DId you update the bios?
Try saving your settings and even create a fan profile to use at windows startup.
Printable View
I am watercoooling mine and it runs on max slider with ATi catylyst overtool
http://cpcff.co.uk/phpBB2/viewtopic....16267&start=60
if you want to edit your Bios have a look at this
http://www.xtremesystems.org/forums/...d.php?t=167966
Editing the bios doesn't work, it was attempted weeks and weeks back in the "unleash the gladiator" 3870 thread.
The 3870 is NOT the same as a 2900, if anything it is closer to a x1950.
hardware volt mods are out, and those work. But you better have good soldering skills and intermediate knowledge on electricity and ohms law if you are going to attempt them.
I'm still testing, but this far it runs nice on 878/1300, or whatever the core clock was... I'll be back with a small chart showing you how it scales in 25MHz increments on both core and mem.
EDIT: Tests below are run with a E6320 @ 3500MHz, (500x7, mems 1:1 @ 4-4-4-8) and the frequencies listed are as I set them in AMD GPU Clock tool, not absolute frequencies. The XP install is quite ancient, so the scores suck (but then again, it reflects the scaling anyways).
Stock freq (whatever they are): 11057p
850/1200: 11667p
875/1225: 11905p
875/1250: 11936p
875/1275: 12003p
875/1300: 12088p
880/1300: 12162p (the 880MHz freq set in AMD GPU Clock Tool results in the 878MHz core frequency instead of the 860-something with 875MHz).
Beyond this my card wasn't willing to operate, but it's just a rough test to get a feel for the scaling. The RAM frequencies don't matter that much as you see a 50MHz increase on the rams resulting in "only" a 152 point increase. It's not bad though. :)
I found out from a review that these samsung chips run at default 1250mhz at 0.8ms latency.
I've been experimenting with 27mhz incriments and I have found the card to be extremely stable at 877/1304.
Remember, 3dmark06 does not test full stability since you are not using any of your input controls (keyboard, mouse). it is more of a test for stability of viewing graphics rather than controlling them.
I am confident that with proper ram and pcb cooling, and increasing the latency should allow a higher clock on the memory.
But with that thermal goop being more of an insulator, don't expect the memory to gain that much performance when oc'ing.
The core is where the performance is at. Hardware volt that thing, and the card will be flying!:up:
we want the power color bios.... somebody please post it....
if i were you, i'd watch techpowerup or ask the member there to upload the bios. i just flashed my x800gto to an x850xt using the bios from this site, they have a huge collection.
at first glance all i could find was the powercolor 512mb 3850 bios, but they might be there.
http://www.techpowerup.com/vgabios/i...CI-E&memSize=0
yeah but people likely to buy that specific card will hit this thread at some point when researching their card....
Do not run the fan at 100%, odds are you will kill it. I have a HIS 3870 and when I ran the fan at 100% using Rivatuner, it died after one 3DM06 run..lol. :D
So I replaced the HSF with a TR-HR03.
My results are here:-
http://www.ocforums.com/showthread.php?t=538373&page=2
Best result:
This card rocks the socks of anything I've had so far, especially the way the core is clocking.
13162 3DM06
http://img225.imageshack.us/img225/9...33dmui1.th.jpg
AFAIK, the GeCube, Visontek and HIS cards are all based on the same reference PCB. One look at the Inductors, Polymer Fujitsu, Nichion and Chemicon caps and their placement tells the quick story. If you need my BIOS, let me know.
PCI-E Frequency 110MHz...
Its been around in my mind for a while, but still, someone get this clear to me!
PCI-E freq. default is 100MHz right? in some case, lowering or give a little more helps to avoid mobo bugs, like p5b mobos, when pci-e speed drops to 1x instead of 16x.. bit still, does it give any real performance boost if I set it to 110 rather than 100...?
i run mine at 120, would run it higher if i wernt using sata drives.
i would always say the faster the better but you would need to test using
3dmark or something similar to see if your scores improved.
if your score doesnt improve or you see no improvement in any bench leave it alone.
be careful raising your PCI-E if you have SATA drives. i would imagine you could corrupt your OS if you set it too fast. i tried mine at 120 once and couldn't even boot into Windows. i put it back to 100 and everything appeared okay. i decided to leave it at 100 since then.
I received my Sapphire 3870 today. Did not flash the BIOS and set clocks in CCC. Settled on 860/1300 for a quick run of 3dmark06 and the result is 9825. This puts me second in the Result Analyser comparison thing. To keep temps down I set the fan to run at 70% in Rivatuner, which kept everything under 50 degrees.
Raising the PCI-E frequency helps stabilize high OC clocks.
It really does not provide a performance increase.
Setting the PCI-E too high might get you more clicks, but may cause instability..and if you use a 965 chipset, you can corrupt your SATA drives and loose all your data.
what maker of 3870 would clock to 900?
I wouldn't recommend raising PCI-E frequency. The speed difference does not justify the risk IMO.
This is a great thread along with the "Unleash the gladiator" thread I think I'm armed to make the 3870s rock. I just received two Duorbs and some AC Mx-2 and will try this out tonight using the PLL .71 BIOS I've downloaded.
Very exciting, hopefully I can get mid 900's GPU on air.
Thanks for the info Slugzkea, Stilt, Random, et al.
10e
By the way I voted "yes" :)
So far here is what I've done:
* Installed the dual duorbs with AC MX-2 paste. They are great, and don't impede the CF bridges at all
* Pencil modded each 3870 to just around 1.4v
* Flashed each with Winflash to the new PLL 0.71 BIOS (from "gladiator" thread)
* Used RivaTuner 2.06 to go up to 905mhz on each card and then crossfired
* Turned off RT and used CCC to go all the way up to 885 fully stable
I'm getting between 35 - 38 in idle and 50 - 52 on load with this config at 905, pencil modded, fully stable, in crossfire
Memory could only go to 1276 but that's ok. I didn't touch the mem heatsinks, but I might later on. This is not a big deal to me.
So far I'm impressed with how much more playable this makes Crysis at 1920x1080, DX9, high settings, no motion blur, 8x AF, no AA. Benches only added 3 to 5 FPS, but the game is now playable in single player in all tested situations. Prior to this it would get choppy at times and make it painful to aim.
Tonight I go for 1Ghz on air using some of the OC bioses based on the .71 using CCC in CF.
Thanks,
10e