AMD Radeon Instinct MI60 vs AMD Radeon Pro WX 9100 Comparison

The Verdict

The AMD Radeon Instinct MI60 is the clear performance pick. Across the two shared benchmark tests, it wins both, with a 38.9% lead in Geekbench OpenCL and a 69% lead in Geekbench Vulkan. Its average benchmark score of 92,466 places it in the 93rd percentile of all GPUs, while the Radeon Pro WX 9100 sits at 64,212 average, in the 89th percentile. If the workload is compute-heavy and the card will live in a server or render node, the MI60 justifies its position with roughly 44% higher average performance.

The Radeon Pro WX 9100 is the more practical choice for a workstation with display needs. It has six mini-DisplayPort 1.4a outputs versus the MI60's single output, and its 230 W power draw with a 550 W suggested PSU is friendlier for a desktop chassis. The MI60 demands 300 W and a 700 W suggested PSU. For someone running multi-monitor setups or who needs to drive several displays from one card, the WX 9100's output configuration is decisive. Its launch MSRP was 1,599 USD, but pricing should not factor into the decision here.

The data indicates that the MI60 is for pure compute throughput, while the WX 9100 is for visual workloads with multiple displays. Neither card is current production; both are end-of-life, so the choice is about matching existing hardware to workload.

Architecture Differences

Both cards use AMD GCN architectures, but they are on different revisions and process nodes. The MI60 uses Vega 20 on GCN 5.1, built on TSMC's 7 nm process. The WX 9100 uses Vega 10 on GCN 5.0, built on GlobalFoundries' 14 nm node. This node difference explains much of the performance gap: the MI60 packs 13,230 million transistors into a 331 mm² die, yielding a density of 40.0M transistors per mm². The WX 9100 has 12,500 million transistors on a 495 mm² die, for 25.3M per mm². The MI60 is smaller physically but denser, which contributes to its higher boost clock.

The memory subsystems diverge sharply. The MI60 has 32 GB of HBM2 on a 4096-bit bus, delivering 1.02 TB/s of bandwidth. The WX 9100 has 16 GB of HBM2 on a 2048-bit bus, for 483.8 GB/s. That is a 2x bus width and roughly 2.1x bandwidth advantage for the MI60. Both have the same compute core counts: 4096 shading units, 256 TMUs, and 64 ROPs. The MI60's clock advantage (1800 MHz boost versus 1500 MHz) and memory bandwidth push its peak rates higher: 115.2 GPixel/s pixel rate and 460.8 GTexel/s texture rate versus 96.00 GPixel/s and 384.0 GTexel/s on the WX 9100.

FP32 throughput tells the same story. The MI60 reaches 14.75 TFLOPS, while the WX 9100 manages 12.29 TFLOPS. Both offer FP16 at a 2:1 ratio, with the MI60 at 29.49 TFLOPS and the WX 9100 at 24.58 TFLOPS. The MI60 also runs on PCIe 4.0 x16, while the WX 9100 is PCIe 3.0 x16. DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3 support are identical across both. Neither card has ray tracing or tensor cores.

Head-to-Head Benchmarks

The recorded benchmark data shows the MI60 winning both shared tests by substantial margins. In Geekbench OpenCL, the MI60 scores 92,488 against the WX 9100's 66,605, a delta of 38.9%. This is the more relevant test for compute workloads, and the gap reflects the MI60's higher clocks, wider memory bus, and newer process node. The WX 9100's score is still respectable, placing it among mid-range workstation cards, but it trails by nearly four-tenths.

In Geekbench Vulkan, the MI60 scores 92,444 versus 54,711 for the WX 9100, a delta of 69%. Vulkan is more sensitive to driver and memory latency, and the MI60's 1.02 TB/s bandwidth likely explains the larger gap here. The WX 9100's Vulkan result is its weakest of the three tests it has recorded; its OpenCL score of 66,605 is higher, and it also has a Metal score of 71,319. The MI60 has no recorded Metal result, so cross-API comparisons are limited to OpenCL and Vulkan.

The average benchmark scores confirm the hierarchy. The MI60's average is 92,466, while the WX 9100's is 64,212. That is a 44% difference in average performance. The MI60's nearest rivals in the database are the NVIDIA RTX A4500 at 91,671 (0.9% behind) and the RTX A4500 Mobile at 91,134 (1.5% behind). The WX 9100's nearest rivals are the NVIDIA CMP 30HX at 63,842 (0.6% ahead) and the AMD Radeon RX 9060 XT LP at 63,830 (0.6% ahead). The MI60 competes with modern mid-range workstation GPUs, while the WX 9100 sits near much older or lower-tier parts.

