Overclock.net › Forums › Graphics Cards › NVIDIA › brettjv's 470 SLI vs 670 Bench-Off (both OC'd) ... (Now with 7970 OC'd!)
New Posts  All Forums:Forum Nav:

brettjv's 470 SLI vs 670 Bench-Off (both OC'd) ... (Now with 7970 OC'd!) - Page 7

Poll Results: What are you thinking: Used Fermi for SLI, or go with Kepler?

 
  • 18% (20)
    Thinking of getting adding another cheap, used Fermi for SLI!!!
  • 64% (70)
    Thinking of selling my Fermi and going with Single High-End Kepler (670/680)!!!
  • 17% (19)
    Thinking of waiting to see what lower-end/cheaper Keplers bring to the table!!!
109 Total Votes  
post #61 of 209
Quote:
Originally Posted by brettjv View Post

And just for Windfire I went and bought a Kill-A-Watt on my way home, and took a baseline for the 470's, along with temp tests (CPU, System, Exhaust) while running the Crysis bench so we'll be able to see the changes with the new card.
thumb.gif (+ virtual rep)
This should give some credibility to Nvidia's claim that Kepler is much more energy efficient than Fermi.
347
So, hopefully this is not just an accidental thing but continues to Maxwell.

Quote:
Originally Posted by brettjv View Post

Although having only one card looks BIZARRE to me at this point. Had SLI for so long ...
Hmmm....is this a disguise to say 'I want another GTX670' ?
This is definitely a prelude to something. biggrin.gif

Also, perhaps when you are finished doing all the tests for the GTX670, you might add back a GTX470 as a Physx card to test its usefulness. It is most likely not needed and not worth it. But this test should be a good footnote to your Gigabyte GTX670 OC review.
post #62 of 209
Thread Starter 
Kay, I guess I shouldn't leave y'all totally hanging ... here's a stock 3dMark11 (stretched) compared with the previous 470 SLI (800/900) run. A little disappointing, but not crushing by any stretch. Oddly, their System Tool now misreports my CPU clock headscratch.gif

The GPU Score was 9560, nothing to sneeze at.

EDIT: Okay so I'm up to 1296 core ATM, running windowed Heaven ... but my voltage hasn't budged from 1.162V. I've tried maxing the volt slider .... which does nothing. Do only some of these cards go to 1.175, or do I need to use Precision, or ... how does this work?

Oh, and Windy, you're gonna LOVE the news from the Kill-A-Watt when I post it in the OP biggrin.gif

477
Edited by brettjv - 5/14/12 at 11:52pm
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
post #63 of 209
Thread Starter 
Okay, Heaven bench passed perfect at 1358 core (boost = +163), but artifacts started at around 1359. Going to back her down to 1350 and see what I can do with a Memory OC on top of it.

Voltage is still at 1.162V and hasn't stepped to 1.175 even with the slider at +100 ... what's the deal w/that?

Here's a 3dMark11, passed at (1350/3005), compared to stock (1189/3005)

Stock GPU Score: 9560
OC'd GPU Score: 10594

Lets have a quick look at the (oft maligned) Kepler Scaling here, shall we?

Our clock speed improvement = 161/1189 = 13.54%, memory is left at stock

Lets peep our FPS improvements:

(50.73 - 45.51) / 45.51 = 11.47% = 84.7% OC scaling
(50.32 - 45.06) / 45.06 = 11.67% = 86.19% OC scaling
(64.34 - 58.38) / 58.38 = 10.2% = 75.33% OC scaling
(31.54 - 28.62) / 28.62 = 10.2% = 75.33% OC scaling

I dunno where you guys stand on this, but an average of around 80% scaling with NO memory overclock is actually pretty friggin solid OC scaling in my book. Turns out when it's calculated correctly, Kepler scaling is pretty comparable to that of Fermi thumb.gif

515
Edited by brettjv - 5/15/12 at 7:57am
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
post #64 of 209
If you average all the FPS numbers you got it comes out to a 10.9% increase.
If you then take your clock speed improvement of 13.54%, we can calculate the percent performance increase per percent core clock improvement, or 10.9% fps / 13.54% core change = 0.81% fps increase per 1% core change. That's a pretty hefty improvement over what i got when i didn't yet have the boosted max core value of a stock 670 so i had to just use the unboosted core values. I got .69% fps increase per 1% core change, so that shows that the boost definitely doesn't scale linearly when you overclock. So that might mean the 670 becomes increasingly more powerful the more you overclock because of how boost scales. We would need to know your stock unboosted core frequency and the OC unboosted core frequency to attempt to find this out though.
post #65 of 209
Thread Starter 
Card finally gave up with driver crash at 1350/3305 ... I'll pick up from here tomorrow ... I got bored and was just bumping 25MHz at a time on memory, I suspect final OC will be around 1350/3275 or thereabouts ...

