[HOWTO] ATI Radeon overclocking

[mirko_3]

As long as you know the risks, and don't overvolt too much...
I'm assuming you want to overvolt the gpu...
http://www.ocforums.com/showthread.php? ... 0+volt+mod

I only gave it a cursory look, if you don't find what you need there, you may search the forums (I believe those forums are a gold mine for any overclocker...) or PM Hell-Fire -XS- (on overockers' forums, not these of course) as he seems to know everything about voltmodding ATIs...

Also, improved cooling would be better...

[TheRealElch]

Great work!
Sadly no go on my Ati Radeon 9700 pro too
I'v tested up to version 0.4b but it gives me always "No ATI card found."
Here is my lspci -v output

Code: Select all

0000:01:05.0 VGA compatible controller: ATI Technologies Inc Radeon R300 ND [Rad eon 9700 Pro] (prog-if 00 [VGA])
        Subsystem: ATI Technologies Inc: Unknown device 0002
        Flags: bus master, stepping, 66Mhz, medium devsel, latency 255, IRQ 17
        Memory at e0000000 (32-bit, prefetchable)
        I/O ports at b000 [size=256]
        Memory at f1000000 (32-bit, non-prefetchable) [size=64K]
        Capabilities: [58] AGP version 2.0
        Capabilities: [50] Power Management version 2

0000:01:05.1 Display controller: ATI Technologies Inc Radeon R300 [Radeon 9700 P ro] (Secondary)
        Subsystem: ATI Technologies Inc: Unknown device 0003
        Flags: stepping, 66Mhz, medium devsel
        Memory at e8000000 (32-bit, prefetchable) [disabled]
        Memory at f1010000 (32-bit, non-prefetchable) [disabled] [size=64K]
        Capabilities: [50] Power Management version 2

[Hasw]

@TheRealElch: I know why, because it's the 6th device (currently it only checks the usual first device on the AGP bus). What board/card do you have?

EDIT: I updated to version 0.5a, which uses the information from /proc/bus/pci/devices. This should work better.

Nah, AGP voltage does 2 things:
1-Increase chipset voltage
2-Increase AGP voltage (of course)
Both only help you reach very high FSB speeds (like in my case, where I have 180FSB on a 133 board).

I think 1 depends on the mainboard layout/chipset. On my nForce2 board the AGP voltage is controlled differently from chipset core voltage. AGP voltage increases there only the voltage of the AGP I/O buffers.

[TheRealElch]

Using version 0.5a works, but it only detects the card. When calculating the clocks it throws out a "Gleitkomma-Ausnahme"
Here is my output:

Code: Select all

Radeon overclock 0.5a by Hasw (hasw@hasw.net)

Found ATI card on 01:28, device id: 0x4e44
I/O base address: 0xfffffffc
Video BIOS shadow found @ 0xc0000
Reference clock from BIOS: 27.0 MHz
Gleitkomma-Ausnahme

I also tried the -x option with 1432 and 2970 as parameters - no go
My Hardware:
Chaintech 7KDD AMD MPX760 Chipset
His Ati Radeon 9700PRO 128MB

Code: Select all

0000:00:00.0 Host bridge: Advanced Micro Devices [AMD] AMD-760 MP [IGD4-2P] System Controller (rev 11)
        Flags: bus master, 66Mhz, medium devsel, latency 32
        Memory at d0000000 (32-bit, prefetchable)
        Memory at f4000000 (32-bit, prefetchable) [size=4K]
        I/O ports at c000 [disabled] [size=4]
        Capabilities: [a0] AGP version 2.0

0000:00:01.0 PCI bridge: Advanced Micro Devices [AMD] AMD-760 MP [IGD4-2P] AGP Bridge (prog-if 00 [Normal decode])
        Flags: bus master, 66Mhz, medium devsel, latency 32
        Bus: primary=00, secondary=01, subordinate=01, sec-latency=32
        I/O behind bridge: 0000b000-0000bfff
        Memory behind bridge: f0000000-f1ffffff
        Prefetchable memory behind bridge: e0000000-efffffff
        Expansion ROM at 0000b000 [disabled] [size=4K]

[oggialli]

Thank you ! My Mobility FireGL T2 in my laptop went nicely from 307/200 to a whopping 480/230 ! Which gave an insane performance boost in UT2004.

