Anyone have one of the new 6 series cards yet?

[Dan3345]

It could be drivers.. Would make sense that a new card would have very immature drivers, it just seems unlikely that they would cause the computer to shut down. Has he checked the temps?

[Phalanx108]

No, it doesn't beat two 580, it can't touch them, and it can't touch the 590. Not even 680 or 7970 can touch 590 or 6990.

 

It does beat the 580. But on the downside 670 has pretty crappy build, with a dodgy PCB and a worthless cooler.

 

Would like to see your evidence and results on this. Dont have a test bench? No worries.

 

Also, I agree with you on the second point. For a less than excellent company the Gigabyte card is currently the best buy alongside the Gainward Phantom model, simply because of the non reference cooler.

Edited May 22, 2012 by Phalanx108

[FMod]

I have no idea what his PSU is; the PC kept shutting down with a 7970, and he told me it was because of power consumption with a 1000w PSU.

The amount of current required to overload a 1000W PSU is such that the card would burn, as in smolder and catch on fire. It's kind of easy to notice. Not just keep shutting down.

 

Although I can think of some very old PSU that had 18A current limited lines, and if somehow both his GPU aux power and his CPU or ATX power were on the same line (but how?), it would not be able to put out more than a couple hundred watt. I don't think they ever made these PSU rated for 1000W, however.

 

 

Also, I agree with you on the second point. For a less than excellent company the Gigabyte card is currently the best buy alongside the Gainward Phantom model, simply because of the non reference cooler.

Really the best 670 if you have to get one is the Asus DC2 model. It's got a better PCB than others, with 6+2 phase power.

 

No, it doesn't beat two 580, it can't touch them, and it can't touch the 590. Not even 680 or 7970 can touch 590 or 6990.

Would like to see your evidence and results on this. Dont have a test bench? No worries.

Here you go:

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_670/1.html

 

The relevant comparison of high-end cards in high-res looks like this:

 

http://img833.imageshack.us/img833/6131/tmp1.png

 

I don't have to go to the Moon in person to know it's about 1/4 Earth's radius and 200,000 miles away.

 

Now, since none of the reviewers seem to have bothered to pitch 580 SLI versus 680 directly, here's a couple comparisons between 590 and 580 SLI:

 

http://www.xbitlabs.com/articles/graphics/display/geforce-gtx-580-sli_7.html#sect0

http://www.hardocp.com/article/2011/03/24/asus_geforce_gtx_590_video_card_review/9

 

As you can see, 580 SLI devastates 590, which is effectively 570 SLI performance-wise. Which, again, is faster than a single 680.

[Phalanx108]

Really the best 670 if you have to get one is the Asus DC2 model. It's got a better PCB than others, with 6+2 phase power.

Here you go:

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_670/1.html

The relevant comparison of high-end cards in high-res looks like this:

http://img833.imageshack.us/img833/6131/tmp1.png

Now, since none of the reviewers seem to have bothered to pitch 580 SLI versus 680 directly, here's a couple comparisons between 590 and 580 SLI:

As you can see, 580 SLI devastates 590, which is effectively 570 SLI performance-wise. Which, again, is faster than a single 680.

 

We havent been able to get any of the Asus cards into the store unfortunately, so i cant comment on its performance.

 

Also, as for online benchmarks, i tend to find that the ones we get done in store come up with radically different numbers. Some for the better, but a shockingly high amount for the worse. More common on the late AMD 6XXX cards, and for the lower end NVidia 5XX's.

[FMod]

Also, as for online benchmarks, i tend to find that the ones we get done in store come up with radically different numbers. Some for the better, but a shockingly high amount for the worse.

Which indicates that your personnel is probably not as well-equipped for or not as experienced at doing these benchmarks and doesn't manage to extract the full amount of performance that the cards are capable of.

 

Do you use 2x00K or 3xx0 K/X CPU when testing video cards?

Are they overclocked to 5 GHz sharp and is hyper-threading turned off?

Do you install exactly one RAM module per channel and operate it at 2133 MT/s?

Is this done with Asus ROG motherboards, which can deliver marginally better performance?

Do you use low ripple, stable voltage power supplies like high-end Seasonics for your overclocking tests?

Is the specialist performing them experienced in overclocking ranging from hard voltmods to timings fine-tuning?

Do you use clean OS without as much as a third-party firewall, disconnected from the internet where not required?

Do you make sure to perform comparative testing on the same day with the same drivers and not recycle old results?

Are your OS, games and benchmark applications installed on one Intel or Crucial drive in AHCI configuration?

Do you actually connect 120 Hz displays for 1080p and below tests and 30" ones for 2560x1600 tests?

Do you use wired PS/2 keyboards and simple USB optical mice with standard 125 Hz polling rate?

Is your framerate monitoring software configured and verified not to cause any fps loss?

 

For a fair comparison of video cards, you should provide real loading conditions and eliminate all other bottlenecks.

Now, most sites don't tick all of these boxes. Sometimes that's fine, because they don't all have equal effect on performance, and TPU is actually a good one. Other times you run into a mess like h--dware---ven that doesn't even care to script their game tests. You're probably in between, but might not be ticking some of these boxes that dedicated reviewers do. And when talking about single percentage margins of difference, even these small ones matter.

 

I know to a very high degree of certainty that the comparative performance order is as indicated in that chart, at least in the cropped section. This not only comes from tens of benchmarks all agreeing on this, and not only from thousands of user-submitted benchmark results, both of which you can use yourself to verify it. For me it also comes from known trustworthy reviews, including by people I know in person, from some of my own experience, and from the personal reports of fellow overclockers who recently or not so recently switched some of these cards for others on the list.

