AMD Instinct MI300X vs NVIDIA RTX PRO 6000 Blackwell Server Comparison

The database contains no shared benchmark entry for the AMD Instinct MI300X and the NVIDIA RTX PRO 6000 Blackwell Server. The MI300X is recorded only in Geekbench OpenCL, where it scores 317994 and ranks at the 100th percentile, while the RTX PRO 6000 Blackwell Server is recorded only in 3DMark Steel Nomad DX12, where it scores 5996 and ranks at the 34th percentile. Because these are different test suites, the two percentile values cannot be compared directly. The specification data, however, shows two very different server accelerators: a 192 GB HBM3 AMD part with 5.32 TB/s of memory bandwidth and an NVIDIA part with 96 GB GDDR7, 126.0 TFLOPS FP32, and full graphics API support.

Head-to-Head Benchmarks

Neither card has a recorded head-to-head win over the other. No shared workload exists in the database, so a direct score comparison is impossible from the current data.

The MI300X's only benchmark is Geekbench OpenCL. Its score of 317994 puts it 5% behind the NVIDIA H200 NVL (334891), 8% behind the NVIDIA B200 (345482), 7.5% ahead of the NVIDIA L40S (295763), and 10.7% ahead of the NVIDIA RTX 6000 Ada Generation (287237). The largest recorded win in this pool is the 10.7% margin over the RTX 6000 Ada; the largest deficit is the 8% gap to the B200.

The RTX PRO 6000 Blackwell Server's only benchmark is 3DMark Steel Nomad DX12. Its score of 5996 is effectively level with its nearest rivals: the NVIDIA GeForce GTX 770M (6000, 0.1% ahead), the AMD Radeon RX 6400 (6001, 0.1% ahead), the AMD FirePro W4100 (5987, 0.2% behind), and the NVIDIA Quadro K4000M (5986, 0.2% behind). The largest recorded win in this pool is 0.2% over the FirePro W4100 and Quadro K4000M; the largest deficit is 0.1% to the GTX 770M and RX 6400.

Architecture Differences

Both chips are fabricated by TSMC on a 5 nm process, but the underlying designs diverge. The MI300X uses the Aqua Vanjaram chip with the CDNA 3.0 architecture; the RTX PRO 6000 Blackwell Server uses the GB202 chip with the Blackwell 2.0 architecture.

The MI300X integrates 153,000 million transistors on a 1017 mm² die, for a transistor density of 150.4M / mm². The RTX PRO integrates 92,200 million transistors on a 750 mm² die, for a density of 122.9M / mm². The AMD part therefore has the larger transistor budget and die area.

Memory is a major architectural split. The MI300X uses 192 GB of HBM3 on an 8192 bit bus, delivering 5.32 TB/s. The RTX PRO uses 96 GB of GDDR7 on a 512 bit bus, delivering 1.79 TB/s. The NVIDIA part has a higher memory clock, listed as 1750 MHz 28 Gbps effective, versus 1300 MHz 5.2 Gbps effective for the AMD part.

The compute arrays also differ. The MI300X has 19,456 shading units, 1,216 TMUs, and 0 ROPs, with no listed RT cores or tensor cores. The RTX PRO has 24,064 shading units, 752 TMUs, 192 ROPs, 188 RT cores, and 752 tensor cores. The NVIDIA card supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, while the MI300X lists N/A for DirectX, OpenGL, and Vulkan. The RTX PRO also has 4x DisplayPort 2.1b outputs; the MI300X has no outputs.

Clock speeds favor the NVIDIA part: base 1590 MHz, boost 2617 MHz versus base 1000 MHz, boost 2100 MHz. The RTX PRO reaches 126.0 TFLOPS FP32 and 126.0 TFLOPS FP16, while the MI300X reaches 81.72 TFLOPS for both. Pixel rate is 502.5 GPixel/s for the RTX PRO versus 0 MPixel/s for the MI300X, but texture rate is higher on the MI300X at 2,553.6 GTexel/s versus 1,968.0 GTexel/s.

Where Each One Wins

Because there are no shared benchmarks, the database cannot assign a direct win to either card. The only recorded wins are within each card's separate rival group. In OpenCL, the MI300X beats the L40S and RTX 6000 Ada by 7.5% and 10.7%, respectively. In 3DMark Steel Nomad DX12, the RTX PRO beats the FirePro W4100 and Quadro K4000M by 0.2% each.

The specification data gives each card a clear area of advantage. The MI300X wins on memory capacity (192 GB versus 96 GB), memory bus width (8192 bit versus 512 bit), memory bandwidth (5.32 TB/s versus 1.79 TB/s), transistor count (153,000 million versus 92,200 million), die size (1017 mm² versus 750 mm²), and texture rate (2,553.6 GTexel/s versus 1,968.0 GTexel/s). The RTX PRO wins on shading units (24,064 versus 19,456), FP32 throughput (126.0 TFLOPS versus 81.72 TFLOPS), FP16 throughput (126.0 TFLOPS versus 81.72 TFLOPS), pixel rate (502.5 GPixel/s versus 0 MPixel/s), ROPs (192 versus 0), RT cores (188 versus none listed), tensor cores (752 versus none listed), base and boost clocks, and API and display support. It also has a lower TDP at 600 W versus 750 W.

