Geforce GTX 460: Dual-strike against the Radeons
You're looking for a potent DirectX-11-card, fit for Full-HD-gaming for around 200 US-Dollar? Enter the Geforce GTX 460! The PCGH-Test reveals strengths and weaknesses of both models compared to a host of other options and older cards.
While being a feature beast and quite a bit faster than AMDs single GPU flagship, Nvidias Fermi launch back in March left the majority of gamers feeling underwhelmed. Partly, because the Geforce GTX 480 and 470 were noisy under load and took quite a hefty sip of electricty from the outlet, especially when only surfing the net or watching movies.
With Geforce GTX 460, dubbed "the hunter" by Nvidia opposed to "the tank", which was GF100), the californian company reworked major parts of the inner architecture of the chip while trying to keep the edge it has in tessellation-heavy DX11 workloads; but first things first.
Based on the GF104 GPU, the Geforce GTX 460 in general is a closer fit for the gamers' needs. It achieves performance on par with the GF100-based GTX 465 with only a fraction of the die size and equally lessened strain on your electricty bill. In order to make the most use out of the chips 1.95 billion transistors, Nvidia decided to only divide the chip into two parts called GPC compared to the GF100's four GPCs. "GPC" stands for Graphics Processing Cluster and in general consists of four SIMDs, called shader multiprocessors (SMs) by Nvidia, and more importantly, a rasterizer and a triangle setup unit of its own. Those are responsible for converting geometry data into pixels and the parallel execution of this task removes a serious bottleneck from the rendering pipeline all previous cards have suffered from.
Each shader multiprocessor now consists of three groups of 16 ALUs, or Cuda Cores, as Nvidia calls them, compared to 32 in GF100. Additionally, the number of texturing units per SM have been doubled and each one has been beefed up in order to process up to FP16-formats at full speed. For yield (and thus cost) reasons, out of the 2x 4 SMs in GF104 only seven are enabled in current GTX 460 products. So, each one has 336 ALUs and 56 TMUs working for your gaming pleasure.
The Geforce GTX 460 comes in two flavors: One with 768 MByte and one with a full gig of GDDR5 video memory. Which at first glance doesn't sound too surprising becomes more intriguing when considering the peculiarities of Nvidias current architecture. In Fermi-chips, memory channels and thus memory size, are closely tied to the chips' memory controllers which in turn are attached to L2-caches and ROP-Partitions. The latter are largely responsible for antialiasing and color blending performance. So, apart from having less memory and bandwidth at its disposal, a 768 MB GTX 460 will be performing a bit slower than its better equipped sibling. After reviewing our benchmarks, you should carefully consider, if the lost potential is worth a saving of about 20 to 30 dollars.
We have prepared a table providiging the most important theoreticals at a glance:
|Model||GTX 470||GTX 465||GTX 460/1G||GTX 460/768||HD 5850||HD 5830||HD 5770/1G|
|Codename||GF100||GF100||GF104||GF104||Cypress Pro||Cypress LE||Juniper XT|
|Process tech||40 nm||40 nm||40 nm||40 nm||40 nm||40 nm||40 nm|
|Geometry throughput (Mio. Tris/s.)||2.432||1.824||1.300||1.300||725||800||850|
|Pixel throughput (GPix/s.)||17,0||13,4||9,45||9,45||23,2||12,8||13,6|
|Texel throughput (GTex/s.)||34,0||26,8||36,4||36,4||52,2||44,8||34,0|
|Memory bandwidth (GByte/s.)||133,9||102,7||115,2||86,4||128,0||128,0||76,8|
|Clockrate Chip (Megahertz)||608||608||650||650||725||800||850|
|Clockrate ALUs (Megahertz)||1.215||1.215||1.350||1.350||725||800||850|
|Clockrate Vid-Mem (Megahertz)||1.674||1.604||1.800||1.800||2.000||2.000||2.400|
|Memory Interface (Bit)||320||256||256||192||256||256||128|
|PSU-Connectors||2x 6-pin||2x 6-pin||2x 6-pin||2x 6-pin||2x 6-pin||2x 6-pin||1x 6-pin|
|Power Consumption (TDP)||215 watts||200 watts||160 watts||150 watts||170 watts||175 watts||108 watts|
Now, in case you're wondering as to why we are listing much lower pixel rates than most other websites: The connections between the shader engine and the ROPs is only 2 pixels per clock wide per SM, thus in a 7 SM-part, only 14 pixels may pass any given cycle.
Without further ado, let's head straight to a few carefully selected benchmarks, depicting the performance you will experience in game with the given settings. More about our benchmarking philosophy and exact procedures of the tests (including youtube videos) can be found in this article (in german), more benchmarks are located in the original german version of this article. There you'll also finde more about power consumption and noise of Nvidias Geforce GTX 460 (you might want to use an online translator such as the offering from google for better understanding).
Grafikkarten-Rangliste 2016: 32 Radeon- und Geforce-GPUs im Benchmarkvergleich [Oktober]