SAVRN Model Hub · DeepSeek-V3.2-NVFP4-mixed-CT-AutoRound
DeepSeek-V3.2-NVFP4-mixed-CT-AutoRound GPU Requirements
DeepSeek-V3.2-NVFP4-mixed-CT-AutoRound needs about 927 GB of GPU memory at 16-bit, 464 GB at 8-bit and 232 GB at 4-bit: its 386.3B parameters plus a 20% working margin. The cheapest setup at 16-bit is 4x MI325X from $8.00 an hour, about $5,840 a month around the clock. None of the 9 accelerators the SAVRN Index prices holds it on one card at 16-bit.
Every Accelerator, Every Precision
How many cards of each accelerator the SAVRN Index prices it takes to hold DeepSeek-V3.2-NVFP4-mixed-CT-AutoRound, and what that many cards cost an hour at the lowest listed on-demand price. Memory needed: 927 GB at 16-bit, 464 GB at 8-bit, 232 GB at 4-bit.
| Accelerator | Memory per card | Lowest price per card | 16-bit | 8-bit | 4-bit |
|---|---|---|---|---|---|
| H100 Voltage Park |
80 GB | $1.99 | More than 8 | 6 cards $11.94/hr |
3 cards $5.97/hr |
| H200 GMI Cloud |
141 GB | $2.60 | 7 cards $18.20/hr |
4 cards $10.40/hr |
2 cards $5.20/hr |
| B200 Vultr |
180 GB | $3.50 | 6 cards $21.00/hr |
3 cards $10.50/hr |
2 cards $7.00/hr |
| GB200 NVL72 GMI Cloud |
186 GB | $8.00 | 5 cards $40.00/hr |
3 cards $24.00/hr |
2 cards $16.00/hr |
| MI300X Vultr |
192 GB | $1.85 | 5 cards $9.25/hr |
3 cards $5.55/hr |
2 cards $3.70/hr |
| MI325X Vultr |
256 GB | $2.00 | 4 cards $8.00/hr |
2 cards $4.00/hr |
1 card $2.00/hr |
| B300 Massed Compute |
268 GB | $6.60 | 4 cards $26.40/hr |
2 cards $13.20/hr |
1 card $6.60/hr |
| GB300 NVL72 Verda |
279 GB | $8.89 | 4 cards $35.56/hr |
2 cards $17.78/hr |
1 card $8.89/hr |
| MI355X Vultr |
288 GB | $2.59 | 4 cards $10.36/hr |
2 cards $5.18/hr |
1 card $2.59/hr |
Running It Around the Clock
| Precision | Cheapest setup | Per hour | Per month (730 hours) |
|---|---|---|---|
| 16-bit | 4x MI325X (Vultr) | $8.00 | $5,840 |
| 8-bit | 2x MI325X (Vultr) | $4.00 | $2,920 |
| 4-bit | 1x MI325X (Vultr) | $2.00 | $1,460 |
One copy of the model on rented cards, busy or idle. Serving more users at once takes more copies or more memory for their contexts.
Questions
How much VRAM does DeepSeek-V3.2-NVFP4-mixed-CT-AutoRound need?
About 927 GB at 16-bit; about 464 GB at 8-bit; about 232 GB at 4-bit: the weights plus 20% for the runtime and a short context. A long context needs more.
What is the cheapest GPU setup to run DeepSeek-V3.2-NVFP4-mixed-CT-AutoRound?
At 16-bit, 4x MI325X from $8.00 an hour, at the lowest on-demand price the SAVRN Index lists.
Can DeepSeek-V3.2-NVFP4-mixed-CT-AutoRound run on a single H100?
No. An H100 has 80 GB, and DeepSeek-V3.2-NVFP4-mixed-CT-AutoRound needs 232 GB even at 4-bit, so it takes more than one card.
Memory is the weights at that precision plus 20% for the runtime and a short context. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Sep 18, 2026. Setups beyond eight cards, one server, are not listed.