NVIDIA Tesla

Tesla是一個NVIDIA顯示核心系列品牌,主要用於伺服器高性能電腦運算,用於對抗AMDFireStream系列。這是继GeForceQuadro之后,第三个顯示核心商标。NVIDIA將顯示核心分為三大系列。GeForce用於提供家庭娛樂;Quadro用於專業繪圖設計;Tesla用於大規模的並聯電腦運算。

Tesla
634218 Badge Tesla 3D.jpg
2007年開始使用的Tesla標誌
研发者NVIDIA
生产日期2007年
类型圖形處理器

Tesla以發明家尼古拉·特斯拉的名字命名。

產品系列编辑

 
nVIDIA Tesla C870圖形處理器

目前,Tesla有三個系列:

  • Tesla GPU运算处理器 - 外形與普通顯示卡大致相同,C870採用GeForce 8顯示核心,而C1060採用GeForce 200顯示核心,不設任何顯示輸出。
  • Tesla GPU Deskside Supercomputer - 桌面平台用,外形與QuadroPlex相似,D870包含兩張C870运算处理器,可透過接線互联多個裝置。Tesla 10系列中沒有相關產品。
  • Tesla GPU Server - 服务器用,外形與1U伺服器相似,S870包含四張C870运算处理器,而S1070包含四張C1060运算处理器,可透過接線互联多個裝置。

較早的時候,人們已意識到GPU能運算大量數據。所以开发者通过图形语言,利用顯示核心,來进行并行计算,亦即是GPGPU(通用繪圖核心)。但开发者需要有一定程度的图形处理知識,才能發揮顯示核心效能。隨後,NVIDIA推出了CUDA。开发者利用C++语言,再通過CUDA编译器,就能利用顯核運算。开发者可忽略图形处理技術,而直接利用熟悉的C++语言。开发者和科学家,就可以利用顯示核心,研究物理生化勘探等領域。

Tesla比較專注於高性能運算,并且C1060以上(G200)系列能支援双精度浮点格式。另一方面,CUDA被所有的NVIDIA顯示核心支援,包括GeForce和Quadro系列。