Specification Differences

The table lists the recorded fields that differ between the two cards.

| Field | AMD Instinct MI300X | NVIDIA RTX PRO 6000 Blackwell Server |

|---|---|---|

| Manufacturer | AMD | NVIDIA |

| Chip | Aqua Vanjaram | GB202 |

| Architecture | CDNA 3.0 | Blackwell 2.0 |

| Transistors | 153,000 million | 92,200 million |

| Die size | 1017 mm² | 750 mm² |

| Transistor density | 150.4M / mm² | 122.9M / mm² |

| Base clock | 1000 MHz | 1590 MHz |

| Boost clock | 2100 MHz | 2617 MHz |

| Memory clock | 1300 MHz 5.2 Gbps effective | 1750 MHz 28 Gbps effective |

| Memory size | 192 GB | 96 GB |

| Memory type | HBM3 | GDDR7 |

| Memory bus width | 8192 bit | 512 bit |

| Memory bandwidth | 5.32 TB/s | 1.79 TB/s |

| Shading units | 19456 | 24064 |

| TMUs | 1216 | 752 |

| ROPs | 0 | 192 |

| RT cores | Not listed | 188 |

| Tensor cores | Not listed | 752 |

| Pixel rate | 0 MPixel/s | 502.5 GPixel/s |

| Texture rate | 2,553.6 GTexel/s | 1,968.0 GTexel/s |

| FP32 | 81.72 TFLOPS | 126.0 TFLOPS |

| FP16 | 81.72 TFLOPS (1:1) | 126.0 TFLOPS (1:1) |

| TDP | 750 W | 600 W |

| Slot width | OAM Module | Dual-slot |

| Power connectors | None | 1x 16-pin |

| Suggested PSU | 1150 W | 1000 W |

| Display outputs | No outputs | 4x DisplayPort 2.1b |

| API support | DirectX N/A, OpenGL N/A, Vulkan N/A | DirectX 12 Ultimate (12_2), OpenGL 4.6, Vulkan 1.4 |

| Dimensions | Not listed | 267 mm 10.5 inches, 111 mm 4.4 inches, 40 mm 1.6 inches |

| Production status | Not listed | Active |

| Release date | 2023-12-05 | 2025-03-17 |

| Predecessor | Radeon Instinct | Server Hopper |

| Successor | Not listed | Server Rubin |

FAQ

Q: What benchmark is recorded for the MI300X and how does it compare to its nearest rivals?

A: The MI300X scores 317994 in Geekbench OpenCL and sits at the 100th percentile. It is 5% behind the H200 NVL (334891), 8% behind the B200 (345482), 7.5% ahead of the L40S (295763), and 10.7% ahead of the RTX 6000 Ada Generation (287237).

Q: What benchmark is recorded for the RTX PRO 6000 Blackwell Server and how does it compare?

A: The RTX PRO scores 5996 in 3DMark Steel Nomad DX12 and sits at the 34th percentile. The GTX 770M (6000) and RX 6400 (6001) are 0.1% ahead; the FirePro W4100 (5987) and Quadro K4000M (5986) are 0.2% behind.

Q: Which card has more memory and bandwidth?

A: The MI300X has 192 GB of HBM3 on an 8192 bit bus with 5.32 TB/s of bandwidth. The RTX PRO has 96 GB of GDDR7 on a 512 bit bus with 1.79 TB/s of bandwidth.

Q: Which card has higher FP32 and FP16 throughput?

A: The RTX PRO has 126.0 TFLOPS FP32 and 126.0 TFLOPS FP16. The MI300X has 81.72 TFLOPS for both FP32 and FP16.

Q: Does the MI300X support graphics APIs or display outputs?

A: No. Its DirectX, OpenGL, and Vulkan entries are N/A, and it has no outputs. The RTX PRO supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, and has 4x DisplayPort 2.1b outputs.

Q: What are the power specifications for each card?

A: The MI300X has a 750 W TDP, uses an OAM module, has no power connectors, and lists a suggested PSU of 1150 W. The RTX PRO has a 600 W TDP, uses a dual-slot design, has one 16-pin connector, and lists a suggested PSU of 1000 W.

The Verdict

The recorded data does not provide a direct performance comparison. The MI300X is a 192 GB HBM3 accelerator with 5.32 TB/s of memory bandwidth and 81.72 TFLOPS FP32 and FP16, built for memory-capacity-bound compute. The RTX PRO 6000 Blackwell Server is a 96 GB GDDR7 card with 126.0 TFLOPS FP32 and FP16, 188 RT cores, 752 tensor cores, and full graphics API support, suited to workloads that need rendering, ray tracing, or display output.

In the database's separate benchmark pools, the MI300X sits at the 100th percentile in Geekbench OpenCL, while the RTX PRO sits at the 34th percentile in 3DMark Steel Nomad DX12. These percentiles are not directly comparable. The data supports selecting the MI300X for massive memory residency and memory bandwidth, and the RTX PRO for higher raw FP32 throughput, graphics features, and a lower power envelope.