providing plenty of hardwares allowing us to do our thing!

 

After a first session, I must say that I've got mixed feelings about this card. For sure, this card has great potential and some of the features of the card are very, very handy and should be adopted by other vendors, but the lack of software support can make it quite difficult for people who are not connected to get the card running at high speeds. After seeing Stummerwinter's 1625MHz result, I found new hope that it will be able to match the Lightning (although there are much more higher clocking Lightnings).

 

SF3D tried to convince me over MSN that the Matrix card is one of the easiest cards to bench under LN2. Of course, I'm so stubborn that I didn't want to believe him. But, after my first session I was quite amazed that he was indeed correct! You just need the right tools and software to get it to run so easily. So, here's what you do:

 

1. Solder on the back of the PCB to disable OCP and enable higher PWM frequency

 

The picture is pretty much self-explanatory. You can do the soldering without removing the backplate of the card!

 

2011-08-051801401.jpg

 

2. Flash the card to the 580 coldslow BIOS

 

Next, you have to flash the card with the coldslow bios for the GTX580 Matrix, which can be found in the coldslow/coldbug bios thread here at hwbot

 

Link: Coldslow and Coldbug VGA BIOS collection - HWBOT forum

 

3. Prepare the card for extreme overclocking

 

This is quite important. Before you flash the firmware of the card, you must prepare the card for extreme overclocking and make sure you have LN2 available. The firmware updates so it will give you 1.5V Vgpu and for sure you don't want that to be pushed through your air cooled card.

 

If you update the firmware on your air-cooled card, you are just stupid and absolutely don't deserve to get any replacement card. So, if it happens to you, I hope you get a refusal and head-smack when applying for RMA ... if you catch my drift.

 

As for insulation, this card is again quite well designed. You can use the backplate as pressure tool to push the insulation to the back of the PCB. At Assembly, I just used 3 double layers of paper towel in between PCB and backplate. It worked perfectly. Just remember to remove the paper towel when you're done, as the condensation water will enter the paper after you're done.

 

4. Update the firmware of the card to 1.5V/1.3V/1.8V Vgpu/Vpex_vdd/Vmem

 

This fourth step is what I perhaps hate the most about this card. To have a really easy bench session, you need to update the firmware to one that has elevated/unlocked voltage, but the firmware seems to be distributed only to those-who-know where to get it. It took me a long time to find the right file (and someone who was okay for this to be updated), but in the spirit of Hipro5-open-sharing, here it is.

 

As I mentioned in the previous section (but doing it again because it's so important): "If you update the firmware on your air-cooled card, you are just stupid and absolutely don't deserve to get any replacement card. So, if it happens to you, I hope you get a refusal and head-smack when applying for RMA."

 

https://hwbotdownloads.s3.amazonaws.com/downloads/tools/matrix_xoc.rar

 

- run soft.bat to update your gputweakit software to allow higher voltage

- run stock.bat to return the firmware to original setting

- run fwup1.bat to update firmware to higher voltage settings

 

Once you run the fwup1.bat, on the next reboot the voltages will be fixed to:

 

- vgpu = 1.5V

- pex_vdd = 1.3V

- vmem = 1.8V

 

There are other firmware files floating around, but those come with higher Vmem (1.9V or 2.0V). Unfortunately, I only had the 2.0V for the event and it gave me quite a lot of problems to push the memory beyond 1250 MHz. I was told the 1.8V version would allow for better clocks ... I'll try later this week.

 

With this firmware update, though, overclocking the Matrix is pretty much reduced to temperature and frequency control. Easy benching for sure !

 

5. Use Nvidia Inspector to finetune voltage and frequency.

 

The last step is start to overclock. ASUS have their own GPUTweakIT software to push this vga, but I would recommend to use Nvidia Inspector. First of all, because it's a lot less Taiwan-ish (colors, colors, colors!), secondly because I'm used to Inspector and I'm a conservative overclocker so I like to use what I know. Thirdly, because I don't understand 5200MHz memory (you will understand when opening the software) and fourthly because it's a lot easier (imho) to finetune the Vgpu in Inspector.

 

Basically, I update the firmware and then set the voltage in Inspector to 0.963V. This way, I still had finetuning possibilities to 1.15V if necessary, which would yield me 1.6V+ on the core. For me it worked quite well, but I can believe others would prefer other software.

 

6. Break records (or die trying) like a pro

 

My result: Massman`s 3DMark11 - Performance score: 11049 marks with a GeForce GTX 580

 

Stummerwinter's result: stummerwinter`s 3DMark11 - Performance score: 12488 marks with a GeForce GTX 580

 

Clearly I have some work to do. I will order some new pro-skills online; hopefuly I have enough funds so I can update my pro-level to something like Stummer's level. I think it will be difficult :D

 

6373081.jpg

 

2011-08-05172755.jpg

 

More coming soon!

 

//edit: in conclusion:

 

Loving the ...

 

- soldering pads which are accessible without having to remove anything

- pwm heatsink is not connected to any other heatsink -> usable for ln2

- firmware = super-relaxed overlocking

- backplate is perfect to use for insulation pressure

 

Hating the ...

 

- firmware not public? are they scared people would show good results ???