將來,顯示核心能普及化地,輔助中央處理器,進行视频压缩数据库搜索等工作。並支援更多程式語言,例如FortranC++JAVAPython等。

完整型號列表编辑

Model 微架構 Launch Chips Core clock
(MHz)
Shaders Memory Processing power (每秒浮點運算次數)[a] CUDA
compute
ability[b]
热设计功耗
(watts)
Notes, form_factor
Cuda cores
(total)
Base clock (赫兹) Max boost
clock (赫兹)[c]
Bus type Bus width
(位元)
Size
(吉字节)
Clock
(Transfer (computing)英语Transfer (computing))
Bandwidth
(吉字节/s)
單精度浮點數
(MAD+MUL)
單精度浮點數
(MAD or 乘積累加運算)
雙精度浮點數
(乘積累加運算)
Units MHz MHz W
C870 GPU Computing Module[d] Tesla (microarchitecture)英语Tesla (microarchitecture) May 2, 2007 1× G80 600 128 1350 不適用 GDDR3 384 1.5 1600 76.8 518.4 345.6 1.0 170.9 Internal PCIe GPU (full-height, dual-slot)
D870 Deskside Computer[d] May 2, 2007 2× G80 600 256 1350 不適用 GDDR3 2× 384 2× 1.5 1600 2× 76.8 1036.8 691.2 1.0 520 Deskside or 3U 19-inch rack英语19-inch rack external GPUs
S870 GPU Computing Server[d] May 2, 2007 4× G80 600 512 1350 不適用 GDDR3 4× 384 4× 1.5 1600 4× 76.8 2073.6 1382.4 1.0 1U 19-inch rack英语19-inch rack external GPUs, connect via 2× PCIe (×16)
C1060 GPU Computing Module[e] April 9, 2009 1× GT200 602 240 1296[2] 不適用 GDDR3 512 4 1600 102.4 933.12 622.08 77.76 1.3 187.8 Internal PCIe GPU (full-height, dual-slot)
S1070 GPU Computing Server "400 configuration"[e] June 1, 2008 4× GT200 602 960 1296 不適用 GDDR3 4× 512 4× 4 1538.4 4× 98.5 3732.5 2488.3 311.0 1.3 800 1U 19-inch rack英语19-inch rack external GPUs, connect via 2× PCIe (×8 or ×16)
S1070 GPU Computing Server "500 configuration"[e] 1440 不適用 4147.2 2764.8 345.6
S1075 GPU Computing Server[e][3] June 1, 2008 4× GT200 602 960 1440 不適用 GDDR3 4× 512 4× 4 1538.4 4× 98.5 4147.2 2764.8 345.6 1.3 1U 19-inch rack英语19-inch rack external GPUs, connect via 1× PCIe (×8 or ×16)
Quadro Plex 2200 D2 Visual Computing System[f] 2× GT200GL 648 480 1296 不適用 GDDR3 2× 512 2× 4 1600 2× 102.4 1866.2 1244.2 155.5 1.3 Deskside or 3U 19-inch rack英语19-inch rack external GPUs with 4 dual-link DVI outputs
Quadro Plex 2200 S4 Visual Computing System[f] 4× GT200GL 648 960 1296 不適用 GDDR3 4× 512 4× 4 1600 4× 102.4 3732.5 2488.3 311.0 1.3 1200 1U 19-inch rack英语19-inch rack external GPUs, connect via 2× PCIe (×8 or ×16)
C2050 GPU Computing Module[4] Fermi (microarchitecture)英语Fermi (microarchitecture) July 25, 2011 1× GF100 575 448 1150 不適用 GDDR5 384 3[g] 3000 144 1030.4 515.2 2.0 247 Internal PCIe GPU (full-height, dual-slot)
M2050 GPU Computing Module[5] July 25, 2011 不適用 3092 148.4 225
C2070 GPU Computing Module[4] July 25, 2011 1× GF100 575 448 1150 不適用 GDDR5 384 6[g] 3000 144 1030.4 515.2 2.0 247 Internal PCIe GPU (full-height, dual-slot)
C2075 GPU Computing Module[6] July 25, 2011 不適用 3000 144 225
M2070/M2070Q GPU Computing Module[7] July 25, 2011 不適用 3132 150.336 225
M2090 GPU Computing Module[8] July 25, 2011 1× GF110 650 512 1300 不適用 GDDR5 384 6[g] 3700 177.6 1331.2 665.6 2.0 225 Internal PCIe GPU (full-height, dual-slot)
S2050 GPU Computing Server July 25, 2011 4× GF100 575 1792 1150 不適用 GDDR5 4× 384 4× 3[g] 3 4× 148.4 4121.6 2060.8 2.0 900 1U 19-inch rack英语19-inch rack external GPUs, connect via 2× PCIe (×8 or ×16)
S2070 GPU Computing Server 不適用 4× 6[g]
K10 GPU accelerator[9] Kepler (微架构) May 1, 2012 2× GK104 不適用 3072 745 ? GDDR5 2× 256 2× 4 5000 2× 160 4577 190.7 3.0 225 Internal PCIe GPU (full-height, dual-slot)
K20 GPU accelerator[10][11] November 12, 2012 1× GK110 不適用 2496 706 758 GDDR5 320 5 5200 208 3524 1175 3.5 225 Internal PCIe GPU (full-height, dual-slot)
K20X GPU accelerator[12] November 12, 2012 1× GK110 不適用 2688 732 ? GDDR5 384 6 5200 250 3935 1312 3.5 235 Internal PCIe GPU (full-height, dual-slot)
K40 GPU accelerator[13] October 8, 2013 1× GK110B 不適用 2880 745 875 GDDR5 384 12[g] 6000 288 4291–5040 1430–1680 3.5 235 Internal PCIe GPU (full-height, dual-slot)
K80 GPU accelerator[14] November 17, 2014 2× GK210 不適用 4992 560 875 GDDR5 2× 384 2× 12 5000 2× 240 5591–8736 1864–2912 3.7 300 Internal PCIe GPU (full-height, dual-slot)
M4 GPU accelerator[15][16] Maxwell (microarchitecture)英语Maxwell (microarchitecture) November 10, 2015 1× GM206 不適用 1024 872 1072 GDDR5 128 4 5500 88 1786–2195 55.81–68.61 5.2 50–75 Internal PCIe GPU (half-height, single-slot)
M6 GPU accelerator[17] August 30, 2015 1× GM204-995-A1 不適用 1536 722 1051 GDDR5 256 8 4600 147.2 2218–3229 69.3–100.9 5.2 75–100 Internal MXM GPU
M10 GPU accelerator[18] 4× GM107 不適用 2560 1033 ? GDDR5 4× 128 4× 8 5188 4× 83 5289 165.3 5.2 225 Internal PCIe GPU (full-height, dual-slot)
M40 GPU accelerator[16][19] November 10, 2015 1× GM200 不適用 3072 948 1114 GDDR5 384 12 6000 288 5825–6844 182.0–213.9 5.2 250 Internal PCIe GPU (full-height, dual-slot)
M60 GPU accelerator[20] August 30, 2015 2× GM204-895-A1 不適用 4096 899 1178 GDDR5 2× 256 2× 8 5000 2× 160 7365–9650 230.1–301.6 5.2 225–300 Internal PCIe GPU (full-height, dual-slot)
P4 GPU accelerator[21] 帕斯卡 (微架构) September 13, 2016 1× GP104 不適用 2560 810 1063 GDDR5 256 8 6000 192.0 4147–5443 129.6–170.1 6.1 50-75 PCI Express card
P6 GPU accelerator[22][23] March 24, 2017 1× GP104-995-A1 不適用 2048 1012 1506 GDDR5 256 16 3003 192.2 6169 192.8 6.1 90 行動PCI Express模組 card
P40 GPU accelerator[21] September 13, 2016 1× GP102 不適用 3840 1303 1531 GDDR5 384 24 7200 345.6 10007–11758 312.7–367.4 6.1 250 PCI Express card
P100 GPU accelerator (mezzanine)[24][25] April 5, 2016 1× GP100-890-A1 不適用 3584 1328 1480 高頻寬記憶體 4096 16 1430 732 9519–10609 4760–5304 6.0 300 NVLink card
P100 GPU accelerator (16 GB card)[26] June 20, 2016 1× GP100 不適用 1126 1303 8071‒9340 4036‒4670 250 PCI Express card
P100 GPU accelerator (12 GB card)[26] June 20, 2016 不適用 3072 12 549 8071‒9340 4036‒4670
V100 GPU accelerator (mezzanine)[27][28][29] 伏打微架構 1× GV100-895-A1 不適用 5120 未知 1455 HBM2 4096 16 or 32 1750 900 14899 7450 7.0 300 NVlink card
V100 GPU accelerator (PCIe card)[27][28][29] June 21, 2017 1× GV100 不適用 未知 1370 14028 7014 250 PCIe card
T4 GPU accelerator (PCIe card)[30][31] 图灵微架构 September 12, 2018 1× TU104-895-A1 不適用 2560 585 1590 GDDR6 256 16 未知 320 8100 未知 7.5 70 PCIe card
RTX A40[32] 安培微架构 2020年10月5日
Model 微架構 Launch Chips Core clock
(MHz)
Shaders Memory Processing power (每秒浮點運算次數)[a] CUDA
compute
ability
热设计功耗
(watts)
Notes, form factor
Cuda cores
(total)
Base clock (赫兹) Max boost
clock (赫兹)[c]
Bus type Bus width
(位元)
Size
(吉字节)
Clock
(Transfer (computing)英语Transfer (computing))
Bandwidth
(total)
(吉字节/s)
單精度浮點數
(MAD+MUL)
單精度浮點數
(MAD or 乘積累加運算)
雙精度浮點數
(乘積累加運算)

