Man did nvidia cheap out on the reference gtx670

[rickon66]

I don't think there's a requirement to follow reference design. I mean, if some vender is making GTX 670s can they simply refuse to offer a reference model? Isn't that what Asus did or does Asus have a reference model I don't know about?

For certain Gigabyte did not produce a reference, but offered an OC version with a great cooler at MSRP

 

[cmdrdredd]

As long as it works and can run OC who cares.. 2 years later when it dies, doesn't matter!! It's no longer under warranty!

ps. Srsly don't worry... u can always bake it back to life.

I think generally the reference card works as intended, but when you start overclocking and going past the stock settings things can start to go awry. I haven't heard too many people complain about MSI or Galaxy reference cards. Maybe the occasional somebody who thinks the fan makes noise but nothing really broken that I've seen. Although I'll admit I didn't specifically search for this.

 

[Lepton87]

When it comes to reference cards X2900XT was the first card that really used a high quality PCB, it was sort of necessary with 512bit bus, the cooler was a bit lacking though.

 

[lopri]

The build quality and quality components could turn out to be rather a big deal for Kepler especially.

http://www.anandtech.com/show/5699/nvidia-geforce-gtx-680-review/4
http://www.behardware.com/articles/857-4/gpu-boost-turbo-non-deterministe.html

By all accounts, NV seems to have been able to achieve very smart load balancing in terms of voltages/frequencies, to the point that overclocking looks meaningless. I do like this development a lot, but the hardware needs to be built to last. If not, it could be a recipe for performance degradation in relatively short time since the mechanics put in there depend on environment.

 

[toyota]

The build quality and quality components could turn out to be rather a big deal for Kepler especially.

http://www.anandtech.com/show/5699/nvidia-geforce-gtx-680-review/4
http://www.behardware.com/articles/857-4/gpu-boost-turbo-non-deterministe.html

By all accounts, NV seems to have been able to achieve very smart load balancing in terms of voltages/frequencies, to the point that overclocking looks meaningless. I do like this development a lot, but the hardware needs to be built to last. If not, it could be a recipe for performance degradation in relatively short time since the mechanics put in there depend on environment.

I can tell you that in 3 weeks I went from 1295/7000 stable in Heaven to not even being able to run 1277/6800 without crashing. thats on the same settings and with the same or even cooler ambient temps.

 

[Zanovar]

I can tell you that in 3 weeks I went from 1295/7000 stable in Heaven to not even being able to run 1277/6800 without crashing. thats on the same settings and with the same or even cooler ambient temps.

Im nervous,looping heaven as we speak.

 

[lopri]

I can tell you that in 3 weeks I went from 1295/7000 stable in Heaven to not even being able to run 1277/6800 without crashing. thats on the same settings and with the same or even cooler ambient temps.

1295 MHz..?! Didn't you push a bit too much? I thought these things normally overclock around 1100~1200 MHz.

 

[toyota]

1295 MHz..?! Didn't you push a bit too much? I thought these things normally overclock around 1100~1200 MHz.

it was 1189 boost out of the box and many go to 1229 even. 1250-1300 is about the typical oc range. mine did 1301 for 3D mark but needed 1295 for Heaven when I first got it. but like I said the max oc I could maintain in those benchmarks got lower and lower over the weeks.

 

[lopri]

@Toyota: Mind if you sharing what brand of your 670 is?

 

[toyota]

@Toyota: Mind if you sharing what brand of your 670 is?

it was the Evga FTW.

 

[lopri]

ty

 

[Cookie Monster]

Interesting observation toyota. Maybe Ill re-bench the GTX680 I have tonight to see if Im experiencing this.

 

[Grooveriding]

I can tell you that in 3 weeks I went from 1295/7000 stable in Heaven to not even being able to run 1277/6800 without crashing. thats on the same settings and with the same or even cooler ambient temps.

This is true. My cards can no longer run the clocks they could when I first got them, or at least one of them cannot. Was too lazy to isolate it.