- gputweakit memory shows ddr speeds x4. I don't know what 4800MHz memory is. I do know what 1200 MHz memory is ...

- so far, it doesn't seem to match lightning speeds (which is hitting 1600+ even on retail cards)

Edited by Massman
Link to comment
Share on other sites

  • Replies 87
  • Created
  • Last Reply

Top Posters In This Topic

You can remove the mods to send it back later on :)

 

You used Rogiers new soft tool ? On air it does nothing special... maybe it needed the new firmware...

 

I think the issue is the girlie tissue paper, it's holding you back... Tri sli soon ?

 

That tool is the GPUTweakIt and I'm not using it. I prefer the simplicity of Inspector over the colorblast of Tweakit.

 

3-Way ... maybe. Need moar LN2.

Link to comment
Share on other sites

  • Crew

:D

 

Great to read...:)

 

Just to add:

 

I tested right now 5 cards under LN2 in preparation for GC in Cologne...

 

1 x 1675/1400 3D03

1 x 1575/1300 3D11

1 x 1625/1350 3D11

1 x issues below -160°C

1 x not shure, tested different thermal grease

 

These cards I got from Christian (via Michael) from ASUS Germany, thanx btw, just retail, nothing selected...

 

Serial numbers: B5CVCM068782 ~ B5CVCM068788 (not all are present for me) and 68843

 

Found out, thermal grease in combination with container ist a issue, my big container is simly to wide, causes condensation at PWM!

 

big one:

 

GPU-Pot-03.jpg

 

slim version:

 

SLI-Pot-3.jpg

 

The slim version works much better, but has lack of "cold power", but anyway, 1625/1350 where doable...;)

 

Agree with MM, fw-voltage tweak ist not a really nice thing, but works somehow. I didn't use tweakit, simply pushed buttons and EVGA-tool to oc...;)

 

I think, we should work on you insulation...;)

 

BTW, JP: I failed also while ocing, the card which did the 1625/1350 in 3D11 was not modded OCP/PWM...:D

Edited by stummerwinter
Link to comment
Share on other sites

BTW, JP: I failed also while ocing, the card which did the 1625/1350 in 3D11 was not modded OCP/PWM...:D

 

SOF! Ehr, I mean, FAIL! :D

 

Did you test the different FW versions? Any change between 1.8V Vmem and higher?

 

That 1400 card seems quite solid ... :D

Link to comment
Share on other sites

Yesterday i tested my Matrix .

 

I flashed the coldslow bios and tested the tree diffrent firmware . Irs verry strange i can not go below -70 degree . If i do this the frames in 01 are verry slow . So i did exact what mass describes in his first post . Any sugestions ?

 

Thanks ,

 

Roger

Link to comment
Share on other sites

Massman

This happened with me, and with coorect BIOS.

Cars is too freeze....

You have to hot the card, about -100 C and run benchmark for a lit bit, then return to 135 ~145 and set again.

Mine matrix can t work well less that -155

;)

It can cause lots of issues...includes Coldslow (with correct bios too)

 

Sorry the horrible english

Link to comment
Share on other sites

  • Crew
that pot looks so nice, can i have one? :D

 

Sry, not possible, because to expensive to produce...

 

SOF! Ehr, I mean, FAIL! :D

 

Did you test the different FW versions? Any change between 1.8V Vmem and higher?

 

That 1400 card seems quite solid ... :D

 

Some cards like or need 2,0 V (I don't have the 1,9 V hack), to run at lower temps...

 

Well, to be honest the firmware thing is a bit silly. Whoever got idea about it, should have thought twice.

Card would definitely need a switch with normal firmware and oc firmware. *Hint*

 

Then it would be perfect.

 

God hint!

 

[X] *sign*

 

:D

 

I woulnd't use the metal backplate, causes to much condensation...

 

That's the way I do it...

 

plastic backplate covered with armaflex and an adhesive rubber foam (~ 3 mm thick) on the front, a litte rubber at PWM:

 

matrix-iso-1.jpg

 

slim container:

 

matrix-iso-2.jpg

 

and some more armaflex:

 

matrix-iso-3.jpg

 

most important tweak...;)

 

matrix-iso-4.jpg

Edited by stummerwinter
Link to comment
Share on other sites

Tested 4 Matrix cards during this weekend. All these new cards are doing 1550+ core and 1350MHz memory, so these will be suitable for 4-way attempts.

I am not sure what went that badly wrong week a go, when elmor and I tried these cards in assembly summer 2011 event.

 

Great cards for sure. Now I just need to get correct platform for single GPU runs.

Testbed was x58 based and scores are not good. Best score was 11699.

 

I did some ln2 pictures:

typpi%20p.png

typpi%202%20p.png

typpi%205%20p.png

 

Pictures in fullsize, if you like to see them closer:

http://sf3d.pp.fi/images/SF3D%20OC%20pots/Ln2%20pics/typpi.png

http://sf3d.pp.fi/images/SF3D%20OC%20pots/Ln2%20pics/typpi%202.png

http://sf3d.pp.fi/images/SF3D%20OC%20pots/Ln2%20pics/typpi%205.png Look at the small flying bubbles

 

 

I will try SLI action during next week and next weekend I will propably try 4-way, if everything goes like planned.

Edited by SF3D
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.


×
×
  • Create New...