Notes

  1. ^ 1.0 1.1 To calculate the processing power see Tesla (microarchitecture)英语Tesla (microarchitecture), Fermi (microarchitecture)英语Fermi (microarchitecture), Kepler (微架构), Maxwell (microarchitecture)英语Maxwell (microarchitecture), or 帕斯卡 (微架构). A number range specifies the minimum and maximum processing power at, respectively, the base clock and maximum boost clock.
  2. ^ Core architecture version according to the CUDA programming guide.
  3. ^ 3.0 3.1 GPU Boost is a default feature that increases the core clock rate while remaining under the card's predetermined power budget. Multiple boost clocks are available, but this table lists the highest clock supported by each card.[1]
  4. ^ 4.0 4.1 4.2 Specifications not specified by Nvidia assumed to be based on the NVIDIA GeForce 8GTX
  5. ^ 5.0 5.1 5.2 5.3 Specifications not specified by Nvidia assumed to be based on the NVIDIA GeForce 200
  6. ^ 6.0 6.1 Specifications not specified by Nvidia assumed to be based on the Quadro FX 5800
  7. ^ 7.0 7.1 7.2 7.3 7.4 7.5 With ECC on, a portion of the dedicated memory is used for ECC bits, so the available user memory is reduced by 12.5%. (e.g. 4 GB total memory yields 3.5 GB of user available memory.)


