G70 Core Clock Variance

Thanks for the info Unwinder.

So if you want to overclock the whole card to the same amount at present you need to go to 459, 486, 513, 540 etc or else your shader and Rop domains will be at one level down from these.

16 27 432
17 27 459
18 27 486
19 27 513
20 27 540
21 27 567
22 27 594
23 27 621
24 27 648
25 27 675

I suppose Gt3 on 3dmark05 or nature Gt4 on 3dmark03 should show spiky jumps in fps as you hit these boundaries then .. ie non linear increase.

Can anyone test this ?
 
dizietsma said:
Thanks for the info Unwinder.

So if you want to overclock the whole card to the same amount at present you need to go to 459, 486, 513, 540 etc or else your shader and Rop domains will be at one level down from these.

16 27 432
17 27 459
18 27 486
19 27 513
20 27 540
21 27 567
22 27 594
23 27 621
24 27 648
25 27 675

I suppose Gt3 on 3dmark05 or nature Gt4 on 3dmark03 should show spiky jumps in fps as you hit these boundaries then .. ie non linear increase.

Can anyone test this ?
Click to expand...

I hope that is is just the driver's limitation and there will be more flexible ROP frequency control in new drivers. The PLL they use for geometric domian is able to generate clock via: 2 feedback dividers, 2 reference dividers and 1 post divider.
The PLL they use to clock ROP domain is simplier, but it also contains 1 reference divider allowing more accurate glock frequency generation. In fact the clock generated by this PLL is defined with the following formula:

((oscillator_clock (27MHz) * feedback_divider (16)) / reference_divider (1))>>post_divider

I can understand the reason of not using post divider (it is often limited with maximum VCO frequency), but I have no ideas why do they always set the reference divider to 1 for this PLL. This can be done for increasing generated clock stability / working around some hardware issues as well as it can be a simple flaw in their clock frequency generation algorithm. Let's hope that it is the second case.
 
digitalwanderer said:
Lemme guess, that was either a BB or DP reply....right? ;)
Click to expand...

This is the honest truth, but funnily enough I've never spoken to them before :D
 
2 Unwinder

1) currently NVIDIA driver is able to adjust clocks of these domains with 27MHz (oscillator frequency) step only
2) Perf. level 1 : 415(+35)MHz

(432-27=405) So let's hope :).
 
SomeBody Else said:
2 Unwinder

1) currently NVIDIA driver is able to adjust clocks of these domains with 27MHz (oscillator frequency) step only
2) Perf. level 1 : 415(+35)MHz

(432-27=405) So let's hope :).
Click to expand...

What's that 405 about? You mean the GT?
 
SomeBody Else said:
No, I mean 415 MHz is not equal 432-27 :).
Click to expand...

Small info update:

According to report of testers using 15.7, the driver sets ROP clock to 418MHz in low power 3D mode (27 * 31 / 2). So hardware does seem to be able to generate more accurate clocks with no problems, at least in low power 3d mode,
It is still unclean why such rough step is used for 3d mode. As a quick idea: probably BIOS defines personal reference dividers for each performance level.
 
I knew it! I knew it! :)
We believe in you, master Unwinder! ;)

P.S. Happy vacation! А какой у тебя отпуск длинный... :)
 
digitalwanderer said:
Thanks Pharma, I just frontpaged this one.

Did y'all see nVidia's response to Hiiiiilbert's inquiry?

Update 1:

Answer from NVIDIA:

Hey Hilbert,

As our chips become more advanced, we are implementing more complex clocking inside the chip. 430MHz is the primary clock speed of the chip and can be verified by fill rate tests.

We will work with Rivatuner to read the correct registers in order to report the right clock.

Hope this makes sense.

Now if you disect that answer (as vague as it honestly is) we can make note of the fact that NVIDIA uses "Primary" clock speed. Obviously, there's a secondary clockspeed running also. Very likely there are surely different clocks for different pipes.

If 430 remains the primary clock then why can Asus (check that here) advertise it as 470 ?

That would be somewhat misleading to you, the consumer as you think it's an uber Ultra version or something like that. They sincerely use this to advertise:

Engine Clock 470 MHz**
**NV clock(430MHz)+Geometric Delta clock(40MHz)

As it seems nobody can explain the 40 MHz differential properly, yet is is being used as marketing gimmick, consumers automatically assume it's 470 MHz where that's not 100% true.
Click to expand...
Click to expand...

