Review: GTX 280's single GPU gets a boost
XFX was kind enough to send us two of their most powerful single-GPUs based on GT200 core. We know them as GTX 280 and GTX 260, but it’s their price and performance that made sure we don’t forget them. The cards themselves didn’t change since the launch, but the price did, and we’re talking about a price drop of over €100 for GTX 280. The new price adjustments made GTX 280 much more affordable for an end user, and if you want the fastest GPU around – look no further than this baby. Of course, if you want even more speed then XFX’s offering might better suit your needs, and today we’re talking about XFX GF GTX 280 670M 1GB GDDR3 DUAL DVI TV PCI-E XXX.
670M in the card’s name suggests the core speed of 670MHz. XFX did a core overclock of almost 70MHz whereas the shaders got a boost from reference 1296MHz and now run at 1458MHz. The next picture shows a dual slot XFX 670M XXX card with reference cooling. The cooler performs well but it’s a bit too loud for our taste. The temperatures on the overclocked card went up to 86 degrees Celsius.
GTX 280 has a 1GB frame buffer and uses 512bit memory interface coupled with a fast GDDR3 memory running at 2214MHz. XFX has taken this a step further so the memory on their GTX 280 card runs at 2500MHz. So, the memory bandwidth also got a boost from 140GiB to about 160GiB.
XFX did a nice paintjob on their card, and you can see the silhouette of a wolf in the background. The back of the card also got a nice make-up job, so it’s worth taking a look at it.
Nvidia stuck to 65nm process with their GT200 chip and its 240 stream processors, and they managed to pack as much as 1400 million transistors to a surface of almost 600mm2. That’s how this chip got a title of the largest single GPU ever made. Compared to the G80, GT200 has more than twice the number of transistors, where most of them are used to improve the GPU’s computing capabilities. Compared to the last generation, GT200 brings more threads, stream processors, better shading and texturing and more memory and memory bandwidth.
240 stream processors are connected to 32 ROPs, which are further connected to the 512bit memory interface. Here we have 8 ROP partitions, each packing 4 ROP units (8 x 4 = 32 ROPs), where each ROP partition has a 64bit connection to the main memory. 8 ROP partitions x64bit interface results in 512bit memory interface. GTX 260 (the weaker GT200 GPU) has 7 ROP partitions with the same 64bit memory interface per partition, which tells us that GTX 260 has a 448 bit memory interface. GTX 260 has 896MB of GDDR3 and 192 stream processors.
Each ROP partition on GT200 GPU is capable of processing 4 pixels per clock. Although they packed 6 ROP partitions the previous Geforce 8 and 9 generations are also capable of 4 pixels per clock. Just like this example of ROPs on GT200, we’ve got much evidence that indicate how GT200 is in many ways a greatly improved and enhanced G80 setup. Compared to last generation’s 128, the number of 240 stream processors is the best indicator of sheer power that GT200 GPU packs.
A view from above once again shows that this is a dual slot design, and towards the end of the card we see the two power connectors. You’ll need a 6pin and 8pin connectors to power this card. Maximum consumption is 236W in the worst case scenario. However, if you’re not using 3D (and have an appropriate chipset), HybridPower technology kicks in and powers down the card passing the task of rendering to the integrated graphics on your motherboard. GT200 can also power down the transistors that are not needed in a given moment and dynamically regulate frequencies and voltages. The card needs only 25W in idle mode whereas HD content viewing will need only 35W. We’re of course talking about pure 2D where the card’s frequencies are also downclocked to 300/100/200MHz (core, shaders, memory).
The card has a PCIe 2.0 interface but no DX 10.1 support. Still, DX 10.1 isn’t that important because it’s still not put to good use, but we might see some apps that’ll require it. GTX 200, as well as G80 and G92 based cards, feature PhysX support that helps the GPU with physics acceleration, but we currently don’t have any games to try it out. In order to use GPU’s PhysX acceleration you’ll need PhysX drivers. For now we can play with Tri SLI support, but three GTX 280 cards are not a cheap pleasure. Two SLI connectors are well hidden on the top of the card, so you can leave them there until they’re needed.
Nvidia thinks it’s quite important to portray GT200 (as well as G80 and G92) as a multifunctional architecture locked and loaded for more than just gaming. We see them use the phrase “beyond gaming” more and more often, and it should mean that we can use this card for many other tasks thanks to Compute Unified Device Architecture (CUDA). Using CUDA, we communicate with the GPU that feeds back the processed data. It’s probably closer to a multifunctional GPU than many think, but we’ll wait until we see the results.
Until then, we can do some gaming, and XFX ships a nice gift with their card – a full version of the game Assassin’s Creed.
In the box you’ll also find a SPDIF cable needed to route sound through an HDMI cable. You’ll also need a DVI-to-HDMI dongle that XFX doesn’t ship with their cards. I/O side features two dual link DVI-I with HDCP and one HDTV out.
The box features the same motif we’ve seen on XFX 670M XXX card.