EmbeddingDense0.44B params512 max context

GTE Large (1024-dim) VRAM Requirements: How Many GPUs & What It Costs

GTE Large (1024-dim) needs 0.9 GB of VRAM for its weights at FP16 — 1.3 GB once a KV cache is loaded. Below is the full breakdown: which GPUs fit, how many you need, what it serves, and what it costs per million tokens. Every figure comes from tokcalc's open-source capacity model.

Weights (FP16)
0.9 GB
1.3 GB with KV cache
Fits on 1 GPU?
Yes
RTX 3080 10GB or larger
Cheapest config
$0.008247
1× RTX 3080 10GB / 1M tokens
Fastest config
177273
tok/s on 1× Cerebras CS-3 (WSE-3)
VRAM requirement by quantization
Weight footprint = params × bytes per param. The “with KV cache” column adds one request at 4K context.
QuantizationWeightsWith KV cacheSmallest single GPU
FP16recommended0.9 GB1.3 GBRTX 3080 10GB (10 GB)
FP8 (E4M3)0.4 GB0.8 GBRTX 3080 10GB (10 GB)
AWQ 4-bit0.2 GB0.6 GBRTX 3080 10GB (10 GB)
Which GPU should you use?
Reference workload: 8K context, 8 concurrent users, continuous batching, FP16. Ranked by fitting first, then by tokens per dollar-hour.
GPUGPUsTotal VRAMDecode tok/sAggregate tok/sCost / 1M tokens
RTX 3080 10GB1×10 GB561.46736$0.008247
RTX 3090 24GB1×24 GB691.48296$0.008370
RTX 5090 32GB1×32 GB132415884$0.008744
RTX 4090 24GB1×24 GB744.58935$0.0124
RTX 4080 16GB1×16 GB529.66355$0.0131
AMD MI300X 192GB1×192 GB391546977$0.0148
V100 SXM2 32GB1×32 GB664.87977$0.0174
A100 40GB SXM41×40 GB114513739$0.0182
AMD MI325X 288GB1×288 GB443253182$0.0183
A100 80GB SXM41×80 GB147717727$0.0188
A10G 24GB1×24 GB443.25318$0.0188
Intel Gaudi 3 128GB1×128 GB243829250$0.0190
How concurrency changes with context length
On 1× RTX 3080 10GB at FP16. KV cache scales linearly with context, so long contexts cut the number of users you can serve from the same GPUs.
ContextConcurrent usersKV per requestFits?
4K22412.3 MB✓ fits
8K11824.6 MB✓ fits
33K23.22 GB✓ fits
66K16.44 GB✓ fits
131K012.88 GB✗ does not fit
Frequently asked questions

Can GTE Large (1024-dim) run on a single GPU?

Yes. At FP16 the model needs 1.3 GB of VRAM, so it fits on a single RTX 3080 10GB or anything with more VRAM. At full FP16 it needs 1.3 GB, which fits on one RTX 3080 10GB or larger.

How many GPUs does GTE Large (1024-dim) need?

At FP16 with 8,192-token context and 8 concurrent users, the cheapest configuration that fits is 1× RTX 3080 10GB (10 GB total VRAM). $0.008247 per 1M output tokens.

How fast can GTE Large (1024-dim) be served?

On 1× Cerebras CS-3 (WSE-3) at FP16, the aggregate throughput is 177273 tokens/sec with continuous batching. That hardware has no published rental price, so its cost per million tokens is not comparable.

How much VRAM does GTE Large (1024-dim) need?

Weights alone: 0.9 GB at FP16, 0.9 GB at FP16. Add KV cache for your context length and concurrency — every concurrent request at 8,192 tokens costs extra VRAM, so budget headroom above the weight figure.

Is it cheaper to self-host GTE Large (1024-dim) than to call an API?

At 1× RTX 3080 10GB the marginal cost is $0.008247 per 1M output tokens, before you add idle capacity, ops staff, or egress. That is typically cheaper than frontier API pricing at sustained volume, but API calls win when your load is bursty or too small to keep a GPU busy. The Build vs Buy tab on tokcalc prices this against a specific provider for you.

Try GTE Large (1024-dim) with your own context and concurrency

The calculator uses the exact same formulas as this page — change the workload and watch every number update.

Open in calculator

GTE Large (1024-dim) on specific GPUs

GTE Large (1024-dim) on RTX 3080 10GBGTE Large (1024-dim) on RTX 3090 24GBGTE Large (1024-dim) on RTX 5090 32GBGTE Large (1024-dim) on RTX 4090 24GBGTE Large (1024-dim) on RTX 4080 16GBGTE Large (1024-dim) on AMD MI300X 192GBGTE Large (1024-dim) on V100 SXM2 32GBGTE Large (1024-dim) on A100 40GB SXM4
Model architecture
The numbers behind the KV-cache and throughput math.
Total parameters0.44B
Active parameters per token0.44B
ArchitectureDense
Attention layoutFull attention (GQA)
Transformer layers24
Hidden dimension1024
Attention heads16 query / 16 KV (GQA)
Head dimension64
Vocabulary size30,522
Max context (catalog)512
KV cache per token @ 8K96.0 KB

Every layer caches keys and values for the whole context, so the KV cache grows linearly with conversation length.