Calculating the key OC points on the 7800

[HeavyH20]

Here are two GTX BIOS revs to try for anyone interested.

KO-Rev3 - based on a newer BIOS build, this BIOS uses a 50 Mhz delta as well as a better blended 3D throttle (432 with 40 delta). The default clock is 490/1200.

KO-Rev4 - same as above with 500/1350 as default clock. This is the one I use.

[Vapor]

Awesome....I'll be sure to try them out.

[krampak]

Is there any bios upate for my Gainward 7800GT?

[-R@mi-]

Nice working my brother.

[HeavyH20]

Is there any bios upate for my Gainward 7800GT?

What clocks can your card handle?

Here is a generic NTSC based GT BIOS with a default clock of 400 and a delta clock of 45 (GT's run at different bumps -20 pipes). It also runs the fan at 100% in 3D.

[krampak]

What clocks can your card handle?

It does 481-482, seems to be the Max of my core clock to don't get artifacts... 1160 on the mem. That's with a Zalman VF700CU.

[HeavyH20]

Nice card. Try this BIOS for the Gainward. 446 root with a 40 delta, 1100 memory, 100% fan in 3D, 75% in 2D. PAL.

[Gorod]

HeavyH20 Excellent work there with key OC points on the 7800 - helped me A LOT ! Whats the next key clock after 621 ? +13 (634MHz) right ?

[rodman]

subscribed Waiting for memory results

[JohnnySK]

hi ppl.. I haven't read the whole post, but if anyone owns a 7800GT like me here's some info (all being MHOs)

MHz in RT <--> 3DM05 score <--> Core/Shader/ROP/MEM in RT monitor
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

400 - 6928 - 441, 405, 405. 501.19
401 - 6987 - 441, 405, 405, 501.19
402 - 6981 - 441,405,405,501.19
403 - 6977 - 441,405,405
404 - 6983 - 445.5, 405,405
405 - 6994 - 445.5,405,405
406 - 7015 - 445.5,405,405
407 - 7027 - 445.5
408 - 7025 - 450,405,405
409 - 7022 - 450,405,405
410 - 7045 - 450,405,405
411 - 7009 - 450,405,105
412 - 7093 - 450,418.5,418.5
413 - 7100 - 453.6,418.5,418.5
414 - 7127 - 455.63, 418.5,418.5
415 - 7130 - 455.63,418.5,418.5
416 - 7127 - 455.63,418.5,418.5
417 - 7148 - 459,418.5,418.5
418 - 7137 - 459,418.5,418.5
419 - 7125 - 459,418.5,418.5
420 - 7125 - 459,418.5,418.5
421 - 7142 - 459,418.5,418.5
422 - 7165 - 464.06,418.5,418.5
423 - 7154 - 464.06,418.5,418.5
424 - 7163 - 465.75,418.5,418.5
425 - 7178 - 465.75,418.5,418.5
426 - 7263 - 468,432,432 <--- Big jump in shader/ROP domain = better score!
427 - 7243 - 468,432,432
428 - 7224 - 468,432,432

The numbers pretty much sum it all up. Tested on my LeadTek 7800GT, threshold set at 40. You have to find the right general core frequency at which the shader/ROP change singnificantly.

I suggest you fire up ATI Tool, start the 3D window, and work yourself up from 400 --> xxx MHz and check when the shader/ROP domain frequency is increased the most, that's the frequency you will get the best performance. You don't need 3DMark and waste 4,5mins for each MHz like I did

And setting a higher threshold IMHO means you will get higher shader/ROP frequency at lower core geometric speeds which again means higher scores.. look at the difference between the frequency I set in RT and the freq the RT monitor is reading out --> it's around 40, the threshold that you can set in the BIOS (I suggest using the latest NiBiTor to do that...)

Does this all make sense to someone ? Btw, is there a thread somewhere around here dedicated to testing out the different memory timings (timing sets) on the 7800GT's?

john

[maverik-sg1]

Man, this information is awesome - really appreciate your information guys, expecially HEAVYH2O.

