Yesterday afternoon was interesting because I spent much of it revisiting the GeForce GTX 590, looking for problems. As you may have heard, some reviewers using early drivers had rather serious, quasi-explosive failures with overclocked GTX 590 cards. Like these guys at Sweclockers:
The trouble there, it seems, is that the card's built-in power-protection scheme didn't, you know, protect. We tested the GTX 590 with somewhat newer drivers, apparently updated to fix that problem, and didn't run into any sparks, flames, or even smoke. In fact, the power protection mechanism kicked in and saved our card from what was probably certain destruction, given what we did to it. From our review:
One has to be careful here, though, because the GF110 chips will definitely reach much higher clock speeds when given enough voltage—we reached 772MHz at 1025 mV, similar to the GTX 580—but you'll also definitely bump up against the GTX 590's power limiting circuitry if you push too hard. The result, as we learned, is that performance drops with the supposedly overclocked config.
That performance result is the preferred outcome, of course, when the alternative is fiery death. The videos going around online about GTX 590 failures have even led Nvidia to issue a note about GTX 590 overclocking protection in the release drivers.
This development, in turn, has led red-tinged conspiracy theorists—of which there are curiously many these days—to speculate, insinuate, and even outright assert that the final, release drivers for the GTX 590 limit clock speeds, either at stock or when overclocked.
The trouble with those claims is that, well, we found a decent amount of performance headroom in our GTX 590, even though it saved our bacon when we overvolted it like mad. Again from our review:
We eventually decided on a more mildly overclocked config in which the GPU core was raised to 690MHz, the GPU core voltage was increased from 938 mV to 963 mV, and the memory clock was tweaked up to 900MHz (or 3.6 GT/s). This setup was easily achieved with MSI's Afterburner software, proved quite stable, and, as you'll see in the following pages, performed consistently better than stock. The only thing left to do then was give these settings a name, since they lacked one. Folks, say hello to Wasson's Intrepid Clock Konfig, Extreme Dually—or WICKED. We've put WICKED and AUSUM head to head to see which is better.
WICKED worked, even though it raised the GTX 590's power draw enough to make us a little uncomfortable. There is some headroom in this card, for those with a good PSU and a measure of intestinal fortitude.
The next iteration of this same rumor came to us late yesterday, with the suggestion that perhaps those web release drivers with overclock protection are limiting clock speeds in games to 550MHz, below the 607MHz base clock for the GTX 590, along with the expected drop in performance. Our response was to install the 269.91 drivers and try them for ourselves. After re-running a portion of our test suite, including AvP, Civ V, and F1 2010, we found zero performance differences between the drivers we used in the review and the public 269.91 package. Also, GPU-Z reported a 607MHz clock speed when we had it log clock frequencies while some of our tests were running. Boring, no? But the GTX 590 still works as expected.
The rumor mill wasn't finished, though. Another assertion of problems reached us yesterday afternoon via different channels, based in part on these sweet thermal camera readings, which clearly show temperatures as high as 112° C at the center of the GTX 590. This info, we were told, proves the GPU's on-die thermal sensors are being programmed to under-report temperatures. The solution? We should try an infrared thermometer aimed at the back of the card, instead.
So we did.
Probably didn't need to, though. If you look carefully at those camera readings, you'll see that the highest temperatures reached in their measurements are for the power regulation circuitry at the center of the card, not the two GPUs on the sides. Our own readings with an IR thermometer showed that the metal plates behind the GPUs were cooler than our prior temperature sensor readings had been. In other words, the sensors were probably not reporting artificially low results. Yes, the power circuitry gets hotter—up to 106° C, in our measurements—but we have no sense that such temperatures constitute a problem. Hot VRMs aren't exactly uncommon.
All of which leads us back to exactly where we started, with no evidence of basic problems in the GTX 590's operation beyond, you know, the initial exploding drivers. Heh. We do have some evidence of additional, sloppily made insinuations of problems, which I suppose shouldn't be too surprising.
Trouble is, a lot of these forum rumors tend to be given tremendous credence by a lot of folks. Heck, every once in a blue moon, one of those rumors blows up into something big because there's a real problem underlying it. That could yet be the case with the GTX 590—or the Radeon HD 6990 or, well, I guess Sandy Bridge already went there. Such rumors are also an intriguing source of information because so many of 'em seem to be planted by a major, engineering-focused organization—you know, a competing firm.
Over the years, we've found that we could spend a huge proportion of our editors' time tracking down rumored problems of this sort with depressingly few interesting results. It's generally just not a good use of our time, especially because picking out the rumor that might actually have legs is terribly difficult to do at first glance. We did burn a few cycles making a run at some of the latest rumors about the GTX 590, though, and we were at least able to verify the product's proper operation. We've done such things in the past, a number of times, without even writing about it.
We're not sure who benefits more from a post like this one: the company that produced the product, because we're refuting negative rumors, or the competition, who gets to see a competing product's basic engineering questioned in the media. We kind of learn toward the latter, which is why we don't expect rumors of this nature to stop seeping from the darker corners of the 'net any time soon. We continue to puzzle over how best to serve you, the reader, in this context.
|1. GKey13 - $650||2. JohnC - $600||3. davidbowser - $501|
|4. cmpxchg - $500||5. DeadOfKnight - $400||6. danny e. - $375|
|7. the - $360||8. Ryszard - $351||9. rbattle - $350|
|10. Ryu Connor - $350|
|The SSD Endurance Experiment: Only two remain after 1.5PB||36|
|Friday night topic: Conspiracy theories||54|
|GeForce 344.11 WHQL drivers support new cards, new games, G-Sync||3|
|Deal of the week: A 23'' IPS monitor for $150, a 200-mm fan for free, and more||23|
|GeForce GTX 970, 980 cards already widely available||15|
|Curved VA panel powers 27'' Samsung monitor||15|
|Android L to encrypt devices by default||6|
|Nvidia's GeForce GTX 980 and 970 graphics cards reviewed||293|
|Thursday Night Shortbread||17|