[Dan3345]

Also, as for online benchmarks, i tend to find that the ones we get done in store come up with radically different numbers. Some for the better, but a shockingly high amount for the worse.

Which indicates that your personnel is probably not as well-equipped for or not as experienced at doing these benchmarks and doesn't manage to extract the full amount of performance that the cards are capable of.

 

Do you use 2x00K or 3xx0 K/X CPU when testing video cards?

Are they overclocked to 5 GHz sharp and is hyper-threading turned off?

Do you install exactly one RAM module per channel and operate it at 2133 MT/s?

Is this done with Asus ROG motherboards, which can deliver marginally better performance?

Do you use low ripple, stable voltage power supplies like high-end Seasonics for your overclocking tests?

Is the specialist performing them experienced in overclocking ranging from hard voltmods to timings fine-tuning?

Do you use clean OS without as much as a third-party firewall, disconnected from the internet where not required?

Do you make sure to perform comparative testing on the same day with the same drivers and not recycle old results?

Are your OS, games and benchmark applications installed on one Intel or Crucial drive in AHCI configuration?

Do you actually connect 120 Hz displays for 1080p and below tests and 30" ones for 2560x1600 tests?

Do you use wired PS/2 keyboards and simple USB optical mice with standard 125 Hz polling rate?

Is your framerate monitoring software configured and verified not to cause any fps loss?

 

For a fair comparison of video cards, you should provide real loading conditions and eliminate all other bottlenecks.

Now, most sites don't tick all of these boxes. Sometimes that's fine, because they don't all have equal effect on performance, and TPU is actually a good one. Other times you run into a mess like h--dware---ven that doesn't even care to script their game tests. You're probably in between, but might not be ticking some of these boxes that dedicated reviewers do. And when talking about single percentage margins of difference, even these small ones matter.

 

I know to a very high degree of certainty that the comparative performance order is as indicated in that chart, at least in the cropped section. This not only comes from tens of benchmarks all agreeing on this, and not only from thousands of user-submitted benchmark results, both of which you can use yourself to verify it. For me it also comes from known trustworthy reviews, including by people I know in person, from some of my own experience, and from the personal reports of fellow overclockers who recently or not so recently switched some of these cards for others on the list.

Props to you Fmod. Here it is - http://videocardz.com/31650/geforce-gtx-685-gk110-features-4gb-512bit-memory Supposedly that is the specs for the rumored GTX 685. Running on the GK110. I gotta say, if this is real its a huge screw up for Nvidia. It never looks good for a company to release a product and claim to stand behind their product and them just as soon as that one came out they release a revised (some would say fixed) version. Makes em look silly.

 

But at least this one appears to be worthy of $400 dollars. No telling of what the PCB will be, but it does have 4 gigabytes of VRAM instead of a measly two and many more cuda cores.

[Phalanx108]

Do you use 2x00K or 3xx0 K/X CPU when testing video cards?

Are they overclocked to 5 GHz sharp and is hyper-threading turned off?

Do you install exactly one RAM module per channel and operate it at 2133 MT/s?

Is this done with Asus ROG motherboards, which can deliver marginally better performance?

Do you use low ripple, stable voltage power supplies like high-end Seasonics for your overclocking tests?

Is the specialist performing them experienced in overclocking ranging from hard voltmods to timings fine-tuning?

Do you use clean OS without as much as a third-party firewall, disconnected from the internet where not required?

Do you make sure to perform comparative testing on the same day with the same drivers and not recycle old results?

Are your OS, games and benchmark applications installed on one Intel or Crucial drive in AHCI configuration?

Do you actually connect 120 Hz displays for 1080p and below tests and 30" ones for 2560x1600 tests?

Do you use wired PS/2 keyboards and simple USB optical mice with standard 125 Hz polling rate?

Is your framerate monitoring software configured and verified not to cause any fps loss?

 

Cant say ive ever gone into that much detail with the rest of the staff. Most of my job is setting it up hardware side and ensuring that any hardware faults are dealt with. I can, however answer most of those questions.

 

"Do you use 2x00K or 3xx0 K/X CPU when testing video cards?" i7 3930k.

"...is hyper-threading turned off?" Yes.

"Is this done with Asus ROG motherboards, which can deliver marginally better performance?" EVGA X79 Classified.

"Do you use low ripple, stable voltage power supplies like high-end Seasonics for your overclocking tests?" The one in the test bench is actually a Seasonic x1050.

"Is the specialist performing them experienced in overclocking ranging from hard voltmods to timings fine-tuning?" He says he is, take it with a grain of salt though, ive never really been fond of the guy.

"Do you make sure to perform comparative testing on the same day with the same drivers and not recycle old results?" Of course. We scrub the results and start fresh if it takes too long.

"Are your OS, games and benchmark applications installed on one Intel or Crucial drive in AHCI configuration?" Not as such. An OCZ RevoDrive X4 240gb in AHCI.

"Do you actually connect 120 Hz displays for 1080p and below tests and 30" ones for 2560x1600 tests?" I believe that the one for 1080p tests is only a 75Hz :\

"Do you use wired PS/2 keyboards and simple USB optical mice with standard 125 Hz polling rate?" Couple of crappy beige ones, yeah.

"Is your framerate monitoring software configured and verified not to cause any fps loss?" Not too sure on this one.

 

 

IMPORTANT EDIT: Turns out one of the 580's used in the benchmark (2 SLI 580 vs 1 670) was dying. Put a little too much voltage through it and it didnt do too well. RMA sent, though none of us are really expecting a replacement, due to it being the "specialist's" cock up.

Edited May 28, 2012 by Phalanx108