My review of MSI Power Edition gtx 670

[tviceman]

Well here is some BS about the next version of Afterburner
http://forums.guru3d.com/showthread.php?p=4360550#post4360550

Awesome. Possible false advertisement lawsuit in the works.

 

[cmdrdredd]

Woah...apparently (if I read it right) they promised Voltage adjustment then expect the programmer to make it work after the card releases (before he even has any idea how to do it).

Maybe EVGA's hardware solution isn't so bad after all? Not to say EVGA is flawless...they are overcharging IMO.

 

[DrBoss]

yea, i don't really understand the whole MSI / programmer relationship. You'd think it would all be done in house, not farmed out.

 

[thilanliyan]

yea, i don't really understand the whole MSI / programmer relationship. You'd think it would all be done in house, not farmed out.

Unwinder is the guy who made Rivatuner, which is Afterburner now right? That would explain the "farming out"...he doesn't live in Taiwan.

 

[DrBoss]

Unwinder is the guy who made Rivatuner, which is Afterburner now right? That would explain the "farming out"...he doesn't live in Taiwan.

I see, thanks for explaining.

You'd think msi would have supplied Unwinder with a GTX 680 Lightning and GTX 670 Power Edition before they were made available to the public... especially considering they were so bold as to advertise the unlocked voltage feature directly on the box.

 

[cmdrdredd]

Unwinder also writes EVGA PrecisionX

 

[Cookie Monster]

Um.. False advertising?? Im surprised they are getting away with this.

 

[tviceman]

Yeah I am too. I'm still going to wait until Uwinder gives a definitive answer, but if this ends up holding true then I am all for some sort of legal action (i.e. refund).

 

[Quantos]

Wow, that actually sounds like false advertising. If it turns out that Unwinder says something along the lines of "sorry, guys, it's just not possible", that'll be a blatant lie on MSI's part, and I for one wouldn't be surprised they get sued.

The thing is that this sounds a lot like a blatant lie, and not just a development mistake. If it was the latter, you'd think somewhere along the line a department or another would have realized the error. The designers had to agree to this, the marketing obviously put the mention of it on the box, etc...

 

[DrBoss]

I ran some benchmarks in Crysis 1 last night to determine how my OC's scale with regard to gaming performance.
All test were performed with my CPU at 4.4 GHz using a my custom Crysis config @ 1920x1200 with 8x AA
The FPS figures below represent the average FPS over 5 loops.

Stock: Core 1215 MHz | Memory 3005 MHz = 56.575 FPS
OC1: Core 1266 MHz | Memory 3305 MHz = 60.830 FPS
OC2: Core 1280 MHz | Memory 3500 MHz = 62.165 FPS
OC3: Core 1300 MHz | Memory 3602 MHz = 64.005 FPS
OC4: Core 1316 MHz | Memory 3703 MHz = 65.115 FPS

My highest stable OC yields 8.54 FPS over the GPU's factory default overclock. (15.10% increase)
The OC i run 24/7 yields 5.59 FPS over the GPU's factory default overclock. (9.88 % increase)


Question: If i underclock the core and memory in Afterburner down to reference 670 settings and run the same Crysis benchmark... will the numbers i get accurately reflect the performance of a reference 670?

 

[cmdrdredd]

I ran some benchmarks in Crysis 1 last night to determine how my OC's scale with regard to gaming performance.
All test were performed with my CPU at 4.4 GHz using a my custom Crysis config @ 1920x1200 with 8x AA
The FPS figures below represent the average FPS over 5 loops.

Stock: Core 1215 MHz | Memory 3005 MHz = 56.575 FPS
OC1: Core 1266 MHz | Memory 3305 MHz = 60.830 FPS
OC2: Core 1280 MHz | Memory 3500 MHz = 62.165 FPS
OC3: Core 1300 MHz | Memory 3602 MHz = 64.005 FPS
OC4: Core 1316 MHz | Memory 3703 MHz = 65.115 FPS

My highest stable OC yields 8.54 FPS over the GPU's factory default overclock. (15.10% increase)
The OC i run 24/7 yields 5.59 FPS over the GPU's factory default overclock. (9.88 % increase)


Question: If i underclock the core and memory in Afterburner down to reference 670 settings and run the same Crysis benchmark... will the numbers i get accurately reflect the performance of a reference 670?

