GeForce 600 Series Optimized for Physics with On-Board PhysX Block

Published: 1st Feb 2012, 16:38 GMT

Physics hardware makes Kepler/GK104 fast | SemiAccurate
A very interesting article was written by Charlie Demerjian (SemiAccurate). He focues on upcoming NVIDIA’s Kepler gpus and how are they being optimized for physics performance.

If you require detailed information please check the link in the bottom of this post. This is an overview of newest leaks pointed by SemiAccurate.

According to S|K sources GK-104, an upcoming NVIDIA mid-range gpu GK-104 is probably optimized for physics performance by implementing PhysX code to graphics processing unit. There is also a possibility that there is a separate block only to compute PhysX operations. However more sources claim that this only a specific code which is to increase performance of PhysX operations.

As for GeForce GTX 660 Ti (GK-104), it will require less than 225 Watts. Earlier this week we had some unverified information that this is probable TDP, but according to SemiAccurate GK-104 will require even less. There is also an information about how heatsink was reduces during newer generations of silicons. First A1 silicon required large heatsink (which was seen by SemiAccurate). According to newest information A2 has a smaller heatsink, while A3 is going to be even smaller.

Computing power of GK-104 is reported to oscillate around 3 Teraflops, which on 256-bit interface is impressive number. Such a great power on such a small die was possible due to increased number of shader operating units. Core clock was set to 800MHz on first engineering samples, however it is now reported to level at 1100 MHz.

GK-104 tested on PhysX enabled applications had a performance of Radeon HD 7970, on the other hand, in games without PhysX support it was 10-20% slower. GK-104 is to handily trounce a 7970, however on no-PhysX games it will probably lose to Pitcairn.

Here is how S|K describe GK-104 performance:

All of the benchmark numbers shown by Nvidia, and later to SemiAccurate, were overwhelmingly positive. How overwhelmingly positive? Far faster than an AMD HD7970/Tahiti for a chip with far less die area and power use, and it blew an overclocked 580GTX out of the water by unbelievable margins. That is why we wrote this article.

SemiAccurate also focus on war between AMD and NVIDIA, pointing out how NVIDIA’s developers fix specific games by implementing special code, which actually disables some functionality on AMD cards.

by WhyCry

Previous Post
Matrox Releases Epica TC20+ and TC48 Graphics Cards for Wyse Thin Client
Next Post
AMD Preparing Radeon HD 7950 Eyefinity 6 Edition

Back to Top ↑