Calculating the key OC points on the 7800

[HeavyH20]

Some companys is giving out new GT bios with no delta clocks.
Dont really know why.

The BFG GT uses no delta, as well. eVGA brought that new BIOS out to deal with artifacts occuring on the GTs. Bit of a marketing trick, really. GTX at 460 or GT at 425 when it is really GTX at 500 or GT at 425. The quad yields are good, so the lower clocking GPU's get the laser and made into GT cards. So, many GT people are claiming clocks like 475 to compare against GTX OC results of 475, and the two are very different. Depends on the BIOS delta. Apples and oranges.

[don_vercetti]

somehow, i am not suprised by that. Mine clocks pretty crappy too. Anyway found yet to reconnect the bad quads?
Does the BIOS actually help real speeds at all? lets say without the clock delta, you could clock the whole thing up to 490, wheras with it, the max clock was 470 with the delta making the highest clock 510. Which is faster?

[Mr. Tinker]

The BFG GT uses no delta, as well. eVGA brought that new BIOS out to deal with artifacts occuring on the GTs. Bit of a marketing trick, really. GTX at 460 or GT at 425 when it is really GTX at 500 or GT at 425. The quad yields are good, so the lower clocking GPU's get the laser and made into GT cards. So, many GT people are claiming clocks like 475 to compare against GTX OC results of 475, and the two are very different. Depends on the BIOS delta. Apples and oranges.

My BFG 7800GT has artifacts at stock. You mention something like that in your post. What do you know about that? Do I just have a bad card?

[HeavyH20]

My BFG 7800GT has artifacts at stock. You mention something like that in your post. What do you know about that? Do I just have a bad card?

The stock BIOS has a fixed fan rate of 43% at 3D. This overheats the card. Also, it has one level, instead of 2. Try this rev2 BFG GT BIOS and let me know how it works for you. Has all the fixes.

[dwwalker14335]

somehow, i am not suprised by that. Mine clocks pretty crappy too. Anyway found yet to reconnect the bad quads?
Does the BIOS actually help real speeds at all? lets say without the clock delta, you could clock the whole thing up to 490, wheras with it, the max clock was 470 with the delta making the highest clock 510. Which is faster?

Since it is a physical cut, there is no way to get it back. As for the clocks, if you want the best performance, the Vertex clock should be faster than the ROP and Shader clocks, so I would hazard to say that the delta based BIOS would be faster.

[don_vercetti]

its not possible to reconnect the pipes, with something conductive?

[einCe]

i noticed while monitoring my speeds before, during, and after running 3dmark05, it seems that my core starts out 481 as i specified in coolbits, but then drops back down to 425 for the rest of the test, is this normal?

[HeavyH20]

No, that is not normal. Which BIOS are you using?

[HeavyH20]

Here are a couple of eVGA BIOS revs with no delta for the regular GTX and GTX KO. The 528 for the normal GTX and the 529 for the KO.

[eclypse]

I cant seem to figure this out..Were else can i read about this topic.

[Jimbo Mahoney]

Does this principle apply to all graphics cards?

[don_vercetti]

no just 7800 series.

[Jimbo Mahoney]

Although surely all gfx cards use a reference clock?

I understand the issue here is that a 27 Mhz refclk is quite a high frequency, leading to larger step sizes.

Sorry for semi-hijack, but does anyone know the refclk on an X850XT? The reason I ask is that looking at ATITool, it says my reference divider is currently 4 and the current minimum step size is 6.75 Mhz. I find it more than a coincidence that 4 x 6.75 = 27 Mhz....

I'm going to do some tedious 3DMark '05 testing anway....

EDIT - Actually, having studied the 1st post some more, it's something to do with the syncrohnisation of multiple cores, right? Rather than some form of multiplier limitation dealing with the 27 Mhz refclk?

[cadaveca]

Yes all cards use a reference clock. The divider of 4 refer to the GDDR interface...basically quad channel, but also a divider from teh reference clock. Each channel runs @ 6.75mhz, in order to stay inline with the crystal on the board...the little silver metal square SMD.

The x800xl, as i posted earlier, has a reference clock of 25mhz...and hence why it does not clock as well as teh x850's, beside the power area differences.

Anyway, i'm pretty sure that this is the base frequency that the card runs @ on the pci bus too.

[don_vercetti]

i flashed my inno3d to a xfx 7800gt, that doesn't have a clock delta. I was able to get my core clock from 470 where it artifacts (delta to 510) up to 495, which is artifact free.
In 3dmark03, my score was 200 points higher, so overall, i'd say not using the delta helped me. Of course, once i vmod things could all change.

[Mr. Tinker]

The stock BIOS has a fixed fan rate of 43% at 3D. This overheats the card. Also, it has one level, instead of 2. Try this rev2 BFG GT BIOS and let me know how it works for you. Has all the fixes.

I did notice the fixed fan rate, but my temps never climbed too high. Like 55C max during gaming. Anyway, my card is out on RMA, so we'll see what happens with the next one. Thanks much for the help.

Sorry for going OT.

