AMD Radeon Pro W5700
amd RDNA1 (Navi 10) 2019
Fields
id
accelerators/amd-radeon-pro-w5700Vendor
amd
Architecture
RDNA1 (Navi 10)
Release year
2019
Process (nm)
7
Form factors
pcie
VRAM (GB)
8
Memory type
gddr6
Memory bus (bit)
256
Memory bandwidth (GB/s)
448
Bandwidth basis
Radeon Pro W5000 series table on the Radeon Pro Wikipedia page, W5700 (Navi 10) row: 8 GB GDDR6, 256-bit, 448 GB/s. The AMD product page the table cites returned 404 on 2026-10-17, so this is the compiled table rather than a live vendor sheet.
Peak FLOPS
precisionfp32tflops8.663denseyesvendor_claimnosource_idamd-radeon-pro-w5700-wikipedia-spec-tablebasisderivedbasis_detailSingle-precision boost figure printed in the same table: 8663 GFLOPS at the 1880 MHz boost clock (2304 shaders, FMA). Base-clock figure in the same cell is 6451 GFLOPS at 1400 MHz. Recorded at boost, which is the number a buyer is quoted.precisionfp16tflops17.33denseyesvendor_claimnosource_idamd-radeon-pro-w5700-wikipedia-spec-tablebasisderivedbasis_detailHalf-precision boost figure in the same cell: 17330 GFLOPS. RDNA1 packed FP16 is 2x FP32, and 17.330 / 8.663 = 2.00, so the two cells agree.
TDP (W)
205
Unified memory
no
Consumer part
no
notes
Workstation Navi 10 card, the pro sibling of the RX 5700 XT with the same 8 GB / 256-bit GDDR6 but a lower board power (205 W vs the XT 225 W) and a 799 USD launch price on 19 Nov 2019. 36 CU, 2304 shaders, base 1400 MHz, boost 1880 MHz, PCIe 4.0 x16. 8 GB at 448 GB/s holds a 7B Q4 and is tight on a 13B-14B Q4. The W5700X is a different Apple-specific board (40 CU, 16 GB, 250 W) and is not this record. No GPU-to-GPU fabric. Added because discussion #10879 now has a llama-bench row for the W5700 and the accelerator did not exist. Spec confidence is capped at 0.7 because the AMD product page 404d and the numbers come from the compiled table.
Notes
Workstation Navi 10 card, the pro sibling of the RX 5700 XT with the same 8 GB / 256-bit GDDR6 but a lower board power (205 W vs the XT 225 W) and a 799 USD launch price on 19 Nov 2019. 36 CU, 2304 shaders, base 1400 MHz, boost 1880 MHz, PCIe 4.0 x16. 8 GB at 448 GB/s holds a 7B Q4 and is tight on a 13B-14B Q4. The W5700X is a different Apple-specific board (40 CU, 16 GB, 250 W) and is not this record. No GPU-to-GPU fabric. Added because discussion #10879 now has a llama-bench row for the W5700 and the accelerator did not exist. Spec confidence is capped at 0.7 because the AMD product page 404d and the numbers come from the compiled table.