GeForce 10 series
![]() ![]() The GTX 1070 Founders Edition reference card. | |
Release date | May 2016 |
---|---|
Codename | GP10x |
Architecture | Pascal |
Models | GeForce GTX Series |
Transistors |
|
Cards | |
Entry-level | GeForce GTX 1050 GeForce GTX 1050 Ti |
Mid-range | GeForce GTX 1060 |
High-end | GeForce GTX 1070 GeForce GTX 1080 |
Enthusiast | NVIDIA Titan X |
API support | |
DirectX | Direct3D 12.0 (feature level 12_1) |
OpenCL | OpenCL 1.2 |
OpenGL | OpenGL 4.5 |
Vulkan | Vulkan 1.0 SPIR-V |
History | |
Predecessor | GeForce 900 series |
GeForce GTX 10 Series is a series of graphics processing units developed by Nvidia as a successor to the GeForce 900 Series. The Pascal microarchitecture is the successor to the Maxwell microarchitecture and incorporates TSMC's 16 nm FinFET technology. [1] It also incorporates Samsung's 14 nm FinFET technology for Nvidia's GP107 chips.[2]
Architecture
The microarchitecture of the GeForce 10 series is named Pascal, after the 17th century French mathematician Blaise Pascal, and was presented on May 6, 2016.[3]
Nvidia has announced that the Pascal GP100 GPU will feature four High Bandwidth Memory stacks, allowing a total of 16 GB HBM2 on the highest-end models,[4] 16 nm technology,[1] Unified Memory and NVLink.[5]
New Features in GP10x:
- CUDA Compute Capability 6.0 (GP100 only), 6.1 (GP102, GP104, GP106, GP107)
- DisplayPort 1.4
- HDMI 2.0b
- Fourth generation Delta Color Compression
- PureVideo Feature Set H hardware video decoding HEVC Main10 (10bit), Main12 (12bit) & VP9 hardware decoding (GM200 & GM204 did not support HEVC Main10/Main12 & VP9 hardware decoding)[6]
- HDCP 2.2 support for 4K DRM protected content playback & streaming (Maxwell GM200 & GM204 lack HDCP 2.2 support, GM206 supports HDCP 2.2)[7]
- NVENC HEVC Main10 10bit hardware encoding
- GPU Boost 3.0
- Simultaneous Multi-Projection
- HB SLI Bridge Technology
- New memory controller with GDDR5X & GDDR5 support (GP102, GP104) [8]
- Dynamic load balancing scheduling system.[9] This allows the scheduler to dynamically adjust the amount of the GPU assigned to multiple tasks, ensuring that the GPU remains saturated with work except when there is no more work that can safely be distributed.[9] [9]
- Instruction-level preemption.[10] In graphics tasks, the driver restricts this to pixel-level preemption because pixel tasks typically finish quickly and the overhead costs of doing pixel-level preemption are much lower than performing instruction-level preemption.[10] Compute tasks get either thread-level or instruction-level preemption.[10] Instruction-level preemption is useful because compute tasks can take long times to finish and there are no guarantees on when a compute task finishes, so the driver enables the very expensive instruction-level preemption for these tasks.[10]
- Triple buffering implemented in the driver level.[11] Nvidia calls this "Fast Sync".[11] This has the GPU maintain three frame buffers per monitor.[11] This results in the GPU continuously rendering frames, and the most recently completely rendered frame is sent to a monitor each time it needs one.[11] This removes the initial delay that double buffering with vsync causes and disallows tearing.[11] The costs are that more memory is consumed for the buffers and that the GPU will consume power drawing frames that might be wasted because two or more frames could possibly be drawn between the time a monitor is sent a frame and the time the same monitor needs to be sent another frame.[11] In this case, the latest frame is picked, causing frames drawn after the previously displayed frame but before the frame that is picked to be completely wasted.[11] This feature has been backported to Maxwell-based GPUs in driver version 372.70.[12]
Successor architecture
Volta
After Pascal, the next architecture will be codenamed Volta, after the 18th century Italian physicist Alessandro Volta. Volta was to be the direct successor to Maxwell but in 2014 Nvidia announced that Pascal was following Maxwell "to take advantage of stacked memory and other innovations sooner."[13]
Products
Founders Edition
Announcing the GeForce 10-series products, Nvidia has introduced Founders Edition graphics cards versions both of GTX 1070 and 1080. These are what was previously known as reference cards, that is which were designed and built by Nvidia and not by its authorized board partners. The Founders Edition cards have a die cast machine-finished aluminum body with a single radial fan and a vapor chamber cooling (GTX 1080 only[14]), an upgraded power supply and a new low profile backplate.[15] Nvidia also released a limited supply of Founders Edition cards for the GTX 1060 that were only available directly from Nvidia's website.[16] Founders Edition cards prices are greater than MSRP of partners cards, however some partners' cards, incorporating a complex design, with liquid or hybrid cooling may cost even more than Founders Edition.