I actually think that the Asus model will clock to 510Mhz when entering a 3D application but someone will have to verify that. The BFC OC goes to 500Mhz when 3D are being executed.
 
digitalwanderer said:
bigz said:
We spoke to David Kirk earlier today:

http://www.bit-tech.net/news/2005/07/07/g70_clock_speed/
Click to expand...
And he basically didn't tell you anything about it. :?
Click to expand...

What do you mean? He said the variable clocks were derived from mobile tech, are the reason for the lower power consumption and heat on the G70, that there are more than just 3 and that they will be providing info to Unwinder on which clocks are most suitable for end-user overclocking. What were you expecting - schematics?
 
trinibwoy said:
What were you expecting - schematics?
Click to expand...
aye, they're not exactly going to turn around and give us exact schematics on how it works... it will be interesting to see how they work together with Unwinder to allow users to get the most out of RivaTuner with G7x based parts.
 
bigz said:
it will be interesting to see how they work together with Unwinder to allow users to get the most out of RivaTuner with G7x based parts.
Click to expand...
That's kind of what I meant. No offense intended towards your interview BigZ, more towards the interviewee. ;)

It just sounded more like PR happy talk than an explanation of the what and whys to me.
 
I'm not sure what you mean, Digi. He's explained that it's a power-savings technique. What more do you want about the why's? As for the what, how is what was said not enough?
 
digitalwanderer said:
It just sounded more like PR happy talk than an explanation of the what and whys to me.
Click to expand...

I still think he covered the what (multiple clocks) and why (lower power / heat) just not the how which honestly isn't any of our business.
 
I did some testing with 3DMark 2k1 Game 4 - Nature, as it is the most precise gfx testing tool for me.

Single Leadtek card, 'cos my SLI setup don't go higher than 485 on gpus. Forceware 80.40, all stock settings, memory clock always 1200 MHz.

MHz - FPS - FPS delta

430 - 239.4 - 0.0
440 - 240.6 - 1.2
450 - 243.2 - 2.6
460 - 245.5 - 2.3
470 - 246.2 - 0.7
480 - 248.6 - 2.4
490 - 249.5 - 0.9
500 - 251.6 - 2.1
510 - 252.7 - 0.9
520 - 253.3 - 0.6

530 is too much for the card.

So, as you can see, the performance difference is really not linear (although this test is extremely precise!) and FPS delta jumps from 0.6 to 2.6 for each 10 MHz. BUT! Predicted clocks for bumps should be 430+27=457, 457+27=484 and 484+27=511. Instead of this we can see highest bumps (over 2 FPS/10MHz) between 440 and 450, 450 and 460, 470 and 480, 490 and 500 which don't fit in theory :(
 
Being a tree hugging Californian, anything that has an impact on heat generation and/or power consumption by limiting these to where they are only needed is of course very encouraging news.

My only concern is what is the sustainable (as in over a number of hours) performance output in 'average' conditions if there is indeed some form of throttling going in the chip.

If benchmarks or reported performance is indeed the 'sustainable' performance of the chip, it's really of no interest to the end user provided it's consistent and transparent. I'd be more interested in tests validating this is indeed the case so as benchmarked performance is the same in iteration on hour 1 as they are in hour 3 or 4 (continuous, to model a 2-3 hour gaming period). If so, this is truly the best of both worlds and a very well received feature/technology!
 
digitalwanderer said:
bigz said:
it will be interesting to see how they work together with Unwinder to allow users to get the most out of RivaTuner with G7x based parts.
Click to expand...
That's kind of what I meant. No offense intended towards your interview BigZ, more towards the interviewee. ;)

It just sounded more like PR happy talk than an explanation of the what and whys to me.
Click to expand...

I would have liked more info too but I suspect that Mr.Kirk will have to be very careful about what he lets the competition know . I'm also not too sure that they'd really want us finding out what the true max performance is at this time .
 
Sharkfood said:
My only concern is what is the sustainable (as in over a number of hours) performance output in 'average' conditions if there is indeed some form of throttling going in the chip.
Click to expand...
I would think that current benchmarks already rule out this as a problem.
 
You must log in or register to reply here.
Back
Top