It certainly does help clear up more than one myth about the 7800's.

Thanks again.

Mav

[HeavyH20]

Well, it is always nice to hear that the research helps others!

Now, thanks to JohnnySK's post, we may have some insight into the way the ROP and Shader clocks act on a GT. Looks like instead of 27 MHz jumps, there are 13.5 MHz jumps. Exactly half. I will have to look into that a little more.

I ran some 3Dmark05 runs with different memory speeds. I have not adjusted the timings, yet, but here are my results with the root core at 500 Mhz and the memory scaled from 1200 to 1440. I ran the tests at 50 Mhz increments except for past 1400.

The test system:
AMD 64 3200+ @ 2.5 Ghx
BFG 7800 GTX OC@ 500 Mhz
81.82 driver
All default drivers settings, no tweaks


Memory Clock________3DMark05_____

1200_________________8441_______
1250_________________8568_______
1300_________________8706_______
1350_________________8792_______
1400_________________8893_______
1420_________________8908_______
1440_________________8945_______

So, the memory can make a big difference, after all. 500 point spread from 1200 to 1440.

[eclypse]

I have a prob.. Seems like the cards are going into 2d speeds half the time and are not changing into the 3d speeds once i run 3dmark..

I've verified this with rivatuner.. I run 3dmark and a min later not even i exit out and check that graph and it shows it never changed from 275Mhz on the cores.

I keep trying and once in a while it will but i dont know why its not everytime. Really buggin me.

Before i was using a modded bios i was using with only the 3d perf enabled and the clocks at 490/1400 at 1.4v and with the volt mods its runing at 1.62 and 2.32v on the ram 24/7.. So i thought changing it back to the normal specs with the 2d and throttle in there would be smarter so the card would run cooler in 2d.. but now i'm getting this stupid prob.

Someone said that the temp bug or throttling bug isint a prob after the 7.77 driver version so i thought i'd give it a shot since i'm using the 78.3 drivers.

Any ideas cause i think it would be annoying be have to keep checking to make sure its going into 3d pref mode when i bench or game.

[HeavyH20]

Make a BIOS with no throttle, but with a 2D. That way you have two levels instead of 3 to deal with. That should take care of the throttle. I had similar problem with my old UEE once it was volt modded.

[eclypse]

Ok will try that thanks.

[JohnnySK]

Just wanted to add that changing memory parameters also help... setting every setting one level lower resulted in 200 3DMarks extra in 3DM05. Although I didn't have time to experiment, it would be interesting to find out whether slower timings BUT higher MEM frequency or the opposite would benefit more to performance.

For example, in comparison to the main PC mem, I have a Hynix that is rated at CAS 2.5 running @ DDR400. However, lowering the CAS to 3.0 BUT setting the MEM to run at near DDR600 I get higher bandwidth than the max frequency I can cram out with CAS 2.5. Might be the same for our little pets, er, 7800's I mean

Oh yes, if anyone wants to test timings, use NiBiTor, it can temporarily set ANY timing you choose without having to flash the BIOS -- VERY neat feature that ATI cards had a long time ago...

L8ers....

[rodman]

Oh yes, if anyone wants to test timings, use NiBiTor, it can temporarily set ANY timing you choose without having to flash the BIOS -- VERY neat feature that ATI cards had a long time ago...

So what your saying is, try relaxing the timings and go for say 1500 memory and see if it is faster due to bandwidth, interesting

[eclypse]

Well i figured out my prob.. The prob was with using Ati Tool 2.5. Overclocking with that must of messed with the cards.. cause as long as i dont change the clocks with that everything is perfect!

[HeavyH20]

That is good to hear. Looks like it is still beta at best.

BTW, I did some testing on the memory timings. I tried to to find all the settings that would tighten the timings but not affect the OC on the memory.

