Important information about the 7800

cambrian

Gawd
Joined
Aug 5, 2004
Messages
647
I first read about this in nvnews. Here's a discussion of it:

http://discuss.futuremark.com/forum/showflat.pl?Cat=&Board=techdisplayadapters&Number=5476916&page=&view=&sb=&o=&fpart=1&vc=1

And here's the primary source:

http://www.digit-life.com/articles2/video/g70-2.html

Basically, what they're saying is that the ROP/Shader of the card operates at 27 MHz while the geometry operates at an independent frequency. So let's say you overclock your core to 490 MHz, the ROP/shader will actually be adjusted to operate at 486 MHz, a multiple of 27 MHz. You're basically downclocked. So any overclock that is not a multiple of 27 will be pointless.
 
I did a bunch of runs at 5 MHz increments on the core. There are definitely certain spots where the score improves. It kind of follows the 27 MHz increases, but they seem to be more like 425, 450, 475 and 500.

Here were the 05 scores at same mem clock and different core speeds:

430 - 7987
435 - 8030
440 - 8048
445 - 8073
450 - 8263
455 - 8265
460 - 8341
465 - 8343
470 - 8344
475 - 8531
480 - 8536
485 - 8611
490 - 8596
495 - 8610
500 - 8769
505 - 8805
 
HeavyH20 said:
I did a bunch of runs at 5 MHz increments on the core. There are definitely certain spots where the score improves. It kind of follows the 27 MHz increases, but they seem to be more like 425, 450, 475 and 500.

Here were the 05 scores at same mem clock and different core speeds:

430 - 7987
435 - 8030
440 - 8048
445 - 8073
450 - 8263
455 - 8265
460 - 8341
465 - 8343
470 - 8344
475 - 8531
480 - 8536
485 - 8611
490 - 8596
495 - 8610
500 - 8769
505 - 8805

It makes sense if you do the math. All the above in red - ad 40 and divide by 27. Works out perfect.
 
MUCHO said:
It makes sense if you do the math. All the above in red - ad 40 and divide by 27. Works out perfect.

Yes, you have it exactly. The best points on the clock are whole multiples of the core frequency oscillation of 27 MHz. The article had it almost right. That calculates to the following root core clocks:

(17 x 27) - 40 = 419
(18 x 27) - 40 = 446
(19 x 27) - 40 = 473
(20 x 27) - 40 = 500

And, here is the proof. See how a single MHz below the whole number threshold affects 3Dmark05 scores:

445 8052
446 8222

472 8404
473 8508

499 8669
500 8753
 
When I get my KO I will have to bump her to 500, I wonder how Kyle made out with his KO on that issue. I am sure we will be seeing results soon?
 
god damn i was planning to get an EVGA 7800GT that runs at 445 core clock but its 1 freaking mhz off from getting that boost.. and i dont want to overclock and void the warranty

1 MHZ !!
 
I think if u overclock with evga it doesnt void warranty. I mean if you think about it you are already getting a card that is overclocked.
 
TomJesusBrady said:
I think if u overclock with evga it doesnt void warranty. I mean if you think about it you are already getting a card that is overclocked.

and if i think about it a little more i realize that if overclocking didn't void a warranty then i could burn cards on will and get new ones repeatedly
 
jacuzz1 said:
When I get my KO I will have to bump her to 500, I wonder how Kyle made out with his KO on that issue. I am sure we will be seeing results soon?


You will get a gain, but I think there is a little more explaination required on why certain clocks are used by the vendors.

The other two cores. How are those controlled?

Well, when you adjust the root clock, you actually adjust all three clocks.

The root clock gets bumped by 40 MHz to represent the real geometry or vertex core clock. So, lets pick the 490 (530 vertex clock) core of the eVGA KO. 513 is the best vertex clock value (19*27) , which translates to 473 root (513-40). So, we easily exceed this value. So, why does eVGA push the card so far past the optimal 473 root frequency? To boost the ROP and shader cores.

The shader and ROP cores start at 415 (really 418.5) in low power, and jump in whole frequency values (ie, 17 x 27 = 459) to a value that is less than or equal to the root frequency. In this case the root frequency is 490. So, this limits the other two cores to (18* 27 ) 486, which coincidentally, is now very close to the KO spec clock. If they had chose 480 for the root clock, then the ROP and shader clocks woudl have been stuck at 459 instead of 486.

With that, we also have the reason why BFG chose 460 as their core clock. It seems marginal over other vendor's 450, but it lets the ROP and shader cores clock up from 432 to 459, which yields the second gain on the vertex plateau range of 446 root (486 real) through 473 root (513 real).

So, a little confusing, but you do have some control over the other two clocks.

So, here is a list of primary - vertex based (red) and secondary - ROP and shader based (yellow) root clock OC targets. The red will yield the larger gain, and you will see a second, albeit smaller gain on the blue clocks.

419
432
446
459
473
486
500
513
527
540
554
567
581
594
608
621
 
Big Fat Duck said:
god damn i was planning to get an EVGA 7800GT that runs at 445 core clock but its 1 freaking mhz off from getting that boost.. and i dont want to overclock and void the warranty

1 MHZ !!

Well, vendors like round numbers for marketing. I bet that when you check the real clock, it will be 446.
 
Big Fat Duck said:
god damn i was planning to get an EVGA 7800GT that runs at 445 core clock but its 1 freaking mhz off from getting that boost.. and i dont want to overclock and void the warranty

1 MHZ !!



dunno bout you but in 20 years of running puters..I have had one bad card myself..

oc it and begone..
 
So is it still the higher we go the better preformance we get just as samller gains or what
So how big of difference is there between

527
540
 
Gun_Strife said:
So is it still the higher we go the better preformance we get just as samller gains or what
So how big of difference is there between

527
540

The biggest gains occur at the key vertex clocks. The next biggest gain is at the GOP and shader clocks. So, if you were at 526 versus 527, you would see about 100 points difference. At 539 versus 540, around 70 points.
 
so on this stock cooling I can get the card up to 553mhz and 1.37ghz, at that high I see artifacts, so what would you say would be the best core speed, btw i'm using coolbits and drivers 78.01
 
Well, for 24 x 7 operations, I run my card at 500/1300. I have had it at 540/1400, but I got a little artifacting during the benchmark run. I would not push it too hard on a daily basis. If you can do 490 /1300 or 500/1300 all day under game load, then you are a step ahead.
 
Back
Top