另見编辑

參考資料编辑

  1. ^ Nvidia GPU Boost For Tesla (PDF). January 2014 [7 December 2015]. (原始内容存档 (PDF)于2020-05-16). 
  2. ^ Tesla C1060 Computing Processor Board (PDF). Nvidia.com. [2015-12-11]. (原始内容存档 (PDF)于2020-11-24). 
  3. ^ Difference between Tesla S1070 and S1075. 31 October 2008 [January 29, 2017]. (原始内容存档于2012-02-26). S1075 has one interface card 
  4. ^ 4.0 4.1 Tesla C2050 and Tesla C2070 Computing Processor (PDF). Nvidia.com. [2015-12-11]. (原始内容存档 (PDF)于2020-11-24). 
  5. ^ Tesla M2050 and Tesla M2070/M2070Q Dual-Slot Computing Processor Modules (PDF). Nvidia.com. [2015-12-11]. (原始内容存档 (PDF)于2020-11-24). 
  6. ^ Tesla C2075 Computing Processor Board (PDF). Nvidia.com. [2015-12-11]. (原始内容存档 (PDF)于2020-11-24). 
  7. ^ Hand, Randall. NVidia Tesla M2050 & M2070/M2070Q Specs OnlineVizWorld.com. VizWorld.com. 2010-08-23 [2015-12-11]. (原始内容存档于2020-08-17). 
  8. ^ Tesla M2090 Dual-Slot Computing Processor Module (PDF). Nvidia.com. [2015-12-11]. (原始内容存档 (PDF)于2020-10-26). 
  9. ^ Tesla K10 GPU accelerator (PDF). Nvidia.com. [2015-12-11]. (原始内容存档 (PDF)于2020-11-24). 
  10. ^ Tesla K20 GPU active accelerator (PDF). Nvidia.com. [2015-12-11]. (原始内容存档 (PDF)于2017-07-12). 
  11. ^ Tesla K20 GPU accelerator (PDF). Nvidia.com. [2015-12-11]. (原始内容存档 (PDF)于2020-03-26). 
  12. ^ Tesla K20X GPU accelerator (PDF). Nvidia.com. [2015-12-11]. (原始内容存档 (PDF)于2020-11-24). 
  13. ^ Tesla K40 GPU accelerator (PDF). Nvidia.com. [2015-12-11]. (原始内容存档 (PDF)于2020-06-17). 
  14. ^ Tesla K80 GPU accelerator (PDF). Images.nvidia.com. [2015-12-11]. (原始内容存档 (PDF)于2018-07-12). 
  15. ^ Nvidia Announces Tesla M40 & M4 Server Cards - Data Center Machine Learning. Anandtech.com. [2015-12-11]. (原始内容存档于2020-11-08). 
  16. ^ 16.0 16.1 Accelerating Hyperscale Datacenter Applications with Tesla GPUs | Parallel Forall. Devblogs.nvidia.com. 2015-11-10 [2015-12-11]. (原始内容存档于2017-07-09). 
  17. ^ Tesla M6 (PDF). Images.nvidia.com. [2016-05-28]. (原始内容存档 (PDF)于2020-10-22). 
  18. ^ Tesla M10 (PDF). Images.nvidia.com. [2016-10-29]. (原始内容存档 (PDF)于2017-05-10). 
  19. ^ Tesla M40 (PDF). Images.nvidia.com. [2015-12-11]. (原始内容存档 (PDF)于2016-10-21). 
  20. ^ Tesla M60 (PDF). Images.nvidia.com. [2016-05-27]. (原始内容存档 (PDF)于2020-10-22). 
  21. ^ 21.0 21.1 Smith, Ryan. Nvidia Announces Tesla P40 & Tesla P4 - Network Inference, Big & Small. Anandtech. 13 September 2016 [13 September 2016]. (原始内容存档于2021-01-16). 
  22. ^ Tesla P6 (PDF). www.nvidia.com. [2019-03-07]. (原始内容存档 (PDF)于2020-07-28). 
  23. ^ Tesla P6 Specs. www.techpowerup.com. [2019-03-07]. (原始内容存档于2020-07-28). 
  24. ^ Smith, Ryan. Nvidia Announces Tesla P100 Accelerator - Pascal GP100 for HPC. Anandtech.com. Anandtech.com. 5 April 2016 [5 April 2016]. (原始内容存档于2016-07-30). 
  25. ^ Harris, Mark. Inside Pascal: Nvidia’s Newest Computing Platform. [13 September 2016]. (原始内容存档于2017-05-07). 
  26. ^ 26.0 26.1 Smith, Ryan. NVidia Announces PCI Express Tesla P100. Anandtech.com. 20 June 2016 [21 June 2016]. (原始内容存档于2020-11-11). 
  27. ^ 27.0 27.1 Smith, Ryan. The Nvidia GPU Technology Conference 2017 Keynote Live Blog. Anandtech. 10 May 2017 [10 May 2017]. (原始内容存档于2020-11-12). 
  28. ^ 28.0 28.1 Smith, Ryan. NVIDIA Volta Unveiled: GV100 GPU and Tesla V100 Accelerator Announced. Anandtech. 10 May 2017 [10 May 2017]. (原始内容存档于2020-11-12). 
  29. ^ 29.0 29.1 Oh, Nate. NVIDIA Formally Announces V100: Available later this Year. Anandtech.com. 20 June 2017 [20 June 2017]. (原始内容存档于2021-01-24). 
  30. ^ NVIDIA TESLA T4 TENSOR CORE GPU. NVIDIA. [17 October 2018]. (原始内容存档于2021-02-04). 
  31. ^ NVIDIA Tesla T4 Tensor Core Product Brief (PDF). www.nvidia.com. [2019-07-10]. (原始内容存档 (PDF)于2020-06-17). 
  32. ^ NVIDIA发布RTX A6000/A40:满血10752个CUDA核心、48GB显存. 

外部連結编辑