[HeavyH20]

I have seen some people using a custom BIOS with no delta so the clocks can be locked instead of dealing with the delta. While this sounds promising, you have to wonder why nVidia chose to use a delta. So, I ran some benchmarks to discover which is really better. In addition to the regular 40 delta, I also wondered how effective the ASUS TOP card was with a 50 Mhz delta, so I created a BIOS with that delta as well.

Here are the results of the delta scaling on a 7800 GTX.

Test System:
AMD 64 3200+ @ 2.5 GHz
MSI K8N Diamond
DDR 3200 2.5-3-3-7, 1T

Settings:
all driver settings default quality, 81.82

3DMark05 results

Clock____0 Delta_______40 Delta_______50 Delta
490_______8347_________8570__________8593
495_______8397_________8588__________8603
500_______8451_________8728__________8772
505_______8475_________8764__________8781
510_______8534_________8769__________8823
513_______8569_________8803__________8847
520_______8577_________8805__________8864
525_______8703_________8852__________8921
527_______8710_________8912__________8980
530_______8728_________8935__________9045
535_______8765_________8983__________9078
540_______8824_________9034__________9123


Summary

The zero delta BIOS inhibits performance, clock for clock. Additionally, the OC limits are not increased with no delta in place.

The best delta value appears not be 40, or none, but 50.

[eclypse]

HeavyH20, thanks for all that testing.. But i think you need to re check your findings more.

Your saying that with a delta is better then without.. but if you check the first line 490+50 delta= 540 clock= 8593 is better then the score you got at a 540 clock with no delta =8824?

another clear example from your findings..

500+50 delta=550 clock= 8772 but 540 no delta beats it out at 8824.

What was your memory clocked at in these tests?

[Vapor]

I think you have it a little mixed up eclypse...the delta gets ADDED the values in the table. A delta of 50 is most definitely the best.....has anybody made a BIOS with an even higher delta?

[eclypse]

Well un mix me up then cause it seems clear to me..

From what i've read around here is that delta adds Mhz to the clock.. so 500 clock with a delta of 40 = a 540 clock.. Am i wrong?

[Vapor]

You're right, but the clock he is declaring in the table is before you add it...so the 500MHz that he lists is 500 with no delta, 540 with a 40MHz delta and 550 with a 50MHz delta. This what you set when you OC it (before the delta is added)....what 3DMark reports is the value after you add the delta.

[HeavyH20]

What you say is true, kind of. If you take the 500 Mhz clock with the delta of 40, which is 540 and compare it to the no delta version of 540, you do get better performance. That goes without saying. That is because the ROP and Shader clock are at 540 and so is the vertex. With the delta BIOS, the ROP/Shader is at 513.

The thing to look at is the ceiling on the no delta BIOS is the same as the 40 MHz delta BIOS. If I am limited to the same max clock with either delta BIOS, then the 40 delta is the better choice. That is the real point of the tests. If the zero delta actually let me go to 580, which is the same as the 40 delta BIOS at 540, then great. But, it does not.

The bottom line is absolute performance, and the delta based BIOS beats the no delta BIOS hands down. The highest score attainable at the limit of the card is what matters. So, no delta, 8824 ; 40 Mhz delta, 9034; or 50 MHz delta, 9045. I have coached some people who had used a zero delta BIOS with the misperception that it would improve performance. It does not. Once they switch to a 50 Mhz BIOS, they improved their scores by 200 to 300 points. That is how my card was able to get 9400 or so on 3Dmark05 with plain old watercooling and an AMD 64 3200+ at 2.7 GHz. And, yes, that is with no LOD cheat.

I tried higher than 50 MHz delta's, but performance dropped since the vertex core was still the same limit. All I ended up doing was dropping the ROP and Shader clocks one step lower.

nVidia created the 40 MHz delta for a reason. It is exactly 27 MHz plus 13 MHz so that no matter what, the vertex core is one whole oscillation (27 MHz) above the ROP/Shader core no matter the clocking. So, 499 gives you 539 on the vertex and 486 on the ROP/Shader. 500 gets you 540 ( a whole multiple of 27) and the ROP/Shader at 513 (513 is closer to 500 than 486).

[eclypse]

Ok so what your saying is that if the max clock my card can do stable is 621Mhz with no delta.. it would actually run stable at 621+ 40 or 50 delta? So that would make the card really 671Mhz and the other 2 cores would stay at the speed it would at 621Mhz. So thats how its really better to run with a delta?

Also is there a sweet spot speed to run the memory at? or is it just basic higher is better and faster?

[HeavyH20]

Ok, we will use your example. 621 is the limit with a no delta BIOS. With a 40 Mhz delta BIOS, the card will run at 661 and 621 on the ROP/Shader. Now here is the reason why the no delta is a tad slower. Let's say the top clock is 608. That is 648 vertex, and the ROP/Shader clocks are at 621 still. You will get better performance with the delta BIOS at 608 than the no delta BIOS at 621.

As for the memory, I am starting to play around with the timings a tad. Similar to running your memory at 2.5-3-3-7 at 270 FSB versus 2-2-2-5 at 200 FSB. It is a balance between latency and bandwidth. All the tests I ran were with default memory timings and 1350 MHz.

Next testing period, memory scaling with different timings.

[eclypse]

Ok i got it now and cant wait to get some time to find that new sweet spot with a delta bios!

I cant wait to see your findings with the memory tests