[Hasw]

Using version 0.5a works, but it only detects the card. When calculating the clocks it throws out a "Gleitkomma-Ausnahme"
Here is my output:

Code: Select all

Radeon overclock 0.5a by Hasw (hasw@hasw.net)

Found ATI card on 01:28, device id: 0x4e44
I/O base address: 0xfffffffc

It doesn't detect the right PCI device, 01:28 should be 01:05. So the I/O base address isn't correct and the calculation failed. Can you please send me a copy of /proc/bus/pci/devices to hasw@hasw.net, so I can check why it fails?

[Hasw]

Update: There's now an ebuild available: http://www.hasw.net/linux/rovclock-0.5b.ebuild

[bgradid]

Thank you! that worked perfectly with my radeon 9100

[Hasw]

I've now added some memory timings in version 0.6a. Use with care, better login via ssh to make changes because it's possible that it corrupts your screen.

http://www.hasw.net/linux/rovclock-0.6a.ebuild
http://www.hasw.net/linux/rovclock-0.6a.tar.bz2

[ericxx2005]

I've now added some memory timings in version 0.6a. Use with care, better login via ssh to make changes because it's possible that it corrupts your screen.

I can't seem to find any timings that don't cause corruption, have you had any luck?

However, this seems interesting, I've never heard of changing memory timings on video cards. I am very interested in helping you develop this, but have very little experience programming. Let me know if there's anything that I can do to help.

[Hasw]

I can change tWR and tCAS one unit down, other changes doesn't work. There are many other timings which maybe working better to change.
I'll add them if I've found out where they are stored.

[Hasw]

Now added more memory timings: http://www.hasw.net/linux/rovclock-0.6b.ebuild.

Code: Select all

hasw@hasw ~ $ sudo /usr/sbin/rovclock -i
Radeon overclock 0.6b by Hasw (hasw@hasw.net)

Found ATI card on 01:00, device id: 0x4e44
I/O base address: 0x9000
Video BIOS shadow found @ 0xc0000
Reference clock from BIOS: 27.0 MHz
Memory size: 131072 kB
Memory channels: 1, CD,CH only: 1
tRcdRD:   6
tRcdWR:   3
tRP:      7
tRAS:     11
tRRD:     4
tR2W-CL:  3
tWR:      3
tW2R:     2
tW2Rsb:   1
tR2R:     2
tRFC:     20
tWL(0.5): 2
tCAS:     4
tCMD:     0
tSTR:     1
XTAL: 27.0 MHz, RefDiv: 12

Core: 276.75 MHz, Mem: 270.0 MHz

You can set multiple memory timings at once, for example:

Code: Select all

rovclock -t tRcdRD:5 -t tRcdWR:2

[ericxx2005]

Works like a charm here! Between o/c'ing and timings (which now work for me) I went from ~450 fps to ~600 in fgl_glxgears. That's a 33% increase! Now we don't have to worry about ati driver speed improvements, just o/c like hell.

Keep up the good work Hasw. You should post this on rage3d.com, as they would be interested as well.

[WareKala]

I've been looking for some good Linux app to overclock my Radeon card...It's an integrated piece of ****, and AFAIK I can't adjust memory timings at all as it shares the system memory. But I should be able to adjust core speed. Well, the problem is that I get en error like:

Radeon overclock 0.6b by Hasw (hasw@hasw.net)

Found ATI card on 00:00, device id: 0xcab2
I/O space not enabled, aborting.

What's wrong, and what to do about it?

[Hasw]

Can you please send me the output of 'lspci -vv' (PM or hasw@hasw.net)?

[WareKala]

I just took a look at rovclock's sources and noticed that it just checks if the vendor id is ATI's and that's why it fails for me. The first device listed by lspci is an AGP bridge, with ATI as the vendor, then comes PCI bridge also made by ATI. It's the last device listed by lspci which is my Radeon card. So, perhaps rovclock should also check the device class if it matches 0300?

