The official GT300/Fermi Thread

NVIDIA GT300 ''Fermi'' Detailed
NVIDIA's upcoming flagship graphics processor is going by a lot of codenames. While some call it the GF100, others GT300 (based on the present nomenclature), what is certain that the NVIDIA has given the architecture an internal name of "Fermi", after the Italian physicist Enrico Fermi, the inventor of the nuclear reactor. It doesn't come as a surprise, that the codename of the board itself is going to be called "reactor", according to some sources.

Based on information gathered so far about GT300/Fermi, here's what's packed into it:

* Transistor count of over 3 billion
* Built on the 40 nm TSMC process
* 512 shader processors (which NVIDIA may refer to as "CUDA cores")
* 32 cores per core cluster
* 384-bit GDDR5 memory interface
* 1 MB L1 cache memory, 768 KB L2 unified cache memory
* Up to 6 GB of total memory, 1.5 GB can be expected for the consumer graphics variant
* Half Speed IEEE 754 Double Precision floating point
* Native support for execution of C (CUDA), C++, Fortran, support for DirectCompute 11, DirectX 11, OpenGL 3.1, and OpenCL

read somewhere its not going to have hardware tesselation engine(s/w emulation instead), nvidia is behind again..

Quote:

---

Originally Posted by netkas

read somewhere its not going to have hardware tesselation engine(s/w emulation instead), nvidia is behind again..

---

You read wrong, it's required as part of DX11. :ROTF:

Quote:

---

Originally Posted by Helloworld_98

anyone else noticed it doesn't say that is has DX11 support in those specs?

---

Yep, not a peep about anything graphics related. Bad sign? Maybe.

Quote:

---

Originally Posted by netkas

read somewhere its not going to have hardware tesselation engine(s/w emulation instead), nvidia is behind again..

---

I presume you have extensive experience with tessellation and have good reasons for implying that a software implementation will be worse than fixed function hardware? I guess the fact that some of the leaders in tessellation research work for Nvidia doesn't matter.

"CoWA cores" sounds better. (Can of Whoop-Ass cores)

lol. Can of whoop-ass cores. That's great. "COWA Cores"

Quote:

---

Originally Posted by Piotrsama

"CoWA cores" sounds better. (Can of Whoop-Ass cores)

---

Haha. Could also call them CoWA-bunga cores.

Sign me up!

Quote:

---

Originally Posted by DilTech

You read wrong, it's required as part of DX11. :ROTF:

---

DX11 requires tesselation engine, dx11 specs doesnt specify if it has to be hardware or software (using "cuda"-cores)

Looks like a beast card if true. Now all that matters is pricing.

Quote:

---

Originally Posted by netkas

DX11 requires tesselation engine, dx11 specs doesnt specify if it has to be hardware or software (using "cuda"-cores)

---

If hardware isn't required, then you know as well as I do there's a good chance the developers won't code the games with ATi's tesselator in mind, right?

We'll know soon enough.

This is the final nail in Larrabee's coffin.

Quote:

---

Originally Posted by Cybercat

This is the final nail in Larrabee's coffin.

---

I doubt it, if the GT300 uses 512 SP's, they will be SIMD unless the die size is even bigger than rumours say and even then raytraced graphics and GPGPU work will be more optimized for larrabee than GT300 due to the fact that it's Intel and x86 is far more wide spread than nvidia's architecture.

Quote:

---

Originally Posted by Helloworld_98

anyone else noticed it doesn't say that is has DX11 support in those specs?

---

If you honestly believe nvidia will release their next gen GPU without support for DX11... :rofl:

Quote:

---

Originally Posted by Helloworld_98

even then raytraced graphics and GPGPU work will be more optimized for larrabee than GT300 due to the fact that it's Intel and x86 is far more wide spread than nvidia's architecture.

---

Apparently, Larrabee's vector cores are NOT x86.

this can't be good for gamers if it's true

Quote:

---

Our appetite, however, has been whetted by what an NVIDIA bigwig had to say recently, commenting on the upcoming GT300 graphics processor as "more like a CPU than a GPU."

---

Quote:

---

Originally Posted by Helloworld_98

I doubt it, if the GT300 uses 512 SP's, they will be SIMD unless the die size is even bigger than rumours say and even then raytraced graphics and GPGPU work will be more optimized for larrabee than GT300 due to the fact that it's Intel and x86 is far more wide spread than nvidia's architecture.

---

Aaaargh!! Larrabee's vector units are not x86. Repeat it 100x until it sticks.

Quote:

---

Originally Posted by Helloworld_98

I doubt it, if the GT300 uses 512 SP's, they will be SIMD unless the die size is even bigger than rumours say and even then raytraced graphics and GPGPU work will be more optimized for larrabee than GT300 due to the fact that it's Intel and x86 is far more wide spread than nvidia's architecture.

---

GT300 has MIMD cores which work in MPMD mode. :comp10:

This could be great for the high-end market like what NV intended for the GTX 260 and up (even though GTX 260 ended up plummeting in price to compete in the market) but how is it going to scale to smaller chips for the midrange and mainstream markets that really matter most? Last I knew NV was releasing mere DX 10.1 chips (GT21x) for the those markets. That's honestly where I'd be looking these days with little need for uber graphics power and ATi's Juniper seems more appealing. Damn you NV and your not doing full line refreshes. It worked fine with the undying G92 which remained competitive but I don't see how they'll fill the $150-200 market without a cut-down (not just disabled parts) GT300 chip.

3 billion transistors huh, i wonder how much longer it will be against the 5870...

^ to see who physically has the larger e-peen?

Quote:

---

Originally Posted by Bobbylite

^ to see who physically has the larger e-peen?

---

lol...

EVGA GTX 380 SuperEnlarged 3P33N edition !

I can already see it coming...

http://www.nvidia.com/object/fermi_architecture.html

512 CUDA cores - Confirmed.

Quote:

---

Originally Posted by ethomaz

http://www.nvidia.com/object/fermi_architecture.html

512 CUDA cores - Confirmed.

---

Gt300 is going to be a monster, looks like I'll have to return these HD5870s :p:

http://i38.tinypic.com/4j9oxj.jpg