The tide is turning
NVIDIA has been hitting all the right notes of late. Taking the graphics-card market as a case in point, the top-end GeForce GTX 500-series cards have been well-received, while the mid-range continues to be populated by the excellent GeForce GTX 460 GPUs. Switch gears and NVIDIA's mobile business, predominantly based around the Tegra chip, is in the ascendency.
The company's nascent success for mid-to-high-end graphics can be attributed to the painful lessons learnt when designing the maligned GeForce GTX 400-series GPUs. The newer 5-series cards - GTX 580 and GTX 570 - trim away much of the fat and show just how potent the underpinning Fermi architecture can be. But the tasty performance of these GPUs is soured by a retail price which starts at £265 and rises to over £400 for the very best GeForce cards.
Bringing in the 5-series GPUs from the top and working down, the next logical step for NVIDIA would be to release a GeForce GTX 560, and this is precisely what it's doing today. Grabbing all the niceties of the new Fermi architecture and using a model-number nomenclature that enthusiasts have confidence in - heck, GTX 460 has been to go-to mid-range GPU for NVIDIA's supporters for the last eight months - the scene is set for a new mid-range champ to be born.
A history lesson
Cast your mind back to the two-flavour GeForce GTX 460. While NVIDIA engineered a sweet price-to-performance ratio by debuting both the 1,024MB- and 768MB-equipped cards with competitive retail prices, the actual GPUs weren't as potent as the Fermi architecture suggested.
GF114 and GF104
You see, this is the GF104 die for the GTX 460 GPUs, and it comprises of what NVIDIA terms eight Streaming Multiprocessors (SM), each composed of 48 CUDA cores, making 384 in total. However, lower-than-expected yields and associated manufacturing issues forced NVIDIA to deactivate one complete SM in each GTX 460, reducing the total CUDA/shader count to 336. Accompanied by conservative core, shader and memory frequencies, NVIDIA provided partners with plenty of wiggle room for factory-overclocked cards.
GeForce GTX 560 - doing mid-range properly
A look back at the GTX 460 is important for realising how NVIDIA has designed the GTX 560. Taking on board the advances in manufacturing and improved yields 10 months on from the initial Fermi launch, the company can now release full-fat GPUs that are augmented with higher clocks: just compare the GTX 580 vs. GTX 480 for confirmation.
NVIDIA's GeForce GTX 560 1GB uses what is known as the GF114 die. It is composed of the same eight SM arrangement - each SM is functionally the same on both GPUs - and general layout of the GF104, shown above, but this time the retail card gets to keep all the goodies. This means 384 shaders, 32 ROPs, and a 256-bit memory bus. However, more than just filling out the basic die - which is the same size for GF114 and GF104 - the 5-series GPUs also boast improvements to enhance z-culling and much higher frequencies than their 4-series brethren. NVIDIA also throws in the hardware monitoring for the 170W TDP GPU, explained right over here.
Kind of confused with all the numbers and stats? Let us explain it all in a table.
|GPU||GeForce GTX 580 1,536MB||GeForce GTX 570 1,280MB||GeForce GTX 480 1,536MB||GeForce GTX 470 1,280MB||GeForce GTX 560 Ti 1,024MB||GeForce GTX 460 1,024MB|
|Memory interface||384-bit, 1,536MB GDDR5||320-bit, 1,280MB GDDR5||384-bit, 1,536MB GDDR5||320-bit, 1,280MB GDDR5||256-bit, 1,024MB GDDR5||256-bit, 1,024MB GDDR5|
|Board power (TDP)||244W||219W||250W||215W||170W||160W|
GeForce GTX 560 comparative performance analysis
First off, the new GPU will only be available with a 1,024MB frame-buffer, which means that NVIDIA drops the two-SKU approach of the GTX 460. We know the company has been very conservative with the GTX 460's default frequencies, with retail cards ramping up from 675MHz core to 800MHz without breaking stride. NVIDIA, then, increases frequencies the core/shader clocks by over 20 per cent for the GTX 560, along with a 10 per cent bump for the memory, translating to more of everything: shading, texturing and bandwidth.
More power is gained by using the full 384 CUDA cores/shaders on the GF114, so we can subjectively add another 10 per cent extra performance on top. We also know that, like its 5-series brothers, the GTX 560 can filter an FP16 texture in one clock cycle, and this is why the rate is so comparatively impressive against many 4-series cards, though the GTX 460 has the same capability.
Break it all down and the GeForce GTX 560's projected performance is akin to a super-duper-clocked GTX 460 1GB, plus a few extra per cent by way of the additional CUDA cores.
Nomenclature time - Titanium, baby
In fact, the GTX 560 is what the GTX 460 really should have been at the get-go, we think. NVIDIA is keen to promulgate the new GPU's effectiveness by bringing back nomenclature from the time of the GeForce 3 (remember that?). Yup, it's not good enough to simply refer to the retail GF114 as the GTX 560. No, sir, it has to be the GeForce GTX 560 Titanium - a name that's supposed to evoke feelings of strength, nimbleness, and plain awesomeness.
NVIDIA's GeForce GTX 560 Ti 1GB graphics card builds on the success of the GTX 460 1GB by using a revised die that's fundamentally more power-efficient. GF114's tweaks mean that it activates the dormant Streaming Multiprocessor - leading to more cores, tessellation, and texture units - and ramps up clocks on all fronts.
Due to be priced at around £199 at launch, GTX 560 Ti will replace the GTX 470 in the upper-mid-range line-up. The pricing puts it in competition with AMD's Radeon HD 6870 1GB GPU, by the way.