AMD Steam Machine GPU vs NVIDIA B300 Comparison

Where Each One Wins

The AMD Steam Machine GPU and NVIDIA B300 occupy entirely separate corners of the GPU landscape, and the recorded data makes the split unambiguous. The AMD Steam Machine GPU is a compact, console-oriented part built for a specific Valve platform, while the NVIDIA B300 is a server-class compute module. There are no shared benchmark results in the database, so the comparison rests on architectural and specification differences rather than direct performance measurements.

The AMD Steam Machine GPU wins in any scenario that demands a self-contained, low-power graphics solution. Its 110 W power draw, lack of external power connectors, and physical dimensions of 156 mm by 152 mm by 162 mm place it firmly in the small-form-factor console space. It carries display outputs, specifically 1x HDMI 2.1a and 1x DisplayPort 2.1, which means it can drive consumer displays directly. Its 8 GB of GDDR6 memory on a 128-bit bus delivers 288.0 GB/s of bandwidth, a figure suited to 1080p-class gaming workloads. The pixel rate of 156.8 GPixel/s and texture rate of 274.4 GTexel/s indicate a part designed for rasterization-heavy interactive workloads, not compute acceleration.

The NVIDIA B300 wins in every metric tied to raw compute throughput and memory capacity. Its 144 GB of HBM3e memory on a 4096-bit bus produces 4.10 TB/s of bandwidth, over fourteen times the AMD part's memory bandwidth. The FP32 throughput of 76.99 TFLOPS dwarfs the AMD Steam Machine GPU's 17.56 TFLOPS. The FP16 figure of 1,231.8 TFLOPS (16:1 ratio) reveals a part optimized for tensor-heavy workloads, something the AMD GPU does not offer at all. The B300's 76.99 TFLOPS FP32 and 1,231.8 TFLOPS FP16 place it in a performance class that the Steam Machine GPU cannot approach.

The data shows a clean split: the AMD Steam Machine GPU wins in the consumer console segment where power efficiency, display outputs, and compact dimensions matter, and the NVIDIA B300 wins in the server segment where memory capacity, tensor throughput, and raw FP32 compute dominate.

The Verdict

The verdict follows directly from the specification differences. The AMD Steam Machine GPU is the correct choice for the Valve console platform it was built for. Its 6 nm process node, 13,300 million transistors, and 204 mm² die size produce a balanced part within a 110 W envelope. The active production status and release date of 2026-06-28 indicate a current-generation product. The absence of any power connectors simplifies installation in a console chassis.

The NVIDIA B300 is the correct choice for server deployments. Its 1400 W TDP and 1800 W suggested PSU reveal a part that assumes a data-center power infrastructure. The SXM Module slot width confirms a mounting format incompatible with consumer cases. The absence of display outputs means it is not intended to drive monitors. The 144 GB HBM3e memory capacity and 4.10 TB/s bandwidth target large-model inference and training workloads. The predecessor and successor entries, Server Hopper and Server Rubin respectively, place it in a defined server product line.

There is no overlap in intended use. A user building a Steam Machine console cannot use the B300, which lacks display outputs and requires a 1400 W power budget. A data-center operator cannot use the Steam Machine GPU, which lacks the memory capacity and tensor cores needed for modern AI workloads. The percentileVsAllGpus field shows both at the 50th percentile, but that figure reflects the absence of benchmark data rather than comparable performance.

Head-to-Head Benchmarks

The headToHeadBenchmarks array is empty, and both winsA and winsB are zero. The database contains no direct comparisons between these two parts. The analysis therefore relies on the specification differences, which are substantial.

The largest gap appears in memory bandwidth. The NVIDIA B300 delivers 4.10 TB/s while the AMD Steam Machine GPU delivers 288.0 GB/s. The B300's 4096-bit bus versus the Steam Machine GPU's 128-bit bus explains this gap. Memory capacity differs by a factor of eighteen, with 144 GB versus 8 GB. Any workload that exceeds 8 GB of working set cannot run on the AMD part at all.

Compute throughput shows a similar divide. The FP32 figures are 76.99 TFLOPS versus 17.56 TFLOPS, a 4.4x difference. The FP16 figures are even more lopsided: 1,231.8 TFLOPS versus 17.56 TFLOPS. The B300's 592 tensor cores versus the Steam Machine GPU's null tensor core count means the AMD part has no hardware path for the tensor operations that dominate server workloads.

Texture and pixel throughput tell a different story. The B300's texture rate of 1,202.9 GTexel/s exceeds the Steam Machine GPU's 274.4 GTexel/s. But the B300's pixel rate of 48.77 GPixel/s falls below the Steam Machine GPU's 156.8 GPixel/s. The B300's 24 ROPs versus the Steam Machine GPU's 64 ROPs explains this inversion. The AMD part dedicates more silicon to pixel output, consistent with a gaming-oriented design.

Clock speeds favor the AMD part. The Steam Machine GPU boosts to 2450 MHz with a game clock of 2250 MHz and a base of 1720 MHz. The B300 boosts to 2032 MHz with a base of 1665 MHz. The Steam Machine GPU also runs its memory at 2250 MHz (18 Gbps effective) versus the B300's 2000 MHz (8 Gbps effective). The AMD part achieves higher clock rates on a smaller, simpler chip.

Shading units and TMUs heavily favor the B300. The B300 has 18,944 shading units and 592 TMUs versus the Steam Machine GPU's 1,792 shading units and 112 TMUs. These counts align with the FP32 throughput difference.

FAQ

Q: Which GPU has higher FP32 compute performance?

