Entry-level Blackwell desktop GPU for small-model AI experiments and CUDA work.
By provider shows one card per company, placed where its best offer ranked. By configuration lists every offer.
What you can rent comes first: in stock, then waitlist, then not reported, then out of stock. Priced offers rank ahead of quote-only, and on-demand ahead of other billing types.
Within a group, five factors set the order:
We match the provider, its country, GPU model, form factor, billing type, availability and instance name. Matching is partial and case-insensitive. Everyday words work too, so "interruptible" finds spot and "sold out" finds out of stock.
No providers match .
Heads up: A provider's own page may quote a different figure, for example on tax, region or a promotion. A monthly-only plan shows a derived hourly rate. Verify before provisioning. More on how we price.
One RTX 5050 has 8 GB of VRAM. In practice, that's enough memory for roughly 2B parameters at 4-bit, assuming a 32K context. Below are some open-weight LLMs, with the estimated memory and GPUs each one needs.
| Model | Memory (INT4 / FP4) | RTX 5050s needed | Cost /hr | Cost /mo |
|---|---|---|---|---|
Qwen3.8-27B
27B
|
18 GB
|
3
|
–
|
–
|
Gemma 4 31B
30.7B
|
22 GB
|
3
|
–
|
–
|
GPT-OSS-120B
117B · 5.1B active
|
67 GB
|
10
|
–
|
–
|
GLM-5.3-Flash
320B · 18B active
|
178 GB
|
25
|
–
|
–
|
MiniMax-M3
428B · 23B active
|
239 GB
|
34
|
–
|
–
|
DeepSeek V4.1 Flash
552B · 16B active
|
306 GB
|
43
|
–
|
–
|
Kimi K3
2.8T · 104B active
|
1,544 GB
|
215
|
–
|
–
|
Estimates based on the median on-demand rate. Memory is weights plus FP8 KV cache at 32K context per request (FP16/BF16 in the 16-bit column). GPU counts assume 90% of advertised VRAM is usable. No guarantee of runtime support, usable performance, or that a matching quantized build exists. Pricing methodology.
Nvidia RTX 5050 · Per GPU
| Compute · dense | |
|---|---|
| FP4 | 210.5 TFLOPS 421 with sparsity |
| FP8 | 52.6 TFLOPS 105.3 with sparsity |
| FP16 / BF16 | 26.3 TFLOPS 52.6 with sparsity |
| INT8 | 105.3 TOPS 210.5 with sparsity |
| FP32 | 13.2 TFLOPS |
| Precision support | FP4FP8FP16BF16TF32FP32INT8 |
| Memory | |
|---|---|
| Capacity | 8 GB GDDR6 |
| Bandwidth | 320 GB/s |
| Bus width | 128-bit |
| ECC | No |
| Silicon | |
|---|---|
| Architecture | Blackwell |
| Shader cores | 2,560 CUDA cores |
| Matrix cores | 80 Tensor cores |
| Compute units | 20 SMs |
| Fabric and host | |
|---|---|
| Host interface | PCIe 5.0 |
| Power | |
|---|---|
| Board power | 130 W |
| Cooling | Active |
Source: official Nvidia RTX 5050 datasheet.
As of November 13, 2026, we track 4 configs from 1 provider. Prices are per GPU per hour.
| Billing type | Configs | Cheapest |
|---|---|---|
On-demand | 1 | $0.07 (in stock, Salad) |
Spot | 3 | $0.03 (not verified in stock, Salad) |
No median for on-demand and spot: we only show this when at least 3 providers list the GPU on that billing type.
The cheapest verified in-stock estimate is $50 per month on-demand, $29 spot.
GetDeploying currently tracks RTX 5050 configs from 1 provider. Salad has verified in-stock on-demand configs. See the full price comparison above for every provider and config.
One RTX 5050 runs models up to roughly 2B parameters at 4-bit quantization, assuming a 32K context. Larger models run across multiple GPUs: the model table above shows the estimated memory and GPU count for popular open-weight LLMs.
Blackwell tensor cores with FP8 and FP4 support in a low-power desktop card. Suited to testing quantized small models, CUDA development and light image generation on current-generation silicon.
8GB of VRAM fits a 7B model only when quantized to 8 or 4 bits, with little room left for context. For larger models or FP16 inference, the RTX 3060 offers 12GB. Subject to Nvidia's consumer EULA, which may restrict data center use.
Nvidia RTX 5060
8 GB GDDR7 · 40% more bandwidth than the RTX 5050
Nvidia RTX 4060
8 GB GDDR6 · 15% less bandwidth than the RTX 5050
Nvidia RTX 3050
8 GB GDDR6 · 30% less bandwidth than the RTX 5050
Last updated
