AMD Radeon Pro Vega 64X vs NVIDIA B200 Comparison

AMD
RADEON

AMD Radeon Pro Vega 64X

CORE STATE Vega 10
VRAM 16 GB
CLOCK SPEED 1468 MHz
TDP 250 W
BUS WIDTH 2048 bit
ARCHITECTURE GCN 5.0
nm
PROCESS 14 nm
LAUNCH DATE 2019
VS
NVIDIA
GEFORCE

B200

CORE STATE GB100
VRAM 90 GB
CLOCK SPEED 1965 MHz
TDP 1000 W
BUS WIDTH 4096 bit
ARCHITECTURE Blackwell
nm
PROCESS 5 nm
LAUNCH DATE

PERFORMANCE BENCHMARKS

geekbench_metal
83,450
N/A
geekbench_opencl
78,467
345,482

Analysis: AMD Radeon Pro Vega 64X vs NVIDIA B200

Head-to-Head Benchmarks

The only directly comparable benchmark in the database is Geekbench OpenCL, and the result is a decisive victory for the NVIDIA B200. The B200 scores 345482, while the AMD Radeon Pro Vega 64X scores 78467. This gives the B200 a 340.3% advantage, meaning it is more than four times faster in this compute-oriented workload. The delta is so large that it effectively places these two cards in entirely different performance tiers.

To put the B200's score into context, its average benchmark score of 345482 places it at the 100th percentile of all GPUs in the database. That is a perfect ranking, indicating no other recorded GPU has matched or exceeded its OpenCL output. The nearest rival, the NVIDIA H200 NVL, scores 334891, which is 3.2% behind. The next closest, the AMD Instinct MI300X, scores 317994, a gap of 8.6%. Even the NVIDIA B300 SXM6 AC, which is ahead of the B200 by 6.6%, does not displace the B200 from the top percentile ranking. The NVIDIA L40S trails by 16.8% with a score of 295763.

The AMD Radeon Pro Vega 64X, by contrast, sits at the 92nd percentile with an average benchmark score of 80959. Its OpenCL result of 78467 is slightly below its Metal score, which reaches 83450. The nearest rivals for the Vega 64X are much closer in performance. The AMD Radeon PRO W6600 scores 81995, which is 1.3% higher. The NVIDIA GeForce RTX 5090 scores 79842, 1.4% higher. The NVIDIA Tesla P100 PCIe 16 GB scores 79605, 1.7% higher, and the Tesla P100 PCIe 12 GB scores 79396, 2% higher. These small deltas show that the Vega 64X is competitive within its own peer group, but that group is nowhere near the B200's performance class.

The head-to-head Geekbench OpenCL result shows the B200 winning the only shared test. The database records 1 win for the B200 and 0 wins for the Vega 64X. There is no Metal benchmark for the B200, and no additional OpenCL variant for the Vega 64X, so the comparison rests entirely on this single workload. Still, the margin is so large that the conclusion is unambiguous: in raw compute throughput, the B200 is in a different league.

FAQ

Q: How much faster is the NVIDIA B200 than the AMD Radeon Pro Vega 64X in Geekbench OpenCL?

A: The B200 scores 345482, while the Vega 64X scores 78467. The B200 is 340.3% faster, a lead of more than four times.

Q: Where does the NVIDIA B200 rank among all GPUs in the database?

A: The B200 is at the 100th percentile of all GPUs, with an average benchmark score of 345482. Its nearest competitor, the NVIDIA H200 NVL, is 3.2% behind.

Q: What is the AMD Radeon Pro Vega 64X's best benchmark result?

A: The Vega 64X scores 83450 in Geekbench Metal, which is higher than its OpenCL score of 78467. Its average benchmark score is 80959.

Q: Which GPUs are closest to the AMD Radeon Pro Vega 64X in performance?

A: The AMD Radeon PRO W6600 is 1.3% ahead, the NVIDIA GeForce RTX 5090 is 1.4% ahead, and the NVIDIA Tesla P100 variants are 1.7% and 2% ahead. All are within a narrow 2% band.

Q: Does the AMD Radeon Pro Vega 64X beat the NVIDIA B200 in any benchmark?

A: No. In the only shared benchmark, Geekbench OpenCL, the B200 wins. The database records 1 win for the B200 and 0 wins for the Vega 64X.

Q: What is the NVIDIA B200's percentile ranking compared to the Vega 64X's?

A: The B200 is at the 100th percentile, while the Vega 64X is at the 92nd percentile. The gap in percentile reflects the massive difference in recorded compute performance.

Where Each One Wins

The NVIDIA B200 wins in raw compute throughput, and it wins by an enormous margin. Its Geekbench OpenCL score of 345482 dwarfs the Vega 64X's 78467, making it the clear choice for workloads that demand maximum floating-point performance. The B200's position at the 100th percentile means it outperforms every other GPU in the database, including the NVIDIA H200 NVL, which trails by 3.2%, and the AMD Instinct MI300X, which trails by 8.6%. This is a card designed for top-tier compute density, and the data reflects that positioning.

The AMD Radeon Pro Vega 64X wins in a different sense: it is a viable option within a much lower performance envelope. Its Metal score of 83450 shows strength in Apple-ecosystem workloads, and its OpenCL score of 78467 is competitive with its immediate rivals. The Vega 64X sits within 2% of the AMD Radeon PRO W6600, NVIDIA GeForce RTX 5090, and both Tesla P100 variants. For applications that target Metal, or for systems where the Vega 64X's integrated form factor is a requirement, it remains serviceable despite being end-of-life.

