selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

Vxtzq logo

Can the RTX 4070 run Crowd-v1?

Crowd-v1·1B·GeForce RTX 4070·12 GB VRAM

Saved in this browser, used on every model page.
Yes

Yes. The RTX 4070 runs Crowd-v1 fully in VRAM at FP16 / BF16 (about 2.8 GB of 12 GB), at roughly 139 tokens/s.

best quant
FP16 / BF16
needs
2.8 GB
usable
12 GB
est. speed
~139 tok/s

Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the Crowd-v1 page →

Every quantisation of Crowd-v1 on the RTX 4070

QuantQualityMemoryFitSpeed
Q2_K Heavy loss1.2 GBfits~843 tok/s
Q3_K_M Noticeable loss1.3 GBfits~567 tok/s
Q4_K_M Small loss1.4 GBfits~457 tok/s
Q5_K_M Near-lossless1.6 GBfits~390 tok/s
Q6_K Near-lossless1.7 GBfits~337 tok/s
Q8_0 Lossless1.9 GBfits~261 tok/s
FP16 / BF16 pickLossless2.8 GBfits~139 tok/s

Speed is a bandwidth-based estimate for generating text (504 GB/s); prompt processing and long contexts are slower.

Measured speed on the RTX 4070

No community reports for this pair yet.

Report your own speed →

Chat models that fit the RTX 4070 comfortably

All that fit →
ModelParamsBest quantMemorySpeedLicence
1BQ8_0≈1.8 GBFits in memory~261 tok/sConditions
752MQ8_0≈2.0 GBFits in memory~347 tok/sCommercial OK
1.2BQ8_0≈2.0 GBFits in memory~211 tok/sConditions
1.1BQ8_0≈1.7 GBFits in memory~237 tok/sCommercial OK
LiquidAI logoLFM3-1.7BLiquidAI
1.7BQ8_0≈2.7 GBFits in memory~151 tok/sConditions
1.5BQ8_0≈2.6 GBFits in memory~169 tok/sCommercial OK

Other GPUs that run Crowd-v1

Frequently asked questions

Can the RTX 4070 run Crowd-v1?
Yes. The RTX 4070 runs Crowd-v1 fully in VRAM at FP16 / BF16 (about 2.8 GB of 12 GB), at roughly 139 tokens/s.
How fast does Crowd-v1 run on the RTX 4070?
An estimate from memory bandwidth gives about 139 tokens/s at FP16 / BF16.
Which quantisation of Crowd-v1 should I use on the RTX 4070?
FP16 / BF16: the largest one that fits. Expected quality: lossless compared with the full-precision weights.

Memory and speed are estimates from the model's size and the card's memory bandwidth; real numbers depend on the runtime, drivers and context length. Community figures are medians of user reports.