Meet Zotac’s GeForce GTX 465
Zotac was kind enough to send over one of its first GeForce GTX 465 cards, based on Nvidia’s reference design. The board is identical to the GeForce GTX 470 Nvidia sent over for launch, measuring 9.5” long (half an inch shorter than the Radeon HD 5850). In fact, everything from the dual-slot form factor to the display output connectivity is the same—you get a pair of dual-link DVI ports and a mini-HDMI connector, but only two outputs are usable at a time.
The rated thermal design power here is 200W—15W less than the GeForce GTX 470 and 50W less than the GeForce GTX 480. In my initial review of the GTX 470 and 480, however, I noted a significant discrepancy between the power ratings of Nvidia’s GF100-based cards and competing boards from AMD. Indeed, it seems that the two companies define their power numbers differently, just like AMD and Intel’s CPU ratings are incomparable. Per Nvidia, TDP is a measure of maximum power draw over time in real-world applications, and does not represent the maximum power draw in cases like FurMark. That helps explain why, as you’ll see in our benchmarks, the GeForce GTX 480 draws 133W more than the GTX 465 under load, despite an official TDP difference of 50W.
Specs In Depth
When Nvidia first introduced us to GF100, one aspect that stood out to me was how modular it looked. The quartet of Graphics Processing Clusters, each with four Streaming Multiprocessors, and the six ROP partitions—it just looked like pieces were meant to be pulled out to create derivative designs. And that’s sort of what Nvidia’s doing here with its GeForce GTX 465, only instead of manufacturing a smaller, cheaper GPU based on the same Fermi architecture, the company is turning off big portions of its pricey GF100.
For more background on the GF100 design, read back to our first preview of the chip from January. For more detail on how that piece of silicon worked its way into the GeForce GTX 480 and 470, check out the review I published in March.
GeForce GTX 465 leverages three GPCs—one is disabled entirely. Of the 12 remaining SMs (remember, there are four SMs per GPC), Nvidia disables one, leaving 11 SMs, each with 32 CUDA cores. That’s where we get the 352-core number.
|Header Cell - Column 0||GeForce GTX 465||GeForce GTX 470||GeForce GTX 480|
|Graphics Processing Clusters||3||4||4|
|Graphics Clock||607 MHz||607 MHz||700 MHz|
|Shader Clock||1,215 MHz||1,215 MHz||1,401 MHz|
|Memory Clock||802 MHz||837 MHz||924 MHz|
|Memory Bandwidth||102.6 GB/s||133.9 GB/s||177.4 GB/s|
|Texture Filtering Rate||26.7 GTexels/s||34 GTexels/s||42 GTexels/s|
|Connectors||2 x DL-DVI, 1 x mini-HDMI||2 x DL-DVI, 1 x mini-HDMI||2 x DL-DVI, 1 x mini-HDMI|
|Power Connectors||2 x 6-pin||2 x 6-pin||1 x 6-pin, 1 x 8-pin|
|Recommended Power Supply||550W||550W||600W|
|Thermal Design Power||200W||215W||250W|
|Thermal Threshold||105 degrees C||105 degrees C||105 degrees C|
Also remember that, in Fermi, texture units are tied to SMs. There are four texture units per SM. Do the math—11 SMs times four units each gives you 44 total texture units. Moreover, geometry performance dips, as we’re only dealing with 11 PolyMorph engines now.
The GF100’s back-end is independent of the GPCs, yet Nvidia makes cuts here too, turning off two of the six ROP partitions and dropping pixel throughput to 32/clock. This has the side-effect of axing two 64-bit memory interfaces, taking what starts as a 384-bit path and reducing it to 256-bits. As a result, we have a nice, even 1GB of GDDR5 memory pushing up to 102.6 GB/s.
Nvidia isn’t changing the graphics and shader clocks versus GeForce GTX 470—they’re still 607 and 1,215 MHz, respectively. Memory clocks do drop slightly though, from 837 down to 802 MHz. Of course, all of this tweaking, tuning, and massaging is a means to an end—performance. Let’s see how the GeForce GTX 465 measures up.