The use-case split is stark. The B200 is for server-scale compute, with a 1000 W TDP and SXM Module slot width, and it delivers performance at the 100th percentile. The Vega 64X is for portable or integrated Mac systems, with a 250 W TDP, no power connectors, and display outputs described as portable device dependent. Neither card is a gaming product: the B200 has no display outputs, and the Vega 64X is an IGP with portable-device-dependent outputs. The B200 wins on sheer number-crunching power; the Vega 64X wins only in the narrow context of low-power integrated graphics workloads.

Specification Differences

The memory subsystems are radically different. The NVIDIA B200 has 90 GB of HBM3e on a 4096-bit bus, delivering 4.10 TB/s of bandwidth. The AMD Radeon Pro Vega 64X has 16 GB of HBM2 on a 2048-bit bus, delivering 512.0 GB/s. That is more than 8 times the capacity and 8 times the bandwidth for the B200. The memory clocks also differ: the B200 runs at 2000 MHz with 8 Gbps effective, while the Vega 64X runs at 1000 MHz with 2 Gbps effective.

The compute resources show a similar disparity. The B200 has 18944 shading units, 592 TMUs, and 24 ROPs. The Vega 64X has 4096 shading units, 256 TMUs, and 64 ROPs. The B200 also has 592 tensor cores, while the Vega 64X has none. Pixel rate favors the Vega 64X at 93.95 GPixel/s versus the B200's 47.16 GPixel/s, but texture rate favors the B200 at 1,163.3 GTexel/s versus 375.8 GTexel/s. Floating-point performance heavily favors the B200: 74.45 TFLOPS FP32 versus 12.03 TFLOPS, and 1,191.2 TFLOPS FP16 versus 24.05 TFLOPS.

The power profiles are in different categories. The B200 has a TDP of 1000 W and a suggested PSU of 1400 W, while the Vega 64X has a TDP of 250 W and no suggested PSU listed. The B200 uses an SXM Module slot width, while the Vega 64X is an IGP. The B200 has no display outputs; the Vega 64X has portable-device-dependent outputs. The bus interfaces differ as well: the B200 uses PCIe 5.0 x16, while the Vega 64X uses PCIe 3.0 x16.

Architecture Differences

The NVIDIA B200 is built on the Blackwell architecture with the GB100 chip, manufactured on a 5 nm process at TSMC. It contains 104,000 million transistors. The AMD Radeon Pro Vega 64X uses the GCN 5.0 architecture with the Vega 10 chip, manufactured on a 14 nm process at GlobalFoundries, and contains 12,500 million transistors. The Vega 64X has a die size of 495 mm² and a transistor density of 25.3M per mm²; the B200's die size and transistor density are not recorded in the database.

The B200 belongs to the Server Blackwell generation, with a predecessor in Server Hopper and a successor in Server Rubin. The Vega 64X belongs to the Radeon Pro Mac generation, with no predecessor or successor listed. The B200 is still active in production, while the Vega 64X is end-of-life, with a release date of March 18, 2019.

API support differs. The Vega 64X supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3. The B200 lists no API support in the database, reflecting its server-oriented design with no display outputs. The Vega 64X's memory type is HBM2, while the B200 uses HBM3e. The Vega 64X's clock speeds are higher at base and boost: 1250 MHz base and 1468 MHz boost, compared to the B200's 700 MHz base and 1965 MHz boost. The B200's higher boost clock, combined with its massive shading unit count, explains its dominant compute performance despite the lower base clock.

DETAILED SPECIFICATIONS

SPECIFICATION
Pro Vega 64X
B200
Core Specs
Shading Units
4,096
18,944 +362.5%
Shaders
4,096
18,944 +362.5%
TMUs
256
592 +131.3%
ROPs
64
24 -62.5%
Compute Units
64
SM Count
148
Clocks
Base Clock
1250 MHz
700 MHz
Boost Clock
1468 MHz
1965 MHz
Memory Clock
1000 MHz 2 Gbps effective
2000 MHz 8 Gbps effective
Memory
Memory Size
16 GB
90 GB
VRAM (MB)
16,384
92,160 +462.5%
Memory Type
HBM2
HBM3e
Memory Bus
2048 bit
4096 bit
Bandwidth
512.0 GB/s
4.10 TB/s
Cache
L1 Cache
16 KB (per CU)
256 KB (per SM)
L2 Cache
4 MB
50 MB
Performance
Pixel Rate
93.95 GPixel/s
47.16 GPixel/s
Texture Rate
375.8 GTexel/s
1,163.3 GTexel/s
FP32 (TFLOPS)
12.03 TFLOPS
74.45 TFLOPS
FP64 (TFLOPS)
751.6 GFLOPS (1:16)
37.22 TFLOPS (1:2)
FP16 (TFLOPS)
24.05 TFLOPS (2:1)
1,191.2 TFLOPS (16:1)
AI/RT
Tensor Cores
592
Power
TDP
250 W
1000 W
TDP (W)
250
1,000 +300.0%
Suggested PSU
1400 W
Power Connectors
None
Architecture
Architecture
GCN 5.0
Blackwell
GPU Name
Vega 10
GB100
Generation
Radeon Pro Mac (Vega Series)
Server Blackwell (Bxx)
Process Size
14 nm
5 nm
Transistors
12,500 million
104,000 million
Die Size
495 mm²
Foundry
GlobalFoundries
TSMC
Density
25.3M / mm²
API Support
DirectX
12 (12_1)
OpenGL
4.6
Vulkan
1.3
OpenCL
2.1
3.0
CUDA
10.0
Shader Model
6.7
Physical
Slot Width
IGP
SXM Module
Outputs
Portable Device Dependent
No outputs
Bus Interface
PCIe 3.0 x16
PCIe 5.0 x16
Other
Production
End-of-life
Active
Predecessor
Server Hopper
Successor
Server Rubin
View Radeon Pro Vega 64X Details View B200 Details