Does a paltry 256MB of RAM matter? Apparently, it does, if you’re talking about Nvidia’s GF104-based GTX 460 cards.
In the October issue, we took a long look at Asus’s sweet GTX 460 768MB card. While that card impressed us for the most part, it did seem a little weak in a few areas, especially when you turned up antialiasing. This 1GB version isn’t just the same chip with another 256MB of GDDR5 memory slapped on.
The memory bus is actually wider than the 768MB card, at 256 bits wide instead of 192 bits wide. That extra bus width is managed by a fourth memory controller on board the chip (the 768MB card has only three memory controllers.) If you’re thinking that the 1GB version of the GeForce GTX 460 should have had its own model—perhaps GTX 463—you’re not alone. A lot of people have wondered why Nvidia would use the same nomenclature for these two different beasts. The chip itself is the same. The 1GB chip is based on TSMC’s 40nm process technology, and has the same 1.95-billion transistor count as the 768MB version.
Graphics cards have gotten faster and added more features. So we have to ask the question: is it really worth adding a second GPU to your system? Will you get enough of a performance boost to justify the extra power draw and added cost? The answer is more complex than a simple yes or no. It all depends on what games you’re running, how much you dial up features like anti-aliasing, whether you’ve dived into the world of stereoscopic 3D and what monitor you’re running.
Perhaps the most important factor in the decision is display resolution. If you’re running a 1680x1050, 22-inch display, a single midrange or high end card will get the job done. Adding a second GPU is overkill. If you’ve got a 30-inch, 2560x1600 display and want to crank up the AA and postprocessing features, then that second GPU can be a big help.
It’s been interesting watching the evolution of Nvidia’s Fermi graphics. We’ve seen a range of cards, all built using variations of the original chip—a 3 billion transistor monster that runs hot and consumes power like a vampire sucking blood from a hapless victim.
Now Nvidia is shipping a new Fermi, previously code-named GF104. Aimed at the hearts and minds of mainstream PC gamers, the GTX 460 is a new chip, ringing in at just under 2 billion transistors and substantially more power-efficient. Two versions of the chip are available, a low-end and a high-end version.
Sharp-eyed Maximum PC readers who care about performance will no doubt notice that Gigabyte’s GV-N470UD-13I GTX 470 runs at stock reference speeds but achieves almost identical benchmark scores to last month’s kick-ass overclocked EVGA GTX 470. Blame it on new drivers versus old.
To be fair, the N470UD-13I isn’t exactly a stock card. While the card ships at reference clock speeds for core, shader, and memory, Gigabyte builds the board using its Ultra Durable manufacturing methods, which includes two-ounces-of-copper PCB technology, Japanese solid capacitors, high-end Samsung or Hynix GDDR5 memory, and low RDS(on) MOSFETs, which are designed to minimize switching resistance for faster capacitor charging and discharging. The PCB itself is blue, unlike many reference GTX 470 cards.
Factory overclocked graphics cards seem too good to be true. You get increased performance plus the manufacturer’s warranty. XFX’s Radeon HD 5870 XXX was the first factory-OC’d version of that GPU we reviewed (May 2010); that card pushed core clocks to 875MHz and memory to 1,300MHz (5,200MHz effective.) Now MSI is jumping into the game, and unlike XFX, builds a custom cooler onto its 1GB R5870 Lightning.
If you have any doubts about the amount of customization MSI put into the R5870, one look at it tells you it’s not your typical reference card. The custom cooler uses two fans instead of one, and the heatsink is a massive chunk of metal that runs the length of the card and features numerous heat pipes. The PCB is also anything but stock, and extends about 3/4 of an inch taller than other Radeon HD 5870 cards. Stock Radeon HD 5870 cards run off an 8-pin and 6-pin power connector. The R5870 features support for two 8-pin connectors for “extreme overclocking.”
GPUs that cost $500 are all well and good, but the sweet spot for high-end graphics cards is in the $350–$400 range. That’s still a good chunk of change, but it can get you a card with close to 90 percent of the performance of high-end cards.
That’s certainly true of EVGA’s GTX 470 SC. Built on a cut-down version of Nvidia’s high-end, DirectX 11 GPU, this card posted eyebrow-raising benchmarks, pretty much putting it into a class of its own.