[Hasw]

Yes, already did it The detection routines are only a quick hack until I have more time to write something better.

http://www.hasw.net/linux/rovclock-0.6c.ebuild
http://www.hasw.net/linux/rovclock-0.6c.tar.bz2

[WareKala]

I apparently can now change the core speed successfully. Thank you for good job Only one complaint: rovclock still says it's version 0.6b

[Hasw]

Only one complaint: rovclock still says it's version 0.6b

Ups, should release a new version to fix this?

[hypnotic]

Hasw,

Thanks for the tool. Here's some feedback. I'm running a Mobility 9700 Pro. I don't intend to actually overclock this GPU, so I can't report whether rovclock works in that regard or not.

I think rovclock is reporting an incorrect memory clock on my machine. According to my research, my specific machine should be clocked at 405 core/270 mem. And more generally, all Mobility 9700 memory clocks should be between 200-275 (400-520 effective). rovclock reports the seemingly correct 405 core speed, but it reports my memory clock at 438.75. So, is 438MHz my effective rate (ie, my clock is actually about 220)? Or is it simply incorrect?

Code: Select all

Found ATI card on 01:00, device id: 0x4e50
I/O base address: 0x3000
Video BIOS shadow found @ 0xc0000
Reference clock from BIOS: 27.0 MHz
Memory size: 262144 kB
Memory channels: 1, CD,CH only: 0
tRcdRD:   4
tRcdWR:   2
tRP:      4
tRAS:     9
tRRD:     3
tR2W-CL:  3
tWR:      3
tW2R:     3
tW2Rsb:   1
tR2R:     2
tRFC:     17
tWL(0.5): 2
tCAS:     4
tCMD:     0
tSTR:     1
XTAL: 27.0 MHz, RefDiv: 6

Core: 405.0 MHz, Mem: 438.75 MHz

[splaq]

I was bored wanted more out of my 9800 pro..so this is what i got...

Code: Select all

glxgears
Default: Core: 378.0 MHz, Mem: 337.50 MHz
22073 frames in 5.0 seconds = 4414.600 FPS
22223 frames in 5.0 seconds = 4444.600 FPS
22216 frames in 5.0 seconds = 4443.200 FPS

OC: Core: 425.25 MHz, Mem: 378.0 MHz
24516 frames in 5.0 seconds = 4903.200 FPS
24883 frames in 5.0 seconds = 4976.600 FPS
24896 frames in 5.0 seconds = 4979.200 FPS

trying for 5000+ fps
Core: 438.75 MHz, Mem: 391.50 MHz
25715 frames in 5.0 seconds = 5143.000 FPS
25726 frames in 5.0 seconds = 5145.200 FPS
25717 frames in 5.0 seconds = 5143.400 FPS

fgl_glxgears
Default: Core: 378.0 MHz, Mem: 337.50 MHz
4164 frames in 5.0 seconds = 832.800 FPS
4310 frames in 5.0 seconds = 862.000 FPS
4331 frames in 5.0 seconds = 866.200 FPS

OC: Core: 425.25 MHz, Mem: 378.0 MHz
4734 frames in 5.0 seconds = 946.800 FPS
4780 frames in 5.0 seconds = 956.000 FPS
4763 frames in 5.0 seconds = 952.600 FPS

trying for 1000+
Core: 438.75 MHz, Mem: 391.50 MHz
4900 frames in 5.0 seconds = 980.000 FPS
4902 frames in 5.0 seconds = 980.400 FPS
4908 frames in 5.0 seconds = 981.600 FPS

never made the 1000 mark with fgl_glxgears...but did get close with i think 991...then the system froze go figure heh..btw..this works with the 9800 pro's also incase you missed that portion at the beginning

[AndreAPL]

great work
rocking here with 9800se

thanks

P.S. anything about softmod's in linux ?

[BetterUnborn]

First of all: THX THX THX

I can confirm it works with X600 Mobility, too!

I really missed a tool like that!
I don't need it for overclocking, I just underclock my X600Mobility ... from 390/250 to 80/80!
While on (2D) desktop I can see no difference at all, and heat reduces significantly. In windows manual underclocking gave me about 30min of extra battery time, I have to figure out what benefit I get here, but it should be close to that! Why take a laptop with intel integrated graphics, when you can save that much power with an ati chip and still have the graphics power when you need it??

