NVIDIA GT300 'Fermi' Specifications Detailed

More

As most of you already know, the GT300 was first unveiled months ago along with some preliminary specifications. Today we get the bigger picture, thanks to bsn. NVIDIA's current next-gen GPU is known as the GT300 or N70. It is now known that the GPU architecture is called "Fermi" by NVIDIA engineers. A quick history lesson: the man behind the name is an Italian physicist, Enrico Fermi who is known for helping develop the first nuclear reactor. This may be why the GT300 board is also known as "reactor". Now, to the important details.

GT300 "Fermi" Architecture GPU Specifications

• 3.0 billion transistors
• 40nm GPU by TSMC
• 384-bit memory interface (6x64-bit memory controllers)
• 512 shader cores (renamed to CUDA Cores)
• 32 CUDA cores per shader cluster
• 1MB L1 cache memory [divided into 16KB Cache - Shared Memory]
• 768KB L2 unified cache memory
• Up to 6GB GDDR5 memory (1.5GB for GeForce and up to 6GB for Quadro/Tesla)
• Half Speed IEEE 754 Double Precision
• 16 Streaming Multiprocessors (new name for the former Shader Cluster) containing 32 cores each

The Fermi architecture also supports C (CUDA), C++, DirectCompute, Fortran, OpenCL, OpenGL 3.1 and OpenGL 3.2. This means that for the first time in history a GPU can run C++ code! NVIDIA's "Fermi" architecture will be released in three tiers, GeForce for high-end consumers, Quadro for commersial use and Tesla for scientific purposes. There is still no word on pricing or availablility, but I'd expect the new Fermi cards to break the bank a bit.

These revelations are huge for NVIDIA and the enthusiast community, now we need to see one in action!

• Email Link
• Print News

nickosha on September 30, 2009 03:50PM I hope it's still on track to release this year. I can't even begin to imagine what the price might be though. reply to this

tkrow21 on September 30, 2009 03:55PM O.O reply to this

XxHellxRaizerxX on September 30, 2009 06:56PM Finally something original o.O I'm actually impressed. And Cache on a GPU? Looks like they paid attention to the Xbox 360 and Gamecube. reply to this

Zertz on September 30, 2009 07:29PM There always was cache on GPUs just like on CPUs, it's just not widely documented. Without cache, accessing data would be extremely slow. reply to this

Waco on October 1, 2009 12:46AM Not an L2 cache there wasn't... They've included one because they've switched from SIMD to MIMD which should increase efficiency by a huge amount but requires faster access to data/instructions (and more of both). reply to this

XxHellxRaizerxX on September 30, 2009 07:55PM Oh, guess that makes sense just thought the GPU was fine with the higher bandwidth of it's memory data path. But since they are listing it is this a jump from the normal amount placed in a GPU? reply to this

_TheAlexO on September 30, 2009 08:22PM Ive always wondered that as well. They have never listed Cache i just assumed it didnt need it reply to this

Zertz on September 30, 2009 09:09PM It's a bump from GT200, that's for sure. ATI doesn't release those details, but judging strictly by transistor count, the NVIDIA cards probably have more cache. The big features I'm seeing there are mostly oriented towards computing. "Half Speed IEEE 754 Double Precision", ATI does double precision at 1/5 the speed of single, so NVIDIA's 1/2 speed is a huge step. Support for C++ is also a huge step for GPGPU computing. reply to this

Zertz on September 30, 2009 09:13PM We are still months from a GT300 launch though, so while GT300 will almost certainly be faster than RV870, ATI (for the first time since... Radeon 9xxx?) got a lengthy head start. reply to this

Yogie P. on October 1, 2009 01:01AM This is a revenge from N Vidia. reply to this

Ersan on October 14, 2009 03:20PM this card is going to own the ati card by far...6gb GDDR5 at 384bit the 3gb and the 1.5gb versions will be higher bit so id reckon 3gb would be 512bit and 1.5 would be 1gbit 3BILLION transistors lmoa L::D:DD: reply to this

This news has comment postings disabled because it is now archived.