Specification Differences

The two cards differ on nearly every major specification except core counts and physical dimensions. Both have 4096 shading units, 256 TMUs, 64 ROPs, dual-slot width, 1x 6-pin plus 1x 8-pin power connectors, and identical dimensions of 267 mm length and 111 mm height. Beyond those, the differences are:

• Process node: 7 nm (TSMC) for MI60, 14 nm (GlobalFoundries) for WX 9100
• Transistors: 13,230 million for MI60, 12,500 million for WX 9100
• Die size: 331 mm² for MI60, 495 mm² for WX 9100
• Transistor density: 40.0M / mm² for MI60, 25.3M / mm² for WX 9100
• Base clock: 1200 MHz for both
• Boost clock: 1800 MHz for MI60, 1500 MHz for WX 9100
• Memory clock: 1000 MHz (2 Gbps effective) for MI60, 945 MHz (1890 Mbps effective) for WX 9100
• Memory size: 32 GB for MI60, 16 GB for WX 9100
• Memory bus: 4096 bit for MI60, 2048 bit for WX 9100
• Memory bandwidth: 1.02 TB/s for MI60, 483.8 GB/s for WX 9100
• Pixel rate: 115.2 GPixel/s for MI60, 96.00 GPixel/s for WX 9100
• Texture rate: 460.8 GTexel/s for MI60, 384.0 GTexel/s for WX 9100
• FP32: 14.75 TFLOPS for MI60, 12.29 TFLOPS for WX 9100
• FP16: 29.49 TFLOPS for MI60, 24.58 TFLOPS for WX 9100
• TDP: 300 W for MI60, 230 W for WX 9100
• Suggested PSU: 700 W for MI60, 550 W for WX 9100
• Bus interface: PCIe 4.0 x16 for MI60, PCIe 3.0 x16 for WX 9100
• Display outputs: 1x mini-DisplayPort 1.4a for MI60, 6x mini-DisplayPort 1.4a for WX 9100
• Release date: 2018-11-17 for MI60, 2017-07-09 for WX 9100
• Predecessor: FirePro Data Center for MI60, Radeon Pro GCN for WX 9100
• Successor: none for MI60, Radeon Pro Vega for WX 9100
• Launch MSRP: none for MI60, 1,599 USD for WX 9100

The MI60's memory advantage is the most striking: double the capacity, double the bus width, and more than double the bandwidth. The WX 9100's display output count is its clear advantage, with six ports versus one.

FAQ

Q: Which card is faster in OpenCL workloads?

A: The MI60 scores 92,488 in Geekbench OpenCL, which is 38.9% higher than the WX 9100's 66,605. The MI60 wins this test decisively.

Q: Does the WX 9100 have any benchmark where it beats the MI60?

A: No. The WX 9100 has three recorded benchmark scores (Metal, OpenCL, Vulkan), and the MI60 has two (OpenCL, Vulkan). In the two shared tests, the MI60 wins both. There is no head-to-head test where the WX 9100 comes out ahead.

Q: Can the MI60 drive multiple displays?

A: No. It has a single mini-DisplayPort 1.4a output. The WX 9100 has six mini-DisplayPort 1.4a outputs, which makes it the only option for multi-monitor setups.

Q: How much memory does each card have?

A: The MI60 has 32 GB of HBM2 on a 4096-bit bus with 1.02 TB/s bandwidth. The WX 9100 has 16 GB of HBM2 on a 2048-bit bus with 483.8 GB/s bandwidth.

Q: What is the power requirement difference?

A: The MI60 has a 300 W TDP and a suggested PSU of 700 W. The WX 9100 has a 230 W TDP and a suggested PSU of 550 W. The WX 9100 is easier to fit into a standard desktop power envelope.

Q: Are these cards still in production?

A: No. Both are end-of-life. The MI60 was released in November 2018, and the WX 9100 was released in July 2017.

Where Each One Wins

The MI60 wins in raw compute throughput. It is ahead by 38.9% in OpenCL and 69% in Vulkan, and its average score is 92,466 versus 64,212. The 32 GB memory capacity and 1.02 TB/s bandwidth make it suitable for large datasets that exceed the WX 9100's 16 GB frame buffer. The PCIe 4.0 interface also gives it a bandwidth advantage to the host system. If the workload is training models, rendering frames in batches, or processing large arrays, the MI60 is the card to pick. Its single display output signals that it is not meant for interactive use; it is a compute accelerator that happens to have a display port for basic setup.

The WX 9100 wins in workstation flexibility. Six mini-DisplayPort outputs let one card drive six monitors or a multi-display visualization wall. Its 230 W TDP and 550 W suggested PSU make it easier to slot into existing workstation builds without upgrading the power supply. The 16 GB memory is still substantial for many professional workloads, and its OpenCL score of 66,605 is not weak; it just lags the MI60 by a wide margin. The WX 9100's Metal score of 71,319 is its strongest recorded result, so for Apple-centric workflows that use Metal, it has a relevant advantage, though no comparison exists for the MI60 on that API.

Use-case split: choose the MI60 for headless compute servers, batch rendering, or any task where memory capacity and raw FP32 throughput dominate. Choose the WX 9100 for interactive workstation use with multiple displays, lower power draw, or when the host system lacks PCIe 4.0 support. Both are end-of-life, so the decision is about what the existing system can support and what the workload demands. The MI60 is the performance leader; the WX 9100 is the connectivity and power-efficiency leader.