NV44
First off, I'll lay out NV44 vs NV43 in a handy table, to get the basics down.| NV44 | NV43 | |
| Process | 110nm @ TSMC | 110nm @ TSMC |
| Transistor Count | 75M | 143M |
| Geometry Pipeline | VS3.0 | VS3.0 |
| Fragment Processor | PS3.0 | PS3.0 |
| Fragment Processor Setup | 2 full ALU (not equal) each with 1 mini ALU, Fog ALU, per pipe | 2 full ALU (not equal) each with 1 mini ALU, Fog ALU, per pipe |
| Fragment Processor Precision | FP32, FP16 | FP32, FP16 |
| Traditional Render Setup | 4 x 1 | 8 x 1 |
| ROPs per quad | 2 | 2 |
| Vertex Shaders | 3 | 3 |
| Basic Texture Filtering | Bilinear | Bilinear |
| Texture Filtering | Bilinear, Trilinear, 16X Anisotropic | Bilinear, Trilinear, 16X Anisotropic |
| Antialiasing | Multi-sampling and super-sampling | Multi-sampling and super-sampling |
| AA Sample Type | Rotated grid up to 8X with supersampling combined at 8X | Rotated grid up to 8X with supersampling combined at 8X |
| Native Bus Support | PEG16X | PEG16X |
| Memory support | GDDR3, DDR | GDDR3 |
| Basic Core Frequency | 350MHz | 500MHz |
| Basic Memory Frequency | 700MHz | 1000MHz |
| Memory Bus Width | 64-bit, memory crossbar, MMU | 128-bit, memory crossbar |
| Basic Pixel Fillrate | 1400Mpixel/sec | 4000Mpixel/sec |
| Basic Multitexture Fillrate | 1400Mtexel/sec | 4000Mtexel/sec |
| Basic Memory Bandwidth | ~5.6GB/sec | ~16.0GB/sec |
There's a lot to talk about with NV44, all relating to its memory setup. The basic core features of the GPU are identical to NV43 and NV40. Three NV4x vertex units feeding four NV4x fragment units, feeding two ROPs (the same balanced setup seen in NV43). 6200's launch on NV43 said that colour and depth buffer compression were disabled for that SKU. The same is correct for NV44. NV44 has Shader Model 3.0 support like NV43, but like the crippled NV43 version of 6200, it's without support for 64-bit floating point buffer formats. So 16bpc FP rendering is impossible, the blend unit not in the ROP again for NV44. You can verify by checking the buffer formats exposed by NV44's initial driver's DX capabilities list.
NVIDIA's UltraShadow II is in and working, allowing (admittedly bandwidth crippled) accleration of rendering techniques that make use of a Z-only pass for depth/shadow rendering, such as the famous Doom 3.
The GPU itself is tiny, smaller in dimensions than the DRAMs that accompany it on the PCB. 10mm by 11mm, it's some 45% smaller than NV43.
Click for a larger version
Overall, NV44 is a balancing act in terms of cost and die space and that's massively apparent in NVIDIA's approach to NV44's memory subsystem. Let's look at that in more detail.