See y'all in the morning for some benching fun!
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
post #66 of 209
Thread Starter 
Quote:
Originally Posted by SeanPoe View Post

If you average all the FPS numbers you got it comes out to a 10.9% increase.
If you then take your clock speed improvement of 13.54%, we can calculate the percent performance increase per percent core clock improvement, or 10.9% fps / 13.54% core change = 0.81% fps increase per 1% core change. That's a pretty hefty improvement over what i got when i didn't yet have the boosted max core value of a stock 670 so i had to just use the unboosted core values. I got .69% fps increase per 1% core change, so that shows that the boost definitely doesn't scale linearly when you overclock. So that might mean the 670 becomes increasingly more powerful the more you overclock because of how boost scales. We would need to know your stock unboosted core frequency and the OC unboosted core frequency to attempt to find this out though.

You're coming to an incorrect conclusion here because you're missing a part of 'the math'

It's actually irrelevant, those other numbers, and I'm certain that the FPS scales linearly with clocks, regardless of whether it's 'boost clocks' or whatever source the extra clock speed comes from.

The part you appear to miss is ... like I told you on the other thread, using unboosted values will inherently lower the apparent scaling ... this is because by using those, you abnormally increase the value of the denominator in the equation, which is the clock % change.

For example, (1100-900)/900 = .222, which is greater than (1300-1100)/1100 = .1818. When you divide the SAME top number, % FPS change, by a larger number, as would be the case when you use unboosted values ... you'll end up with a lower scaling %.

This is enough of a discrepancy to explain your 69% scaling vs. my 81% scaling ...esp if we weren't running the exact same test which could add more variability to the equation thumb.gif
Edited by brettjv - 5/15/12 at 1:56am
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
post #67 of 209
Quote:
Originally Posted by brettjv View Post

You're coming to an incorrect conclusion here because you're missing a part of 'the math'
It's actually irrelevant, those other numbers, and I'm certain that the FPS scales linearly with clocks, regardless of whether it's 'boost clocks' or whatever source the extra clock speed comes from.
The part you appear to miss is ... like I told you on the other thread, using unboosted values will inherently lower the apparent scaling ... this is because by using those, you abnormally increase the value of the denominator in the equation, which is the clock % change.
For example, (1100-900)/900 = .222, which is greater than (1300-1100)/1100 = .1818. When you divide the SAME top number, % FPS change, by a larger number, as would be the case when you use unboosted values ... you'll end up with a lower scaling %.
This is enough of a discrepancy to explain your 69% scaling vs. my 81% scaling ...esp if we weren't running the exact same test which could add more variability to the equation thumb.gif

I probably should have mentioned this when i linked that old post, but i only linked it to compare the WRONG values from before we had the boosted values to work with. There was no other way to calculate it before because the information just wasn't available. So i just wanted to see the difference between doing it properly with the boost values and doing it improperly with just the core values.

But yeah, i hastily formed that hypothesis that the boost scales with total overclock. But still, knowing your stock unboosted core frequency and the OC unboosted core frequency won't hurt anything since those numbers are easy to find tongue.gif
post #68 of 209
@brettjv: these are nice preliminary results, you've hit > 1.3 right away, that's great. thumb.gif Congrats.
post #69 of 209
Thread Starter 
Sean: Ignoring the Kepler Boost (I'm trying to coin a phrase here wink.gif) the number in GPU-Z (I assume that's what you mean) are:

Default - 980c, 1059 boost
Overclocked (+161 offset) - 1141c, 1220 boost

Observed Op Frequency with KB
(0 offset): 1189MHz
(+161 offset): 1350MHz

Kepler Boost amount appears to be 130MHz on this card.

Everyone Else: Just trying to firm up final OC levels then I'll do some runs for y'all.

Does anyone have an answer about my 1.162 question? Did I just get a lower voltage capped card or what?

EDIT:

Adding a screenshot of the gains from a memory OC (+120MHz actual, 240MHz per AB, 480MHz Effective, ) on top of the 1350 Core. The numbers are not impressive. 3dMark11 Perf appears to have very little memory bandwidth limitation going on. I'll try the same test again later w/Heaven.

1350/3005 - GPU Score = 10594
1350/3240 - GPU Score = 10726

490
Edited by brettjv - 5/15/12 at 9:28am
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
post #70 of 209
Thread Starter 
Okay, Official Max OC, Power Consumption and Temp Numbers are now up, along with the first bench result: 3dMark11

Vantage has also been added and guess what ... it's Our First 670 WIN!!!

Moar benches to follow, whatever I can get done before work ...

SOOO ... turns out Precision CAN push the card to 1.175V, unlike Afterburner. Interesting. I'm kinda frightened what the OC will go to with more voltage. I'm going to leave it at 1350 for now though since that's already at the top end of what people expect with these cards.

I'm guessing I picked a bad day to do these benchies ... I wonder if maybe people are a little more concerned with Diablo 3, perhaps tongue.gif
Edited by brettjv - 5/15/12 at 10:21am
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: NVIDIA
Overclock.net › Forums › Graphics Cards › NVIDIA › brettjv's 470 SLI vs 670 Bench-Off (both OC'd) ... (Now with 7970 OC'd!)