But one interesting effect occured: in windows, I use atitool, which allows me to underclock downto 50/50 with no mad effects at all; if I set frequencies to 50/50 with rovclock, the display turns to a flickering grey (my desktop in a VERY distorted view), up to 80/80 this is gone step by step. So why can I do more underclocking in windows?? Do I have to fine-tune the memory parameters, too?? btw, the flickering is also there when I only underclock the core OR the memory below 80.

[shimbob]

Out of curiousity, will underclocking let the chips run cooler?

Two of my machines (an SMP Athlon-MP and a P4-3ghz stuffed into a Pundit-R case) put out a ton of heat. Vast majority of the time I'm not gaming or anything requiring fast video. Heck, the pundit is a headless machine, '-c 20' seems to run okay.

Also, I get this quirk on the Pundit-R

Code: Select all

zlab rovclock-0.6c # ./rovclock  -i
Radeon overclock 0.6b by Hasw (hasw@hasw.net)

Found ATI card on 01:05, device id: 0x5834
I/O base address: 0xd000
Video BIOS shadow found @ 0xc0000
Reference clock from BIOS: 14.32 MHz
Memory size: 31744 kB
Memory channels: 3, CD,CH only: 0
tRcdRD:   5
tRcdWR:   8
tRP:      8
tRAS:     18
tRRD:     3
tR2W-CL:  2
tWR:      1
tW2R:     3
tW2Rsb:   0
tR2R:     2
tRFC:     13
tWL(0.5): 0
tCAS:     6
tCMD:     0
tSTR:     0
XTAL: 14.32 MHz, RefDiv: 6

Core: 300.72 MHz, Mem: 0.0 MHz
zlab rovclock-0.6c #

lspci:

Code: Select all

0000:01:05.0 VGA compatible controller: ATI Technologies Inc Radeon 9100 IGP (prog-if 00 [VGA])
        Subsystem: ASUSTeK Computer Inc.: Unknown device 8107
        Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR+ FastB2B-
        Status: Cap+ 66Mhz+ UDF- FastB2B+ ParErr- DEVSEL=medium >TAbort- <TAbort- <MAbort- >SERR- <PERR-
        Latency: 64 (2000ns min), cache line size 10
        Interrupt: pin A routed to IRQ 11
        Region 0: Memory at c0000000 (32-bit, prefetchable) [size=fdb00000]
        Region 1: I/O ports at d000 [size=256]
        Region 2: Memory at fdc00000 (32-bit, non-prefetchable) [size=64K]
        Expansion ROM at 00020000 [disabled]
        Capabilities: [58] AGP version 3.0
                Status: RQ=256 Iso- ArqSz=0 Cal=7 SBA+ ITACoh- GART64- HTrans- 64bit+ FW+ AGP3+ Rate=x4,x8
                Command: RQ=1 ArqSz=0 Cal=0 SBA+ AGP- GART64- 64bit- FW- Rate=<none>
        Capabilities: [50] Power Management version 2
                Flags: PMEClk- DSI- D1+ D2+ AuxCurrent=0mA PME(D0-,D1-,D2-,D3hot-,D3cold-)
                Status: D0 PME-Enable- DSel=0 DScale=0 PME-

[BetterUnborn]

Underclocking definitely reduces heat!!

Power consumption is proportional to the frequency, so switching to half the frequency should bisect power consumption and thus heat production ... theoretically. I don't know exactly how this is with GPUs, but commonly ICs do have parasite currents that are independent of frequency, so the example above is best-case.

With my laptop, I can feel the effect ... before both chip and memory were constantly running at full speed heating up the left side of my keyboard and forcing the fan to start up. Now the fan is down, and my keyboard is ... cool ... only little warmer than the rest.

I can't say how thankful I am for rovclock ... combined with cpuspeedy and laptop-mode my machine is COMPLETELY silent ... in normal usage, no fan and no mechanical part at all is in motion, absolutely 0 Sone!! This makes linux on laptop a dream!!

Now even noise is better under linux than under windows (there I have no chance of finding something like laptop-mode, hence I can constantly hear my hard disk rotate)