Every card boosts differently even at the same base clock. For example I have two EVGA FTW cards and if I run both of them at +85 I get a base clock of 1091 and boost clock of 1170. Yet my first card will boost to 1241 and my second card will boost to 1267. Both at the same clocks according to PrecisionX and GPU-Z.

 

[tviceman]

Yet more confirmation that MSI used blatantly false advertising and the MSI PE gtx670 cannot utilize software-controlled core, pll, and memory voltage adjustments:

Partners wishing to have a card with a base power target over 195W must use a custom PCB with suitable power circuitry. NVIDIA won’t allow partners to ship higher-power cards using the reference PCB.
Software overvoltage control is forbidden.

Read more at http://www.anandtech.com/show/6096/evga-geforce-gtx-680-classified-review/3#yJRCc8sJe80rxvAx.99

EDIT: I may have spoke too soon. Sounds like the afterburner author has successfully found a work around for Kepler's voltage lockout (at least on the lightning version): http://forums.guru3d.com/showpost.php?p=4368157&postcount=42

 

[Quantos]

Yet more confirmation that MSI used blatantly false advertising and the MSI PE gtx670 cannot utilize software-controlled core, pll, and memory voltage adjustments:



Read more at http://www.anandtech.com/show/6096/evga-geforce-gtx-680-classified-review/3#yJRCc8sJe80rxvAx.99

EDIT: I may have spoke too soon. Sounds like the afterburner author has successfully found a work around for Kepler's voltage lockout (at least on the lightning version): http://forums.guru3d.com/showpost.php?p=4368157&postcount=42

Well I can understand why MSI outsource to Unwinder, he seems to be pretty good at what he does. He's not an actual MSI employee, though, is he? Damn, if I was them I'd put the man under an exclusivity contract (though I guess there might exist such an agreement without official employment). :sneaky:

 

[cmdrdredd]

He got it working, but is trying to make voltage work as an offset so it can idle at a lower power state.

Today, 10:13 | posts: 9,666 | Location: Taganrog, Russia
Some update and good news:

I decided to shift my vacation a bit and my Lightning finally arrived yesterday. Happily I managed to bypass the main problem with new I2C bus access on Keplers in less than two hours after installing the card. So now all 3Lightning voltage controllers (core, memory and PLL) and thermal controllers are visible to software and programmable. There are still some things left to do, right now core voltage control disables dynamic core voltage adjustment, so changing voltage results in setting maximum fixed voltage in idle as well. But anyway even now it is better that "voltage control" offered on any 680 card by other vendors. I hope that I'll be able to implement alternate voltage control in offset form for dynamic Kepler volatge control.

Alexey Nicolaychuk aka Unwinder, RivaTuner creator

 

[DrBoss]

He got it working, but is trying to make voltage work as an offset so it can idle at a lower power state.

yea, he's got it working on a Lightning... i'd still like to hear him say it will also work on the 670 Power Edition.

 

[cmdrdredd]

yea, he's got it working on a Lightning... i'd still like to hear him say it will also work on the 670 Power Edition.

"Happily I managed to bypass the main problem with new I2C bus access on Keplers in less than two hours after installing the card. So now all 3Lightning voltage controllers (core, memory and PLL) and thermal controllers are visible to software and programmable."

Keplers...meaning GTX 600 series. The tweaks should work for the 670 as well.

 

[DrBoss]

MSI Afterburner 2.2.3 with Triple Overvoltage support for the 680 Lightning and 670 Power Edition available now:
http://event.msi.com/vga/afterburner/download.htm

I cannot confirm if it works considering i'm at the office. I will test it tonight, hopefully some of you can get your hands on it before me and post impressions.

 

[SirPauly]

MSI Afterburner 2.2.3 with Triple Overvoltage support for the 680 Lightning and 670 Power Edition available now:
http://event.msi.com/vga/afterburner/download.htm

I cannot confirm if it works considering i'm at the office. I will test it tonight, hopefully some of you can get your hands on it before me and post impressions.