EVGA’s super-clocked GTX 470 GPU ships with 448 shader processors, running at 625MHz, with a shader clock of 1,280MHz. That’s a 3 percent faster core clock and 2.5 percent faster memory frequencies than the stock GTX 470. (The GTX 480 uses 480 shader processors at 700MHz). The 320-bit-wide memory interface pumps data to 1,280MB of GDDR5 running at 850MHz (3,400MHz effective.) Of course, the card supports the usual set of Nvidia features, including hardware SLI, PhysX acceleration, and 3D Vision Surround video.
Just about everyone knows that Nvidia’s hot new Fermi graphics chip is literally hot. So, when Asus bundled its new ENGTX480 card with a custom voltage tweaker for overclocking, we wondered if it was such a good idea.
After all, do you really need the card to run hotter? And with the speed of the ENGTX480, you probably don’t need the higher clocks anyway. The ENGTX480 ships with 32 shader processors (what Nvidia calls “CUDA cores”) disabled, yet the card still manages to be the fastest single-GPU card you can buy today.
We’ve never been major advocates of GPU overclocking, as the minor gains you achieve often don’t justify the added heat and instability. But there’s a clear difference between Billy Joe doing a maximum overclock on his GPU and a vendor overclocking the part at the factory.
So when XFX offered up its XXX Edition of the already-fast Radeon HD 5870, we were naturally curious. XFX pushes the HD 5870 to 875MHz (3 percent over the stock 850MHz) and juices the memory to 1,300MHz (8.3 percent over the stock 1,200MHz). At first blush, a 3 percent core overclock seems minimal. Given that the card costs about $430, versus about $405 for the stock XFX variant, is it worth the extra jingle?
To find out, we compared the performance of the XXX Edition to a standard XFX Radeon HD 5870, which is a stock card in every respect. Save for clock speeds, the two cards are identical: memory (1GB), ports (two DVI, one DisplayPort, one HDMI), and the reference cooling system. Because of the speed bumps to the XXX Edition’s core and memory clocks, its system idle power usage varies from the stock card, reaching 148W versus 141W.
With intense competition in the graphics market, add-in board (AIB) partners often look for ways to distinguish their products from one another, whether it's a nifty bundle or an exotic cooling solution. So what is XFX planning for Fermi? Absolutely nothing, says Legit Reviews.
"This afternoon we received confirmation that XFX, a division of PINE Technologies, will not be releasing any GeForce GTX 400 series graphics cards to the market when the cards become public next month," writes Nathan Kirsch, founder of Legit Reviews. "XFX was not listed as a launch partner for Fermi and did not issue a press release about the upcoming cards, which might come as a shock to many to many of our readers as they are one of the largest Nvidia AIBs in the world!."
Kirsch goes on to say that the decision belonged to XFX, not Nvidia, which should kill off any conspiracy theories that Nvidia's giving XFX the cold shoulder for carrying ATI hardware.
"It looks like XFX thinks that the Radeon HD 5000 series of graphics card is the right card for the high-end market," Kirsch explains. "From our conversation with XFX they mentioned that they have 'yet to see whether the fermented launch will reach an inglorious anti-climax' and that they want to 'Ferm up to who really has the big guns.'"
Interesting choice of words coming from a major player. XFX has been touted for its excellent 'Double Lifetime Warranty' policy, which allows registered users to transfer their warranty to a second owner.
Tomorrow's supposed to be the big day for Fermi (or early April, according to reports suggesting initial availability will be pretty tight), but even though the long wait is almost over, there's still plenty of demand for ATI's HD 5000 series.
"Graphics cards manufacturers who are selling both Nvidia and ATI are confirming that 58x0 and 59x0 series of ATI cards are still selling great and of course the cheaper 5000 series such as 57x0 and slower didn't slow up in sales," Fudzilla reports.
ATI's cards must be selling well, or at least well enough to have a tough time producing enough stock to fill the demand. Trying to snag a dual-GPU HD 5970 is no easy task, even as expensive as it is. And while HD 5870 cards are easier to come by, Newegg and other vendors often show several SKUs as being out of stock.
That's great for ATI, but it also allows the GPU maker to avoid have to slash prices. Could Fermi change that? We'll find out soon enough.