SAVRN Model Hub · Cortex-ai
Cortex-ai GPU Requirements
Cortex-ai needs about 4.0 TB of GPU memory at 16-bit, 2.0 TB at 8-bit and 990 GB at 4-bit: its 1.7T parameters plus a 20% working margin. 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 Cortex-ai, and what that many cards cost an hour at the lowest listed on-demand price. Memory needed: 4.0 TB at 16-bit, 2.0 TB at 8-bit, 990 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 | More than 8 | More than 8 |
| H200 GMI Cloud |
141 GB | $2.60 | More than 8 | More than 8 | 8 cards $20.80/hr |
| B200 Vultr |
180 GB | $3.50 | More than 8 | More than 8 | 6 cards $21.00/hr |
| GB200 NVL72 GMI Cloud |
186 GB | $8.00 | More than 8 | More than 8 | 6 cards $48.00/hr |
| MI300X Vultr |
192 GB | $1.85 | More than 8 | More than 8 | 6 cards $11.10/hr |
| MI325X Vultr |
256 GB | $2.00 | More than 8 | 8 cards $16.00/hr |
4 cards $8.00/hr |
| B300 Massed Compute |
268 GB | $6.60 | More than 8 | 8 cards $52.80/hr |
4 cards $26.40/hr |
| GB300 NVL72 Verda |
279 GB | $10.32 | More than 8 | 8 cards $82.56/hr |
4 cards $41.28/hr |
| MI355X Vultr |
288 GB | $2.59 | More than 8 | 7 cards $18.13/hr |
4 cards $10.36/hr |
Running It Around the Clock
| Precision | Cheapest setup | Per hour | Per month (730 hours) |
|---|---|---|---|
| 8-bit | 8x MI325X (Vultr) | $16.00 | $11,680 |
| 4-bit | 4x MI325X (Vultr) | $8.00 | $5,840 |
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 Cortex-ai need?
About 4.0 TB at 16-bit; about 2.0 TB at 8-bit; about 990 GB at 4-bit: the weights plus 20% for the runtime and a short context. A long context needs more.
Can Cortex-ai run on a single H100?
No. An H100 has 80 GB, and Cortex-ai needs 990 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 Oct 7, 2026. Setups beyond eight cards, one server, are not listed.