GeForce 10 (10xx) series
- Supported display standards are: DP 1.3/1.4, HDMI 2.0b, Dual link-DVI[17]
- Supported APIs are: Direct3D 12.0 (12_1), OpenGL 4.5, OpenCL 1.2 and Vulkan 1.0
Model | Launch | Code name | Fab (nm) | Transistors (billion) | Die size (mm2) | Bus interface | Core config[a] | Clock speeds | Fillrate | Memory | Processing power (GFLOPS)[b] | TDP (watts) | SLI HB support[c] | Release price (USD) | |||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Base core clock (MHz) | Boost core clock (MHz) | Memory (MT/s) | Pixel (GP/s)[d] | Texture (GT/s)[e] | Size (GiB) | Bandwidth (GB/s) | Bus type | Bus width (bit) | Single precision (Boost) | Double precision (Boost) | Half precision (boost)[18] | MSRP | Founders Edition | ||||||||||
GeForce GTX 1050[19] | October 25, 2016 | GP107-300 | 14 | 3.3 | 135 | PCIe 3.0 x16 | 640:40:32 | 1354 | 1455 | 7000 | 43.3 | 54.2 | 2 | 112 | GDDR5 | 128 | 1733 (1862) | 54 (58) | 27 (29) | 75 | No | $109 | — |
GeForce GTX 1050 Ti[19] | GP107-400 | 768:48:32 | 1290 | 1392 | 41.3 | 61.9 | 4 | 1981 (2138) | 62 (67) | 31 (33) | $139 | ||||||||||||
GeForce GTX 1060 3GB[20] | August 18, 2016 | GP106-300 | 16 | 4.4 | 200 | 1152:72:48 | 1506 | 1708 | 8000 | 72.3 | 108.4 | 3 | 192 | 192 | 3470 (3935) | 108 (123) | 54 (61) | 120 | $199 | ||||
GeForce GTX 1060 6GB[20] | July 19, 2016 | GP106-400 | 1280:80:48 | 120.5 | 6 | 3855 (4372) | 120 (137) | 60 (68) | $249 | $299 | |||||||||||||
GeForce GTX 1070[21] | June 10, 2016 | GP104-200 | 7.2 | 314 | 1920:120:64 | 1683 | 96.4[f][22] | 180.7 | 8 | 256 | 256 | 5783 (6463) | 181 (202) | 90 (101) | 150 | 2-way SLI HB[23] or traditional 2/3/4-way SLI[24] | $379 | $449 | |||||
GeForce GTX 1080[17] | May 27, 2016 | GP104-400 | 2560:160:64 | 1607 | 1733 | 10000 | 102.8 | 257.1 | 320 | GDDR5X | 8228 (8873) | 257 (277) | 128 (139) | 180 | $599 | $699 | |||||||
NVIDIA TITAN X[25] | August 2, 2016 | GP102-400 | 12 | 471[26] | 3584:224:96 | 1417 | 1531 | 136 | 317.4 | 12 | 480 | 384 | 10157 (10974) | 317 (343) | 159 (171) | 250 | — | $1200 |
- ^ Shader Processors : Texture mapping units : Render output units
- ^ For calculating the processing power see Pascal (microarchitecture)#Performance.
- ^ SLI HB only supports a maximum of 2-way SLI using SLI HB bridges, however if using traditional SLI bridges it can support a maximum of 4-way SLI but the performance mostly improve in synthetic benchmarks only.