Went from 1235 down to 1190 to stop artifacts that developed.

These cards seems to be running on the edge to begin with, probably why voltage control was locked down on them. It's interesting to see degradation taking place on cards that have hard locked voltage though. Can't complain I guess, unless they degrade below stock performance.

 

[lopri]

WTF are you guys serious? Grooveriding, which brand is your 680s?

 

[Cookie Monster]

Has your temperatures remained the same?

 

[Grooveriding]

Mine are both EVGA reference bought on release day. My temps are completely stable. The highest these cards have seen is 40C under load, usually 35C though.

I think they just couldn't handle much of an overclock. Stock boost clock is 1100 on these two I believe, so I am still getting what I paid for, but not much more than that

 

[felang]

it was 1189 boost out of the box and many go to 1229 even. 1250-1300 is about the typical oc range. mine did 1301 for 3D mark but needed 1295 for Heaven when I first got it. but like I said the max oc I could maintain in those benchmarks got lower and lower over the weeks.

Can 670's be 0verclocked at 6mhz increments? My 680 jumps 12 or 13 mhz at a time.

 

[Zanovar]

This is true. My cards can no longer run the clocks they could when I first got them, or at least one of them cannot. Was too lazy to isolate it.

Went from 1235 down to 1190 to stop artifacts that developed.

These cards seems to be running on the edge to begin with, probably why voltage control was locked down on them. It's interesting to see degradation taking place on cards that have hard locked voltage though. Can't complain I guess, unless they degrade below stock performance.

,anyone else experiencing this?kinda worrying.

 

[Cookie Monster]

Well, im getting the same score I did a month ago. However sometimes driver revisions can affect your overclocks.

 

[Jaydip]

I am getting the same score as well.No issue here

 

[Rvenger]

I don't OC mine so I guess I will never know.

 

[Zap]

Nvidia always has crappy reference designs. Reference in general is usually a bad idea

That is not always true. There have been times when the pendulum has swung the other way, with many enthusiasts/overclockers recommending only reference cards (of a certain GPU).

Agreed. Unless you're an engineer or very technical savvy, it's diffcult for an average Joe, to comprehend what actually is quality.

Exactly! Just being chock full of components doesn't mean it is quality. For instance look inside some real efficient PSUs. Due to design and good components, they don't necessarily need a ton of big heatsinks anymore.

One way to determine "quality" is through PCB layer count. How many of you guys here can determine that just by looking at the graphics card? Trace layout can also affect quality. Component choice - these days of digital VRMs means you may not necessarily need two pounds worth of caps and coils with the requisite PCB space to properly control the voltages.

From now on, I probably won't recommend blower type of cards for single-GPU users with decent cases.

Some of us have known that for years. In fact, some guy at Overclock.net raged at me and then put me on his ignore list just because I stated such. This was BITD when the single PCB GTX 295 first came out. Everyone said the reference cooler sucked because it did not exhaust out the back of the case. I said it was fine because in my testing (I worked for BFG Tech back then) it ran cooler and quieter than the dual PCB version even inside a closed case. I also linked to some review sites where the reviewers (to their astonishment) found the same results.

Yeah, it was as if I spit on that guy's God or something.

 

[cmdrdredd]

I have a gtx 670 ftw and it clocks the same as day one.

 

[Zanovar]

This is true. My cards can no longer run the clocks they could when I first got them, or at least one of them cannot. Was too lazy to isolate it.

Went from 1235 down to 1190 to stop artifacts that developed.

These cards seems to be running on the edge to begin with, probably why voltage control was locked down on them. It's interesting to see degradation taking place on cards that have hard locked voltage though. Can't complain I guess, unless they degrade below stock performance.

Hmm,my 1244 core oc(max rock solid oc) is now unstable,1215 seems stable atm.wtf is going on here?.mem oc is fine btw.Anyone else having to lower their overclocks?

 

[Meaker10]

My heatsink is even smaller: