Engineering:GeForce 400 series

From HandWiki
Short description: Series of GPUs by Nvidia
GeForce 400 series
GTX480.jpg
The GeForce GTX 480, released in 2010 as the flagship unit of the 400 series
Release dateApril 12, 2010; 14 years ago (April 12, 2010)
CodenameGF10x
ArchitectureFermi
ModelsGeForce series
  • GeForce GT series
  • GeForce GTS series
  • GeForce GTX series
Transistors260M 40 nm (GT218)
  • 585M 40 nm (GF108)
  • 1.170M 40 nm (GF106)
  • 1.950M 40 nm (GF104)
  • 1.950M 40 nm (GF114)
  • 3.200M 40 nm (GF100)
Cards
Entry-levelGT 420
GT 430
Mid-rangeGT 440
GTS 450
GTX 460
GTX 465
High-endGTX 470
GTX 480
API support
Direct3DDirect3D 12.0 (feature level 11_0)[1]
OpenCLOpenCL 1.1
OpenGLOpenGL 4.6
History
PredecessorGeForce 200 series
SuccessorGeForce 500 series

Serving as the introduction of Fermi, the GeForce 400 series is a series of graphics processing units developed by Nvidia. Its release was originally slated in November 2009;[2] however, after delays, it was released on March 26, 2010 with availability following in April 2010.

Architecture

Main page: Engineering:Fermi (microarchitecture)

Nvidia described the Fermi microarchitecture as the next major step in its line of GPUs following the Tesla microarchitecture used since the G80. The GF100, the first Fermi-architecture product, is large: 512 stream processors, in sixteen groups of 32, and 3.0 billion transistors, manufactured by TSMC in a 40 nm process. It is Nvidia's first chip to support OpenGL 4.0 and Direct3D 11. No products with a fully enabled GF100 GPU were ever sold. The GTX 480 had one streaming multiprocessor disabled. The GTX 470 had two streaming multiprocessors and one memory controller disabled. The GTX 465 had five streaming multiprocessors and two memory controllers disabled. Consumer GeForce cards came with 256MB attached to each of the enabled GDDR5 memory controllers, for a total of 1.5, 1.25 or 1.0GB; the Tesla C2050 had 512MB on each of six controllers, and the Tesla C2070 had 1024MB per controller. Both the Tesla cards had fourteen active groups of stream processors.

The chips found in the high performance Tesla branding feature memory with optional ECC and the ability to perform one double-precision floating-point operation per cycle per core; the consumer GeForce cards are artificially driver restricted to one DP operation per four cycles. With these features, combined with support for Visual Studio and C++, Nvidia targeted professional and commercial markets, as well as use in high performance computing.

Fermi is named after Italian physicist Enrico Fermi.

Current limitations and trade-offs

The quantity of on-board SRAM per ALU actually decreased proportionally compared to the previous G200 generation, despite the increase of the L2 cache from 256kB per 240 ALUs to 768kB per 512 ALUs, since Fermi has only 32768 registers per 32 ALUs (vs. 16384 per 8 ALUs), only 48kB of shared memory per 32 ALUs (vs. 16kB per 8 ALUs), and only 16kB of cache per 32 ALUs (vs. 8kB constant cache per 8 ALUs + 24kB texture cache per 24 ALUs). Parameters such as the number of registers can be found in the CUDA Compute Capability Comparison Table in the reference manual.[3]

History

On September 30, 2009, Nvidia released a white paper describing the architecture:[4] the chip features 16 'Streaming Multiprocessors' each with 32 'CUDA Cores' capable of one single-precision operation per cycle or one double-precision operation every other cycle, a 40-bit virtual address space which allows the host's memory to be mapped into the chip's address space, meaning that there is only one kind of pointer and making C++ support significantly easier, and a 384-bit wide GDDR5 memory interface. As with the G80 and GT200, threads are scheduled in 'warps', sets of 32 threads each running on a single shader core. While the GT200 had 16 KB 'shared memory' associated with each shader cluster, and required data to be read through the texturing units if a cache was needed, GF100 has 64 KB of memory associated with each cluster, which can be used either as a 48 KB cache plus 16 KB of shared memory, or as a 16 KB cache plus 48 KB of shared memory, along with a 768 KB L2 cache shared by all 16 clusters.

The white paper describes the chip much more as a general purpose processor for workloads encompassing tens of thousands of threads - reminiscent of the Tera MTA architecture, though without that machine's support for very efficient random memory access - than as a graphics processor.

Many users reported high temperatures and power consumption while receiving correspondingly poor performance improves in the GeForce 400 series Fermi GPUs when compared to rival competitor AMD's Radeon HD 5000 Series - leading AMD to create and release a promotional video "The Misunderstanding"[5] to poke fun at the issue. In the video, a police unit is seen commencing a raid on a house with a large thermal profile, indicating a grow operation. However, upon entering the home it is apparent that the source of the high temperature is a Fermi GPU.[6][7] It became a common joke that one could fry an egg on a Fermi GPU at full load.[8]

Products

  • 1 SPs - Shader Processors - Unified Shaders: Texture mapping units: Render output units
  • 2 Each Streaming Multiprocessor (SM) in the GPU of GF100 architecture contains 32 SPs and 4 SFUs. Each Streaming Multiprocessor (SM) in the GPU of GF104/106/108 architecture contains 48 SPs and 8 SFUs. Each SP can fulfil 2 single precision fused multiply–add (FMA) operations per cycle. Each SFU can fulfil four SF operations per cycle. One FMA operation counts for two floating point operations. So the theoretical single precision peak performance, with shader count [n] and shader frequency [f, GHz], can be estimated by the following, FLOPSsp ≈ f × n × 2 (FMA). Total Processing Power: for GF100 FLOPSsp ≈ f × m ×(32 SPs × 2(FMA) + 4 × 4 SFUs) and for GF104/106/108 FLOPSsp ≈ f × m × (48 SPs × 2(FMA) + 4 × 8 SFUs) or for GF100 FLOPSsp ≈ f × n × 2.5 and for GF104/106/108 FLOPSsp ≈ f × n × 8 / 3.[9]

SP - Shader Processor (Unified Shader, CUDA Core), SFU - Special Function Unit, SM - Streaming Multiprocessor.

  • 3 Each SM in the GF100 contains 4 texture filtering units for every texture address unit. The complete GF100 die contains 64 texture address units and 256 texture filtering units[10] Each SM in the GF104/106/108 architecture contains 8 texture filtering units for every texture address unit. The complete GF104 die contains 64 texture address units and 512 texture filtering units, the complete GF106 die contains 32 texture address units and 256 texture filtering units and the complete GF108 die contains 16 texture address units and 128 texture filtering units.[11]

All products are produced on a 40 nm fabrication process. All products support Direct X 12.0, OpenGL 4.6 and OpenCL 1.1. The only exception is the Geforce 405 which is based on the GT218 core only supporting DirectX 10.1, OpenGL 3.3 and no OpenCL Support

Model Launch Code name Transistors (million) Die size (mm2) Bus interface SM count Core config1,3 Clock rate Fillrate Memory configuration GFLOPS (FMA)2 TDP (watts) Launch price (USD)
Core (MHz) Shader (MHz) Memory (MHz) Pixel (GP/s) Texture (GT/s) Size (MB) Bandwidth (GB/s) DRAM type Bus width (bit)
GeForce 405 (OEM) September 16, 2011 GT218 260 57 PCIe 2.0 x16 1 16:8:4 589 1402 1580 2.4 4.7 512
1024
12.6 DDR3 64 44.9 25 OEM
GeForce GT 420 (OEM) September 3, 2010 GF108 585 116 PCIe 2.0 x16 1 48:8:4 700 1400 1800 2.8 5.6 2048 28.8 GDDR3 128 134.4 50 OEM
GeForce GT 430 (OEM) October 11, 2010 GF108 585 116 PCIe 2.0 x16 2 96:16:4 700 1400 1600
1800
2.8 11.2 2048 25.6
28.8
GDDR3 128 268.8 60 OEM
GeForce GT 430 October 11, 2010 GF108 585 116 PCIe 2.0 x16 2 96:16:4 700 1400 1800 2.8 11.2 1024 28.8 GDDR3 128 268.8 49 $79
GeForce GT 440 February 1, 2011 GF108 585 116 PCIe 2.0 x16 2 96:16:4 810 1620 1800
3200
3.24 13.2 512
1024
2048
28.8
51.2
GDDR3
GDDR5
128 311 65 $79
GeForce GT 440 (OEM) October 11, 2010 GF106 1170 238 PCIe 2.0 x16 3 144:24:24 594 1189 1800 14.26 14.26 1536
3072
43.2 GDDR3 192 342.4 56 OEM
GeForce GTS 450 (OEM) October 11, 2010 GF106 1170 238 PCIe 2.0 x16 3 144:24:24 790 1580 1804 18.96 18.96 1024
1536
86 GDDR5 192 455 106 OEM
GeForce GTS 450 September 13, 2010 GF106 1170 238 PCIe 2.0 x16 4 192:32:16 783 1566 1804 12.53 25.06 512
1024
2048
57.73 GDDR3
GDDR5
128 601.3 106 $129
GeForce GTX 460 SE November 15, 2010 GF104 1950 332 PCIe 2.0 x16 6 288:48:32 650 1300 3400 20.8 31.2 1024 108.8 GDDR5 256 748.8 150 $160?-$180?
GeForce GTX 460 (OEM) October 11, 2010 GF104 1950 332 PCIe 2.0 x16 7 336:56:24 650 1300 3400 20.8 36.4 1024 108.8 GDDR5 256 873.6 150 OEM
GeForce GTX 460 July 12, 2010 GF104 1950 332 PCIe 2.0 x16 7 336:56:24 675 1350 3600 16.2 37.8 768 86.4 GDDR5 192 907.2 150 $199
336:56:32 21.6 1024
2048
115.2 256 160 $229
GeForce GTX 460 v2 September 24, 2011 GF114 1950 332 PCIe 2.0 x16 7 336:56:24 778 1556 4008 18.67 43.57 1024 96.2 GDDR5 192 1045.6 160 $199
GeForce GTX 465 May 31, 2010 GF100 3200 529 PCIe 2.0 x16 11 352:44:32 607 1215 3206 19.42 26.71 1024 102.6 GDDR5 256 855.4 200 $279
GeForce GTX 470 March 26, 2010 GF100 3200 529 PCIe 2.0 x16 14 448:56:40 607 1215 3348 24.28 34 1280 133.9 GDDR5 320 1088.6 215 $349
GeForce GTX 480 March 26, 2010 GF100 3200 529 PCIe 2.0 x16 15 480:60:48 700 1401 3696 33.60 42 1536 177.4 GDDR5 384 1345 250 $499

On November 8, 2010, Nvidia released the GF110 chip, along with the GTX 580 (480's replacement). It is a redesigned GF100 chip, which uses significantly less power. This allowed Nvidia to enable all 16 SMs (all 16 cores), which was previously impossible on the GF100 "Nvidia GeForce GTX 580". http://www.nvidia.com/object/product-geforce-gtx-580-us.html.  Various features of the GF100 architecture were only available on the more expensive Quadro and Tesla series of cards.[12] For the GeForce consumer products, double precision performance is a quarter of that of the "full" Fermi architecture. Error checking and correcting memory (ECC) also does not operate on consumer cards.[13] The GF100 cards provide Compute Capability 2.0, while the GF104/106/108 cards provide Compute Capability 2.1.

Chipset table

Discontinued support

Nvidia announced that after Release 390 drivers, it will no longer release 32-bit drivers for 32-bit operating systems.[14]

Nvidia announced in April 2018 that Fermi will move to legacy driver support status and be maintained until January 2019.[15]

See also

Notes

References

  1. Killian, Zak (July 3, 2017). "Nvidia finally lets Fermi GPU owners enjoy DirectX 12". Tech Report. https://techreport.com/news/32191/nvidia-finally-lets-fermi-gpu-owners-enjoy-directx-12. 
  2. "OFFICIAL: NVIDIA says GT300 on schedule for Q4 2009, yields are fine - Bright Side Of News*". Brightsideofnews.com. http://www.brightsideofnews.com/news/2009/9/25/nvidia-says-gt300-on-schedule-for-q4-20092c-yields-are-fine.aspx. 
  3. Compute Capability Comparison Table in "Page 147-148, Appendix G.1, CUDA 3.1 official reference manual". http://developer.download.nvidia.com/compute/cuda/3_1/toolkit/docs/NVIDIA_CUDA_C_ProgrammingGuide_3.1.pdf. . Page 97 in Appendix A lists the older NVIDIA GPUs and shows all G200 series to be compute capability 1.3, while Fermi-based cards have compute capability 2.x (page 14, Section 2.5).
  4. http://www.nvidia.com/content/PDF/fermi_white_papers/NVIDIA_Fermi_Compute_Architecture_Whitepaper.pdf
  5. https://www.youtube.com/watch?v=2QkyfGJgcwQ
  6. https://amp.hothardware.com/news/amd-pokes-fun-of-nvidias-fermi-gpu-heat-output-in-the-misunderstanding-video
  7. https://www.zdnet.com/article/nvidia-fermi-gf100-gpus-too-little-too-late-too-hot-and-too-expensive/
  8. https://www.tomshardware.co.uk/gf100-fermi-egg-frying-gtx-480,news-33106.html
  9. siliconmadness.com (2010). "Nvidia Announces Tesla 20 Series". http://www.siliconmadness.com/2009/11/nvidia-announces-tesla-20-series.html. 
  10. NVIDIA's GeForce GTX 480 and GTX 470: 6 Months Late, Was It Worth the Wait?
  11. NVIDIA’s GeForce GTX 460: The $200 King
  12. "Statement by NVIDIA on their General CUDA GPU Computing Discussion forum". http://forums.nvidia.com/index.php?showtopic=165055. 
  13. "NVIDIA Tesla C2xxx webpage". http://www.nvidia.com/object/product_tesla_C2050_C2070_us.html. , note from the description one may infer that on Teslas, ECC may be switched on and off using 1/8 of existing on-board memory, unlike standard ECC memory modules which requires 1/8 extra memory chips (that is, one extra chip to be mounted on the printed circuit board for every 8).
  14. http://nvidia.custhelp.com/app/answers/detail/a_id/4604/
  15. http://nvidia.custhelp.com/app/answers/detail/a_id/4654

External links