Thanks!

 

[tviceman]

Take note that this update has only been tested with lightning. Uwinder specifically states he has not been able to test this on the 670 PE because he does not have one. I am offering him to send my card to him, but I have not been given a response yet.

 

[DrBoss]

any confirmation that this enables voltage control on the PE?

the release notes for the new AV (2.2.3) clearly suggest it does:
"Added core, memory and auxiliary PEXVDD voltage control for custom design MSI N670GTX Power Edition series graphics cards with uP6262 voltage regulators"

but then again, the packaging/box for my PE also indicated the voltage was unlocked.

 

[SirPauly]

It seems so!

One can manually change the core, memory and aux voltage with the Power Edition with 2.2.3 -- how this translates into real world testing -- don't know yet.

 

[tviceman]

downloading now! I will test it out real quick just to verify the changes. I probably won't run into extensive testing until this heat wave breaks.

EDIT: I haven't done any testing yet, but Nvidia has already issued a new bios for Lightning cards that will re-impose the hard voltage limitation. http://www.guru3d.com/article/msi-gtx-680-lightning-voltage-tweaking/ Nvidia is serious about locking the voltage down. Very bad news for tweakers and overclockers. I suspect, though, that going into the future AMD is going to do the same. If it can save a considerable amount of cost associated with RMA's, of course companies are going to adpot these kinds of policies.

 

[cmdrdredd]

http://forums.anandtech.com/showthread.php?t=2259776

Discussion here with controversial moves by Nvidia.

 

[tviceman]

The voltage slider IS adjusting the voltage output on my 670PE. I am not pushing the card to it's limits since it is so damn hot in my house, but I undervolted the card and it immediately crashed during a heaven loop. That is proof enough for me that it works now.

 

[RussianSensation]

The voltage slider IS adjusting the voltage output on my 670PE. I am not pushing the card to it's limits since it is so damn hot in my house, but I undervolted the card and it immediately crashed during a heaven loop. That is proof enough for me that it works now.

Be careful, the MSI Lightning overvoltage Guru3D write-up stated that the +100mV bump is applied NOT to stock 1.175V voltage of GTX670/680 cards but to whatever offsetting voltage the card boosts to with GPU boost. The worst part, it appears you can't accurately to measure this in software every time. If your card runs at 1260-1270mhz as it is, it may not be worth it to accidentally pump 1.3-1.32V into it to try and get another 50-60mhz. I probably wouldn't do it. Chances are you'll start losing some GPU Boost bins since with such voltage your card may exceed 70*C and also it's doubtful that there will be linear GPU scaling since at some point the 670/680 cards will start running into their memory bandwidth bottleneck.

Also, remember HD6950's being flashed into a 6970 using the 6970 BIOS? Later on some of those 6950s began to file. Eventually, users found out the memory voltage was changed from 1.0V to 1.1V, and despite the cards being "identical" on the outside some 6950 cards couldn't handle the 1.1V (hence the HD6950 to HD6970 "Shader unlock" came out to alleviate these issues). There may be other reasons why NV locked out GPU voltage for Kepler cards (maybe fragile 28nm TSMC process, etc.). Remember when 45nm Quad-Core E8000/Q9000 chips overclocked really well compared to 65nm E6600/Q6600/6700? Well later on people running them on older boards started killing them and we found out 45nm Intel CPUs couldn't handle as high of FSB Termination voltage (VTT) as the 65nm chips could. Sometimes the more refined manufacturing process can overclock better out of the box but will fail quicker with higher voltages. My HD6950 40nm GPU had a stock voltage of 1.175V, the same as the 28nm Kepler and HD7970 cards. It could be that the 1.175V voltage for current 28nm GPUs was already set towards the higher end of the spectrum for that 28nm node. While some 7970 users benchmark at 1.25-1.3V, I don't see many of those running their cards full out at those settings at home. Just my 2 cents.

P.S. I think Unwinder lives in Taganrog. If you send him your card, I bet it would take a month to turn around and cost a lot of $ in shipping fees.