 |
 |
|
| Name | Nvidia Tesla K80 | Nvidia Tesla V100 |
| Release Date | 17 November 2014 | 21 June 2017 |
| Price | USD 240.50 | USD 7,254.07 |
| GPU Score | ||
| Brand | Nvidia | Nvidia |
| Series | Tesla K80 | Tesla V100 |
The Tesla Series of graphics processors is another hallmark of Nvidia’s superiority in the graphics computing landscape. The two most prominent additions to this series were the Tesla K80 and Tesla V100. Although these GPUs were released over half a decade ago, they still have substantial use in many cases, as newer professional graphics cards can be expensive.
The Tesla V100 essentially continues the same recipe found in the K80. However, it’s much more robust, reliable, and powerful at churning the same datasets. But it is priced exponentially more than the K80, and for those with enthusiasm for AI development and working with large datasets, K80 still does the job pretty well.
Nevertheless, both graphics accelerators are excellent at what they do, but the V100 outperforms and outmatches the K80 in direct competition.
The Tesla K80 was built on the 28nm chip developed by TSMC and utilizes the Kepler 2.0 Architecture. The fabrication is quite extensive by modern standards, but the K80 isn’t held back by it. This features a 561mm2 die and packs in more than 7K million transistors, which allows excellent performance and efficient computing along with multiple cores and pipelines.
The Tesla V100 utilizes the Volta microarchitecture, working on a 12nm TSMC and features a three time higher transistor count of 21,100 Million. This, with the addition of several more cores and pipelines, results in a much larger overall die size and making it a lot better in performance and power efficiency than the K80.
| GPU Name | Nvidia Tesla K80 | Nvidia Tesla V100 |
| GPU Variant | GK210-885-A1 | - |
| Architecture | Kepler 2.0 | Volta |
| Manufacturing Size | 28 nm | 12 nm |
| Numbers of Transistors | 7,100 million | 21,100 million |
| Die Size | 561 mm² | 815 mm² |
The K80 and V100 also have highly distinctive clock speeds. The V100 features a substantially higher core clock with an extensive range and a base of 1245 MHz, which can boost to 1380 MHz with ample power and thermal room. The K80 has a decent core frequency but fails to catch up with the V100. Its core clock range is between 562 and 824 MHz.
Additionally, the K80 flips the case in its favor with its memory clock. It has a higher memory clock of 1250 MHz, and it is rated to be more effective. Meanwhile, the V100 features a barely nominal memory frequency of 876 MHz with much less effectiveness.
| Base Clock | 562 MHz | 1245 MHz |
| Boost Clock | 824 MHz | 1380 MHz |
| Memory Clock | 1253 MHz 5 Gbps effective | 876 MHz 1752 Mbps effective |
Compared to the 12 GB memory of the K80, the V100 brings a higher memory count of 16 gigabytes, with a substantially faster type of memory. The V100 entails an HBM2 memory. Meanwhile, the K80 works on the GDDR5 memory. Now, GDDR5 memories are good, but the HBM2 memory is on a different level of superiority with its significantly higher memory bandwidths.
Like their memories, the K80 and V100 also have different bus widths. The K80 uses the more traditional bus size of 384-bit, but the V100 dwarfs it with a 4096-bit memory bus. This also directly determines memory bandwidth, and to no one’s surprise, the V100 is capable of much higher bandwidths than the K80.
| Memory Size | 12 GB | 16 GB |
| Memory Type | GDDR5 | HBM2 |
| Memory Bus | 384 bit | 4096 bit |
| Bandwidth | 240.6 GB/s | 897.0 GB/s |
Both graphics accelerators take two slots and feature a similar length of around 267mm. They also feature identical bodies and look with different color variations.
Identically, both have a power consumption of 300 Watts and require a 700-watt PSU at minimum to efficiently work in tandem with other necessary hardware. The difference lies in the power delivery. Where the K80 uses a single 8-pin connection, the V100 requires two 8-pin connections direct from the power supply, which weren’t really mainstream back in 2017.
Also, as graphics accelerators made solely for computation and were intended for specialized use cases, these GPUs don’t feature any display outputs. They cannot be used like mainstream GPUs without some additional tinkering.
| Number of Slots | Nvidia RTX 3090 | AMD RX 5700 XT |
| Lenght | 267 mm | 267 mm |
| Width | - | - |
| Height | - | - |
| TDP | 300 W | 300 W |
| Power Connectors | 1x 8-pin | 2x 8-pin |
| Suggested PSU | 700 W | 700 W |
| Display Outputs | No outputs | No outputs |
The V100 is well-suited for modern API support and supports almost all modern programming tools, such as Vulkan 1.3, DirectX 12, etc., and the latest shader model support. Meanwhile, the official support for K80 is a bit limited in the case of Direct X, Shader Model, and Vulkan.
| Direct X | 12 (11_1) | 12 (12_1) |
| OpenGL | 4.6 | 4.6 |
| OpenCL | 3.0 | 3.0 |
| Vulkan | 1.2.175 | 1.3 |
| Shader Model | 6.5 (5.1) | 6.7 |
