BTC
ETH
USDT
XMR
LTC
GPU comparison · Blackwell vs Blackwell Ultra
NVIDIA B200 vs B300.
Same Blackwell design and NVLink 5; the B300 (Blackwell Ultra) raises memory from 180 GB to 288 GB per GPU and FP4 throughput by half, and gives up almost all FP64. It costs $580 more per GPU per month, but less per GB of memory.
- 180 vs 288 GBGPU memory
- 8 TB/s eachmemory bandwidth
- $3,439 vs $4,019per GPU, per month
B200 vs B300: which one to rent.
Choose the B200 if
- Your model and context fit in 180 GB per GPU (1.4 TB in an 8-GPU server).
- You need FP64: 37 TFLOPS on the B200 against 1.25 on the B300.
- You want the lower monthly price: $3,439 per GPU against $4,019.
Choose the B300 if
- You need 288 GB per GPU: 2.3 TB in an 8-GPU server for the largest open models and long contexts.
- You run FP4 inference: 13.5 PFLOPS dense per GPU against 9.
- You want the lowest price per GB of HBM: $13.95 against $19.11 a month.
B200 vs B300 specs, side by side.
Figures from NVIDIA’s published specifications, per GPU. Tensor figures are peak dense throughput.
| Per GPU | NVIDIA B200 | NVIDIA B300 | B300 / B200 |
|---|---|---|---|
| Architecture | Blackwell, TSMC 4NP | Blackwell Ultra, TSMC 4NP | — |
| Form factor | SXM module (HGX B200) | SXM module (HGX B300) | — |
| GPU memory | 180 GB HBM3e | 288 GB HBM3E | 1.6× |
| Memory bandwidth | Up to 8 TB/s | Up to 8 TB/s | Same |
| GPU-to-GPU link | NVLink 5, 1.8 TB/s per GPU | NVLink 5, 1.8 TB/s per GPU | Same |
| FP4 tensor (dense) | 9.0 PFLOPS | 13.5 PFLOPS | 1.5× |
| FP8 tensor (dense) | 4,500 TFLOPS | 4,500 TFLOPS | Same |
| FP16 / BF16 tensor (dense) | 2,250 TFLOPS | 2,250 TFLOPS | Same |
| FP64 | 37 TFLOPS | 1.25 TFLOPS | 0.03× |
| Max power | Up to 1,000 W, configurable | Not published | — |
| GPUs per server | 1, 2, 4, 8 | 1, 2, 4, 8 | — |
| Per GPU in our servers | 28 vCPU · 288 GB RAM · 3.84 TB | 32 vCPU · 384 GB RAM · 3.84 TB | — |
Sources: NVIDIA HGX · NVIDIA DGX B200 · NVIDIA HGX · Blackwell Ultra technical blog. Full sheets: B200 · B300
B200 vs B300 rental price.
Our monthly prices, set 30% below the market median of public on-demand prices and paid in crypto. Hourly figures are the monthly price divided by 730 hours.
| CryptGPU | NVIDIA B200 | NVIDIA B300 | B300 / B200 |
|---|---|---|---|
| Price per GPU, per month | $3,439 | $4,019 | 1.17× |
| Equivalent per GPU-hour | $4.71 | $5.51 | — |
| Per GB of GPU memory, per month | $19.11 | $13.95 | 0.73× |
| Per PFLOPS of dense FP8, per month | $764 | $893 | 1.17× |
| Largest server | 8× · $27,512/mo | 8× · $32,152/mo | — |
| Market median, per GPU | $4,920 | $5,745 | — |
| Below the median | −30% | −30% | — |
Same price per GPU at every server size, no hourly metering. How we compare prices · Configure an B200 server · Configure a B300 server
Which models fit on each.
Smallest server that holds each model, by precision. Rule of thumb with headroom for the KV cache, 92% of GPU memory usable.
| Model (total parameters) | B200 · FP8 | B300 · FP8 | B200 · 4-bit | B300 · 4-bit |
|---|---|---|---|---|
| Qwen3.5-9B | 1× | 1× | 1× | 1× |
| Gemma 4 31B | 1× | 1× | 1× | 1× |
| Llama 3.3 70B | 1× | 1× | 1× | 1× |
| gpt-oss-120b | 1× | 1× | 1× | 1× |
| DeepSeek-V4-Flash | 4× | 2× | 2× | 1× |
| Qwen3.5-397B-A17B | 4× | 2× | 2× | 1× |
| DeepSeek-R1 (671B) | 8× | 4× | 4× | 2× |
| Kimi K2.6 (1T) | 8× | 8× | 4× | 4× |
FP8 ≈ 1.2 bytes and 4-bit ≈ 0.65 bytes per parameter. — = larger than the biggest server of that GPU. Size another model · How much VRAM does an LLM need?
B200 vs B300 FAQ.
More on each GPU: NVIDIA B200 · NVIDIA B300.
Is the B300 faster than the B200?
For FP4 it has 1.5× the dense throughput. NVIDIA lists the same dense FP8 and BF16 rates and the same 8 TB/s bandwidth for both, so the B300 mostly wins through its larger memory and FP4.
Which one for FP64 simulation?
The B200. Blackwell Ultra drops most FP64 hardware: 1.25 TFLOPS against 37 on the B200. For heavy FP64, look at the AMD MI355X or the H200.
What fits in an 8× B300 server?
2,304 GB of GPU memory: by our rule of thumb, a 671B model in BF16 or a 1T-parameter model in FP8 fits in one server.
Other GPU comparisons.
Rent the B200 or the B300.
Dedicated servers with 1 to 8 GPUs, one monthly price, paid in crypto. Online in under 10 minutes, no KYC.