Rubin vs H200 SXM: specifications compared
Comparison of Rubin and H200 SXM based on data from the manufacturer or provider.
According to the manufacturer, Rubin and H200 SXM differ mainly in memory bandwidth (Rubin: 22,000 GB/s, H200 SXM: 4,800 GB/s) and memory (Rubin: 288 GB, H200 SXM: 141 GB). The architecture also differs: Rubin versus Hopper. In the Clouwino points, which are calculated from the specifications, Rubin gets 92 and H200 SXM gets 45. Checked on 19 October 2026, no prices.
How are the points calculated?
Clouwino points are an arithmetic summary of the specifications the manufacturer states, on a scale of 0 to 100. It is not a benchmark and not a judgement on quality or suitability: a product with fewer points may suit your situation better. Products are compared with other products of the same type. For each property we count how far a product sits between the lowest and highest value in the group:
- FP32 compute (for: consumer), carries heavy weight
- Memory (for: consumer), carries heavy weight
- Memory bandwidth (for: consumer), carries heavy weight
- Compute units (for: consumer), carries light weight
- Boost clock (for: consumer), carries light weight
- Memory (for: professional and data centre), carries medium weight
- Memory bandwidth (for: professional and data centre), carries medium weight
- FP16/BF16 (for: professional and data centre), carries heavy weight
- FP8 (for: professional and data centre), carries medium weight
- FP64 (for: professional and data centre), carries medium weight
If too much data is missing, we do not show points. Points are relative: when a new product is added, they can shift. Read more at how we compare.
Rubin and H200 SXM side by side
| Specificatie | NVIDIA Rubin (per GPU) | NVIDIA H200 SXM |
|---|---|---|
| Overview | ||
| Clouwino points | 92 punten | 45 punten |
| General | ||
| Architecture | Rubin | Hopper |
| Manufacturing process | Not stated by the manufacturer | Not stated by the manufacturer |
| Form | Not stated by the manufacturer | SXM |
| Launch | Not stated by the manufacturer | 2023-11 |
| Compute | ||
| FP32 compute | 130 TFLOPS | 67 TFLOPS |
| FP64 | 33 TFLOPS | 34 TFLOPS |
| FP16/BF16 | 4.000 TFLOPS | 1.979 TFLOPS |
| FP8 | 17.500 TFLOPS | 3.958 TFLOPS |
| FP4 | 35.000 TFLOPS | Not stated by the manufacturer |
| Performance notes | FP16/BF16 4 PFLOPS (4000 TFLOPS) DENSE; FP8/FP6 training 17.5 PFLOPS (17500 TFLOPS) DENSE; FP4 = NVFP4 training 35 PFLOPS (35000 TFLOPS) DENSE. NVFP4 inference 50 PFLOPS (50000 TFLOPS) is SPARSE. TF32 2 PFLOPS dense, INT8 250 TOPS dense, FP32 SGEMM 400 TFLOPS and FP64 DGEMM 200 TFLOPS dense. FP32 130 and FP64 33 TFLOPS without a label. Values converted from PFLOPS to TFLOPS (unit only, no recalculation). | FP64 34 TFLOPS (vector; Tensor Core 67) and FP32 67 TFLOPS: no mention of sparsity. FP16/BF16 1,979 and FP8 3,958 TFLOPS: WITH SPARSITY (footnote 'Values are with sparsity'); dense not on the page and not calculated. FP4 not stated. |
| Memory | ||
| Memory | 288 GB | 141 GB |
| Memory type | HBM4 | HBM3e |
| Memory bandwidth | 22 TB/s | 4,8 TB/s |
| Power and connections | ||
| Power (board power or TDP) | Not stated by the manufacturer | 700 W |
| GPU interconnect | NVLink 6: 3,6 TB/s | NVLink 900 GB/s; PCIe Gen5 128 GB/s |
| Multiple instances | Not stated by the manufacturer | MIG: up to 7 instances of 18 GB |
Note: Figures per GPU from the HGX page. CONTRADICTORY: the Vera Rubin NVL72 page states 19.2 TB/s memory bandwidth and 3 TB/s NVLink; the HGX page and NVIDIA developer blog (21 July 2026) state 22 TB/s and 3.6 TB/s. The HGX/blog values are used here. TDP, process node, form factor and launch date not published (NVIDIA: 'ramping into full production'). No separate datasheet PDF was opened. A100 not included (space reserved for Rubin).
Note: Figures per GPU (SXM column). NVIDIA marks the specifications as 'preliminary, may change'. Announced 13 Nov 2023 (SC23), available from Q2 2024. Process node not stated on the page. H200 NVL (PCIe): up to 600 W, 141 GB, 4.8 TB/s.
Main differences between Rubin and H200 SXM
These are the differences the manufacturer lists in the specifications. The full table is further down.
- Memory bandwidth: Rubin 22,000 GB/s, H200 SXM 4,800 GB/s. Rubin lists about 4.6 times as much.
- Memory: Rubin 288 GB, H200 SXM 141 GB. Rubin lists about 2 times as much.
- Architecture: Rubin Rubin, H200 SXM Hopper.
- Memory type: Rubin HBM4, H200 SXM HBM3e.
What should you look at when choosing?
A higher number is not better for everyone. It depends on what you want to use it for.
- Memory bandwidth. Higher bandwidth means data can move to and from memory faster.
- Memory. Memory determines how much data the GPU can hold at once, for example large models, datasets or textures.
Also look at the rest of your system, such as power supply, cooling and software. Real-world performance depends on more than the specifications alone.
What do the Clouwino points say?
Rubin gets 92 points and H200 SXM gets 45 points. The points are calculated from the manufacturer’s specifications, within the same kind of products. They are not benchmark scores and say nothing about price or availability. Read how we calculate them at How we compare.
Frequently asked questions
What is the difference between Rubin and H200 SXM?
According to the manufacturer, Rubin and H200 SXM differ mainly in memory bandwidth (Rubin: 22,000 GB/s, H200 SXM: 4,800 GB/s) and memory (Rubin: 288 GB, H200 SXM: 141 GB). The architecture also differs: Rubin versus Hopper. In the Clouwino points, which are calculated from the specifications, Rubin gets 92 and H200 SXM gets 45.
Which one has the highest memory bandwidth?
According to the manufacturer, Rubin lists the highest memory bandwidth: 22,000 GB/s versus 4,800 GB/s.
What do Rubin and H200 SXM cost?
Clouwino does not show prices. You can see the current price and availability with the partner, via the button on the product page.
Where does the data come from?
The specifications come from the manufacturers' official pages and were checked on 19 October 2026. The sources are at the bottom of this page.
Last updated on