Testing results:


________Stock___Mod
1350____8792____8920
1440____8945____9058

Here is the spec and changes on the memory:

Samsung GDDR3 1.6 Specification

K4J55323QF-GC16 / K4J55323QG-BC16



________SPEC__Change______________________________ ___________

tRC______27_____25_____Row cycle time(min) - operation (tCK)
tRFC_____34_____32_____Row cycle time(min) - Auto Refresh (tCK)
tRASmin__19_____17_____Row active minimum time (tCK)
tRP_______8______7_____Row precharge time(min) (tCK)
tRCDRD___ 9______8_____Ras to cas delay(min) for Read (tCK)
tRCDWR___5______5_____Ras to cas delay(min) for Write (tCK)
tRRD______7______6_____Row to row delay(min) (tCK)
tGR_______2____________Gapless 2 tCK
tRTW_____12___________Read to Write at same bank - CL=8tCK, tCDLR=4tCK
tRASmax__100000________Row active maximum time (tCK)
tRCDRD____9____________Ras to cas delay(min) for Read (tCK)
tRCDWR___5_____________Ras to cas delay(min) for Write (tCK)
tWR______5_____________Last data in Row precharge (7 tCK)
tCDLR_____4____________Last data in to Read delay (6 tCK)
tCDLW____0____________ Last data in to Write delay (0 tCK)
tCCD______2____________Column address to column address delay (3 tCK)
tCKmin____1.6___________Clock minimum cycle time (ns) - CL=8
tCKmax____6____________Clock maximun cycle time (ns) - CL=8

Register timings on the 7800 GTX (default is timing set 7)

0813211B

0A01060A

20270509

The new settings register values:

07112019

0A01060A

20260508

[eclypse]

Thanks for the info there heavyH2O.. I tried it out with my current mem speed of 1480 and no probs. Tested it first and then flashed both cards with the new mem timings.

Beat my personal best score in 3Dmark 03 by 230+ points! So i'm at 34830+ Now to just try raising the ram more to 1500 and maybe 35,000 can happen! hehe.

[HeavyH20]

For those interested, here is the BIOS I am using. It is a generic BIOS, so it will work with all the reference 7800 GTX cards. XFX, eVGA, BFG, etc.

3 performance levels
500 core plus 50 MHz delta
1350 memory speed
tighter memory timings (does not affect OC capacity)
Throttle at whole 27 MHz multiples (432 root plus 40 MHz delta)

That's about it. This BIOS gets me around 200+ or so on 3Dmark05 versus the stock BIOS.

[sluflyer06]

just to add for the thread i flashed the KO rev. 4 bios and was getting artifacts at 500 and above on a BFG 7800GTX it artifacted at both 1.4V and 1.5V, I think it was the 50mhz delta, I lowered it to 45 and problem was gone.

[HeavyH20]

just to add for the thread i flashed the KO rev. 4 bios and was getting artifacts at 500 and above on a BFG 7800GTX it artifacted at both 1.4V and 1.5V, I think it was the 50mhz delta, I lowered it to 45 and problem was gone.

Yes, it will happen. The rev 3 and 4 BIOS are actually for an upcoming release of the eVGA GTX card. To use Rev 4, you really need to be watercooling the card. Most cards will not handle the clocks for long with the stock HSF. Also, the 1.5V mod does not work on the BIOS, since it is not physically available, unlike the 6800 cards. You need to do a physical VMod for any gains.

[HeavyH20]

Two files posted at this link:

http://www.xtremesystems.org/forums/...15#post1082715

There is a 500 MHz version as well as a 473 MHz version. The latter is better for air and the 500 Mhz really should be used for watercooled cards. If you want to use the 500 Mhz with stock cooling, you may want to underclock the card for 24x7 operation and run it as is for benchmarks.

[JohnnySK]

Btw.. this is for a 7800GT,watercooled, delta 40 stock BIOS, a little bit tigher timings... (clocks decreased by 1-4 CLK)

http://service.futuremark.com/compare?2k3=4329151
http://service.futuremark.com/compare?3dm05=1326354

I don't want to push it further although a few more MHz might up the shared/ROP again.. it's watercooled but I've had enough benchmarking, it's time to play now!