Use the filters below to narrow your results.
page:
model:
quant:
engine:
(Click arrow to add another value)
engine version:
cards:
card count:
topology:
(Click arrow to add another value)
single card ·
engine-managed expert tiering across VRAM, host RAM and NVMe ·
engine-managed split with expert tiering across VRAM, RAM and NVMe ·
layer split (no tensor parallel), no P2P ·
PP2 (layer partition 25,23), no P2P ·
PP2 with PLE CPU offload ·
PP2, MTP 3, no P2P ·
PP2, MTP depth 3 ·
PP3 ·
PP3, no P2P (works at PP2 on two cards) ·
PP8 (also PP9 and PP10), TP1 ·
PP8, partition 5,5,5,5,5,5,5,5 ·
PP8, TP1 ·
TP2 over NCCL (GPU P2P), DFlash2 K7 plus hybrid drafting ·
TP2 with expert parallel ·
TP2 with whole-expert expert parallel ·
TP2 x PP3, partition 14,14,12 ·
TP4 ·
TP4 (default) or PP4 ·
TP4 (default) or PP4 for x1/x4 links ·
TP4 (P2P on); expert parallel off ·
TP4 with expert parallel (96 whole experts per card) ·
TP4 with expert parallel, MTP 3 ·
TP4 x PP2, partition 20,20 ·
TP4, P2P on ·
TP4, peer access off (NCCL_P2P_DISABLE=1), host all-reduce ·
TP5 across four CMP 170HX and one RTX PRO 6000 Blackwell (native backend, reduces through host memory); MTP draft on the PRO ·
TP6 (virtual-head padding), PP1 ·
Other ·
None
context:
(Click arrow to add another value)
262,144 ·
1,048,576 ·
None ·
1,000,000 (YaRN x4 from 262,144) ·
1,048,576 (YaRN x4) ·
1,048,576 FP16 ·
1,048,576 per request ·
262,144 (256K in the post) ·
262,144 (up to 1,048,576) ·
262,144 (weights address 1,048,576) ·
262,144 advertised (tested through 32K) ·
262,144 native (YaRN to about 500k in the recipe's long variant) ·
392,960 (profile q8_384k) ·
392,960 per request; 655,360-token Q8 KV pool shared by two requests ·
about 1,000,000 (8-bit KV cache) ·
Other
decode tps:
prefill tps:
vram per card:
power cap:
driver:
unlock build:
status:
tested by:
tested on:
source:
Showing below up to 1 result in range #1 to #1.
& topology:
TP4 with expert parallel (96 whole experts per card)