NVIDIA A10 24GB
Accelerator profile with 24 GB GDDR6, 9,216 CUDA cores, and 31.2 FP32 TFLOPS.
- VRAM
- 24 GB
- CUDA cores
- 9,216
- FP32
- 31.2
- TDP
- 150 W
Benchmarks
Sample benchmark scores are rendered as horizontal ECharts bars with an HTML fallback table.
| Geekbench OpenCL | 142,000 |
|---|---|
| Blender BMW | 48 s |
Plans with this GPU
Sample GPU tariffs using this accelerator, sorted by monthly price in the active currency.
Affiliate link: we may earn a commission if you buy after clicking. Rankings remain editorially independent. Disclosure
| Plan | Provider | GPU | CPU | RAM | Location | Price | Action |
|---|---|---|---|---|---|---|---|
| No sampled plans for this GPU yet. | |||||||
Compare
GPU-to-GPU comparison pages will attach here when comparison content exists.
What NVIDIA A10 24GB Is Good For
NVIDIA A10 24GB is an Ampere GPU accelerator for rented GPU servers where 24 GB VRAM, PCIe 4.0, and a 150 W power envelope matter more than peak training speed. It suits moderate inference, rendering, remote workstation, and batch jobs that would waste larger A100 or L40S nodes.
Buying Signals
- NVIDIA A10 24GB has 24 GB GDDR6 memory. That is enough for many medium GPU jobs, but it is still a hard VRAM ceiling.
- NVIDIA A10 24GB uses PCIe 4.0 x16. That makes it easier to place in ordinary dedicated GPU servers than SXM-class accelerators.
- NVIDIA A10 24GB has a 150 W TDP. Power and cooling should be cheaper than on 350 W or 450 W cards, but the provider may keep the savings.
- NVIDIA A10 24GB has 600 GB/s memory bandwidth. Check data movement before you treat the card as a cheap substitute for HBM2e hardware.
- A low monthly GPU price can hide paid snapshots, weak egress terms, thin CPU allocation, or noisy storage. The GPU model is not the contract.
Hardware Profile
NVIDIA A10 24GB belongs to the Ampere generation. NVIDIA A10 24GB carries 9,216 CUDA cores. NVIDIA A10 24GB carries 288 tensor cores.
NVIDIA A10 24GB delivers 31.2 FP32 TFLOPS in the current hardware profile. NVIDIA A10 24GB delivers 125 FP16 TFLOPS in the same profile.
NVIDIA A10 24GB uses 24 GB of GDDR6 memory. NVIDIA A10 24GB has 600 GB/s of memory bandwidth. NVIDIA A10 24GB uses a PCIe 4.0 x16 interface.
The listed release date is April 12, 2021. HostScout data for this page was refreshed on July 4, 2026.
Where It Fits in GPU Hosting
Choose NVIDIA A10 24GB when the workload needs a real GPU but does not justify a large accelerator bill. Image processing queues, smaller inference services, video-adjacent render jobs, and remote workstation sessions can fit this profile when 24 GB VRAM is enough.
Skip NVIDIA A10 24GB when the workload is constrained by memory bandwidth or model size. A cheap A10 server is a bad bargain if the job spills past 24 GB VRAM, waits on storage IOPS, or pays heavy egress every month.
Do not buy the A10 label alone. A useful GPU server still needs enough CPU cores, RAM, NVMe throughput, bandwidth, backups, DDoS protection, and a clear SLA. If those terms are vague, the monthly price is marketing.
A10 Versus Nearby GPUs
NVIDIA RTX 4090 24GB also has 24 GB VRAM. NVIDIA RTX 4090 24GB has higher CUDA core count, higher memory bandwidth, and a 450 W TDP in the current hardware profile. That profile can be faster, but power density and provider terms matter.
NVIDIA L40S 48GB has 48 GB VRAM. NVIDIA L40S 48GB has higher FP32 throughput, higher FP16 throughput, and 864 GB/s memory bandwidth. It is the cleaner step up when 24 GB VRAM is the first bottleneck.
NVIDIA A100 80GB is a different class of accelerator. NVIDIA A100 80GB has 80 GB HBM2e memory and 2,039 GB/s memory bandwidth. It is the comparison point for heavier ML and HPC jobs, not a cheap A10 replacement.
Practical Verdict
NVIDIA A10 24GB is the sensible rented GPU when you need 24 GB VRAM, moderate compute, and a lower power envelope in one server. It is not the sensible choice when the workload needs 48 GB or 80 GB VRAM, high memory bandwidth, or predictable large-model training capacity.
Normalize every offer into USD per month before comparing A10 plans across US, UK, and global providers. Then price backups, snapshots, egress, migration time, and minimum contract length. Affiliate links may appear on HostScout, but affiliate status does not change the technical verdict.