A: The NVIDIA B300 delivers 76.99 TFLOPS compared to the AMD Steam Machine GPU's 17.56 TFLOPS, a 4.4x advantage.

Q: Which GPU has more memory bandwidth?

A: The NVIDIA B300 provides 4.10 TB/s from 144 GB of HBM3e on a 4096-bit bus, versus the AMD Steam Machine GPU's 288.0 GB/s from 8 GB of GDDR6 on a 128-bit bus.

Q: Does the AMD Steam Machine GPU support tensor operations?

A: No. The tensorCores field for the AMD Steam Machine GPU is null, while the NVIDIA B300 has 592 tensor cores.

Q: Which GPU has display outputs?

A: Only the AMD Steam Machine GPU has display outputs, specifically 1x HDMI 2.1a and 1x DisplayPort 2.1. The NVIDIA B300 has no outputs.

Q: What is the power consumption difference?

A: The AMD Steam Machine GPU draws 110 W with no external power connectors, while the NVIDIA B300 draws 1400 W and requires an 1800 W suggested PSU.

Q: Which GPU has a higher pixel fill rate?

A: The AMD Steam Machine GPU achieves 156.8 GPixel/s versus the NVIDIA B300's 48.77 GPixel/s, due to the AMD part's 64 ROPs compared to the B300's 24 ROPs.

Architecture Differences

The two GPUs share a foundry but little else. Both use TSMC fabrication, but the AMD Steam Machine GPU uses a 6 nm process while the NVIDIA B300 uses a 5 nm process. The transistor counts differ by nearly an order of magnitude: 13,300 million for the AMD part versus 104,000 million for the B300. The AMD part has a recorded die size of 204 mm² and a transistor density of 65.2M per mm², while the B300 has no die size or density recorded.

The AMD Steam Machine GPU uses the RDNA 3.0 architecture with the Navi 33 chip and the codename Hotpink Bonefish. Its generation is listed as Console GPU (Valve). The NVIDIA B300 uses the Blackwell Ultra architecture with the GB110 chip and no codename. Its generation is Server Blackwell (Bxx).

The AMD part includes 28 ray tracing cores and supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The NVIDIA B300 lists no ray tracing cores, no DirectX version, no OpenGL version, and no Vulkan version. The B300 instead carries 592 tensor cores, which the AMD part lacks entirely. The API support difference reflects the target markets: gaming APIs for the console part, compute frameworks for the server part.

The memory architectures are fundamentally different. The AMD Steam Machine GPU uses 8 GB of GDDR6 with a 128-bit bus. The NVIDIA B300 uses 144 GB of HBM3e with a 4096-bit bus. The B300's memory clock runs at 2000 MHz with 8 Gbps effective, while the AMD part runs at 2250 MHz with 18 Gbps effective. The B300's wider bus compensates for its lower per-pin rate, producing 4.10 TB/s versus 288.0 GB/s.

The NVIDIA B300 uses a PCIe 5.0 x16 bus interface. The AMD Steam Machine GPU has no bus interface recorded, consistent with a soldered console part. The B300's SXM Module slot width confirms a server form factor. The AMD part's dimensions, 156 mm by 152 mm by 162 mm, define a self-contained module.

Specification Differences

The following fields differ between the two parts:

• Chip: Navi 33 versus GB110
• Architecture: RDNA 3.0 versus Blackwell Ultra
• Codename: Hotpink Bonefish versus null
• Generation: Console GPU (Valve) versus Server Blackwell (Bxx)
• Process node: 6 nm versus 5 nm
• Transistors: 13,300 million versus 104,000 million
• Die size: 204 mm² versus null
• Transistor density: 65.2M per mm² versus null
• Base clock: 1720 MHz versus 1665 MHz
• Boost clock: 2450 MHz versus 2032 MHz
• Game clock: 2250 MHz versus null
• Memory clock: 2250 MHz (18 Gbps effective) versus 2000 MHz (8 Gbps effective)
• Memory size: 8 GB versus 144 GB
• Memory type: GDDR6 versus HBM3e
• Memory bus width: 128 bit versus 4096 bit
• Memory bandwidth: 288.0 GB/s versus 4.10 TB/s
• Shading units: 1,792 versus 18,944
• TMUs: 112 versus 592
• ROPs: 64 versus 24
• Ray tracing cores: 28 versus null
• Tensor cores: null versus 592
• Pixel rate: 156.8 GPixel/s versus 48.77 GPixel/s
• Texture rate: 274.4 GTexel/s versus 1,202.9 GTexel/s
• FP32: 17.56 TFLOPS versus 76.99 TFLOPS
• FP16: 17.56 TFLOPS (1:1) versus 1,231.8 TFLOPS (16:1)
• TDP: 110 W versus 1400 W
• Slot width: null versus SXM Module
• Power connectors: None versus null
• Suggested PSU: null versus 1800 W
• Bus interface: null versus PCIe 5.0 x16
• Display outputs: 1x HDMI 2.1a, 1x DisplayPort 2.1 versus No outputs
• DirectX: 12 Ultimate (12_2) versus null
• OpenGL: 4.6 versus null
• Vulkan: 1.4 versus null
• Dimensions: 156 mm x 152 mm x 162 mm versus null
• Release date: 2026-06-28 versus 2025-09-10
• Predecessor: null versus Server Hopper
• Successor: null versus Server Rubin

The shared fields include manufacturer TSMC as foundry, active production status, and the 50th percentile score. The launch MSRP is absent for both parts, so no pricing information exists in the database. The release dates place the NVIDIA B300 approximately nine months earlier than the AMD Steam Machine GPU.