- ^ Pixel fillrate is calculated as the lowest of three numbers: number of ROPs multiplied by the base core clock speed, number of rasterizers multiplied by the number of fragments they can generate per rasterizer multiplied by the base core clock speed, and the number of streaming multiprocessors multiplied by the number of fragments per clock that they can output multiplied by the base clock rate.
- ^ Texture fillrate is calculated as the number of TMUs multiplied by the base core clock speed.
- ^ The GTX 1070 has one of the four GPCs disabled in the die. Losing one of the Raster Engines only allows for the use of 48 ROPs per cycle.
GeForce 10 (10xx) series for notebooks
The biggest highlight to this line of notebook GPUs is the implementation of configured specifications close to (for the GTX 1060-1080) and exceeding (for the GTX1050/1050Ti) that of their desktop counterparts, as opposed to having "cut-down" specifications in previous generations. As a result, the "M" suffix is completely removed from the model's naming schemes, denoting these notebook GPUs to possess similar performance to those made for desktop PCs, including the ability to overclock their core frequencies by the user, something not possible with previous generations of notebook GPUs. This was made possible having lower TDP ratings as compared to their desktop equivalents, making these desktop-level GPUs thermally feasible to be implemented into a notebook chassis with better thermal dissipation solutions.
Model | Launch | Code name | Fab (nm) | Transistors (billion) | Die size (mm2) | Bus interface | Core config | Clock speeds | Fillrate | Memory | API support (version) | Processing power (GFLOPS) | TDP (watts) | SLI support | |||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Base core clock (MHz) | Boost core clock (MHz) | Memory (MT/s) | Pixel (GP/s) | Texture (GT/s) | Size (GiB) | Bandwidth (GB/s) | Bus type | Bus width (bit) | DirectX | OpenGL | OpenCL | Vulkan | Single precision (Boost) | Double precision | Half precision | ||||||||||
GeForce GTX 1050 (Notebook)[27] | January 3, 2017 | GP107 | 14 | 3.3 | 135 | PCIe 3.0 x16 | 640:40:16 | 1354 | 1493 | 7000 | 43.3 | 54.2 | Up to 4 | 112 | GDDR5 | 128 | 12.0 (12_1) | 4.5 | 1.2 | 1.0 | 1733 (1911) | 27 | 14 | 75 | No |
GeForce GTX 1050 Ti (Notebook)[27] | 768:48:32 | 1493 | 1620 | 47.8 | 71.7 | 2293 (2488) | 36 | 18 | |||||||||||||||||
GeForce GTX 1060 (Notebook)[27] | August 16, 2016 | GP106 | 16 | 4.4 | 200 | 1280:80:48 | 1404 | 1670 | 8000 | 67.4 | 112 | Up to 6 | 192 | 192 | 3594 (4275) | 112 | 56 | 80 | |||||||
GeForce GTX 1070 (Notebook)[27] | GP104 | 7.2 | 314 | 2048:128:64 | 1442 | 1645 | 92.3 | 185 | 8 | 256 | 256 | 5906 (6738) | 185 | 92 | 110 | Yes | |||||||||
GeForce GTX 1080 (Notebook)[27] | 2560:160:64 | 1556 | 1733 | 10000 | 99.6 | 249 | 320 | GDDR5X | 7967 (8873) | 249 | 124 | 150 |
Chipset table
See also
- GeForce 400 series
- GeForce 500 series
- GeForce 600 series
- GeForce 700 series
- GeForce 800M series
- GeForce 900 series
- Nvidia Quadro
- Nvidia Tesla
References
- ^ a b "Talks of foundry partnership between NVIDIA and Samsung (14nm) didn't succeed, and the GPU maker decided to revert to TSMC's 16nm process". Retrieved August 25, 2015.
- ^ "Samsung to Optical-Shrink NVIDIA "Pascal" to 14 nm". Retrieved August 13, 2016.
- ^ "NVIDIA Updates GPU Roadmap; Announces Pascal". The Official NVIDIA Blog.
- ^ Harris, Mark (April 5, 2016). "Inside Pascal: NVIDIA's Newest Computing Platform". Parallel Forall. Nvidia. Retrieved June 3, 2016.
- ^ "NVIDIA Pascal GPU Architecture to Provide 10X Speedup for Deep Learning Apps - NVIDIA Blog". The Official NVIDIA Blog.
- ^ "How The New Pascal Architecture Supports Next-Generation Video Playback". May 17, 2016. Retrieved June 11, 2016.
- ^ "Nvidia Pascal HDCP 2.2". Nvidia Hardware Page. Retrieved May 8, 2016.
- ^ Shrout, Ryan (July 14, 2016). "3DMark Time Spy: Looking at DX12 Asynchronous Compute Performance". PC Perspective. Retrieved July 14, 2016.
- ^ a b c Smith, Ryan (July 20, 2016). "The NVIDIA GeForce GTX 1080 & GTX 1070 Founders Editions Review: Kicking Off the FinFET Generation". AnandTech. p. 9. Retrieved July 21, 2016.
- ^ a b c d Smith, Ryan (July 20, 2016). "The NVIDIA GeForce GTX 1080 & GTX 1070 Founders Editions Review: Kicking Off the FinFET Generation". AnandTech. p. 10. Retrieved July 21, 2016.
- ^ a b c d e f g Smith, Ryan; Wilson, Derek (July 20, 2016). "The NVIDIA GeForce GTX 1080 & GTX 1070 Founders Editions Review: Kicking Off the FinFET Generation". AnandTech. p. 13. Retrieved July 22, 2016.
- ^ "Release 370 Graphics Drivers for Windows, Version 372.70" (PDF). August 30, 2016. Retrieved August 30, 2016.
- ^ "Nvidia's Pascal to use stacked memory, proprietary NVLink interconnect". The Tech Report. Retrieved March 26, 2014.
- ^ Hruska, Joel (May 31, 2016). "Nvidia's new GTX 1070 freezes out Fury, trashes Titan X". ExtremeTech. Retrieved June 5, 2016.
- ^ Burnes, Andrew (May 18, 2016). "GeForce GTX 1080 Founders Edition: Premium Construction & Advanced Features". GeForce.com. Retrieved May 19, 2016.
- ^ "A Quantum Leap for Every Gamer: NVIDIA Unveils the GeForce GTX 1060". Nvidia Newsroom. July 7, 2016. Retrieved October 18, 2016.
{{cite web}}
: Cite has empty unknown parameter:|1=
(help) - ^ a b Nvidia. "GTX 1080 Graphics Card". Retrieved May 7, 2016.
- ^ Smith, Ryan (July 20, 2016). "FP16 Throughput on GP104: Good for Compatibility (and Not Much Else) - The NVIDIA GeForce GTX 1080 & GTX 1070 Founders Editions Review: Kicking Off the FinFET Generation". AnandTech. Retrieved July 22, 2016.
- ^ a b Nvidia. "GTX 1050 Graphics Card". Retrieved October 18, 2016.
- ^ a b Nvidia. "GTX 1060 Graphics Card". Retrieved August 18, 2016.
- ^ Nvidia. "GTX 1070 Graphics Card". Retrieved May 18, 2016.
- ^ Smith, Ryan. "Synthetics - The NVIDIA GeForce GTX 1080 & GTX 1070 Founders Editions Review: Kicking Off the FinFET Generation". Retrieved 2016-07-21.
- ^ W1zzard (May 17, 2016). "Nvidia GeForce GTX 1080 8 GB". TechPowerUp. Retrieved May 17, 2016.
{{cite web}}
: CS1 maint: numeric names: authors list (link) - ^ W1zzard (June 21, 2016). "Nvidia GeForce GTX 1080 SLI". TechPowerUp. Retrieved June 21, 2016.
{{cite web}}
: CS1 maint: numeric names: authors list (link) - ^ Nvidia. "NVIDIA TITAN X Graphics Card with Pascal". Retrieved July 21, 2016.
- ^ Smith, Ryan. "Updated: NVIDIA Announces "NVIDIA Titan X" Video Card: $1200, Available August 2nd". AnandTech. Retrieved 25 July 2016.
- ^ a b c d e "GeForce GTX 10-Series Notebooks". geforce.com.