Models · Fast agent when Strata is up
the house fast agent: Unsloth UD-Q6_K_XL. Two 32 GB cards plus 96 GB DDR5 run Q4 at 256K — see Flash requirements. Strata is the cited Hermes path (~32.6 tok/s); llama.cpp is first on this page (~16 tok/s). Two-card Q6 lives on Flash daily. Untested cells say contribute. Q3 is a labeled estimate, not a shopper pick.
How we test — Measured cells cite official amdai-bench/run@1 cards. Estimates stay labeled. Q3 is estimate-only. Q1/Q2 stay off.
Official card (Qwen3.8-Flash-Next): 125B total / 6B active, plus a 51B n-gram embedding llama.cpp can keep in system RAM. 48 layers, 12 sparse-attention layers (GQA 2×256) plus a 12×256 indexer cache (raw + pooled key). 36 linear layers hold a fixed state. Publisher UD-Q6_K_XL folder lists ~169 GB.
Official card: Qwen/Qwen3.8-Flash-Next · GGUF: unsloth/Qwen3.8-Flash-Next-GGUF. Abliterated derivatives are educational only — read /abliterated, never the default on this page.
Kit sizes are real DIMM steps. Measured peaks come from run cards when we have them; otherwise the line is an estimate. Wherever RAM offload appears, both shopper labels stay on the page.
Fast agent · Flash · 262K
Minimum system RAM needed
192 GB kit
Estimate · 159.9 GiB raw · 256K context
RAM needed at full context
192 GB kit
Estimate · 159.9 GiB raw · 256K full context
Estimate. Offloaded weights + KV in RAM + runtime + OS/desktop + headroom, rounded up to a 192 GB DIMM kit. The local desktop ships 256 GB DDR5; measured host-RAM usage is not recorded until Local AI numbers arrive.
UD-Q6_K_XL at 262K. GPU count was not published on the prior first-party pair. llama.cpp first; Strata is the faster cited agent path.
Strata
32.6 tok/s
Prior first-party Flash UD-Q6_K_XL on the local desktop. Card count was not published; recorded on the two-R9700 host.
llama.cpp
16 tok/s
Same prior first-party pair. llama.cpp first on this site; Strata is the faster cited agent path.
Cited house pair only. Empty cells stay contribute.
Q6 / Q5 / Q4, plus Q3 as a labeled estimate. 1 GPU and 2 GPU.
Decode, prefill, peak VRAM per card, VRAM total, and peak system RAM come from
amdai-bench/run@1 cards when a cell is measured. A null field stays not on the card.
Untested cells say no data — contribute
.
Q1 and Q2 stay off.
| Quant | GPUs | Context | Decode | Prefill | VRAM / card | VRAM total | System RAM | Source |
|---|---|---|---|---|---|---|---|---|
| UD-Q6_K_XL | 1 · single | short | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q6_K_XL | 1 · single | 8K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q6_K_XL | 1 · single | 32K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q6_K_XL | 1 · single | 100K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q6_K_XL | 1 · single | 128K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q6_K_XL | 1 · single | 256K | 16 tok/s (llama.cpp) · 32.6 tok/s (Strata) | not on the card | not on the card | not on the card | not on the card | Flash pair House recipes list card_count 2. Recorded on the two-R9700 local desktop. Treat layout as cited on those recipes, not a new bench. |
| UD-Q6_K_XL | 2 · split | short | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q6_K_XL | 2 · split | 8K | 32.8 tok/s (Strata) | not on the card | not on the card | not on the card | not on the card | Flash daily Two-card UD-Q6_K_XL at 262K-capable launch. llama.cpp DDR5-5200 18.5 tok/s · DDR5-3600 15.9 tok/s. Peak VRAM and RAM were not published on that set. |
| UD-Q6_K_XL | 2 · split | 32K | 32.8 tok/s (Strata) | not on the card | not on the card | not on the card | not on the card | Flash daily Two-card UD-Q6_K_XL at 262K-capable launch. llama.cpp DDR5-5200 18.2 tok/s · DDR5-3600 15.6 tok/s. Peak VRAM and RAM were not published on that set. |
| UD-Q6_K_XL | 2 · split | 100K | 31.8 tok/s (Strata) | not on the card | not on the card | not on the card | not on the card | Flash daily Two-card UD-Q6_K_XL at 262K-capable launch. llama.cpp DDR5-5200 17.9 tok/s · DDR5-3600 15.3 tok/s. Peak VRAM and RAM were not published on that set. |
| UD-Q6_K_XL | 2 · split | 128K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q6_K_XL | 2 · split | 256K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q5_K_XL | 1 · single | short | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q5_K_XL | 1 · single | 8K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q5_K_XL | 1 · single | 32K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q5_K_XL | 1 · single | 100K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q5_K_XL | 1 · single | 128K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q5_K_XL | 1 · single | 256K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q5_K_XL | 2 · split | short | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q5_K_XL | 2 · split | 8K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q5_K_XL | 2 · split | 32K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q5_K_XL | 2 · split | 100K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q5_K_XL | 2 · split | 128K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q5_K_XL | 2 · split | 256K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q4_K_XL | 1 · single | short | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q4_K_XL | 1 · single | 8K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q4_K_XL | 1 · single | 32K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q4_K_XL | 1 · single | 100K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q4_K_XL | 1 · single | 128K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q4_K_XL | 1 · single | 256K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q4_K_XL | 2 · split | short | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q4_K_XL | 2 · split | 8K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q4_K_XL | 2 · split | 32K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q4_K_XL | 2 · split | 100K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q4_K_XL | 2 · split | 128K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q4_K_XL | 2 · split | 256K | no data — contribute | no data — contribute | no data — contribute | no data — contribute | no data — contribute | Contribute |
| UD-Q3_K_XL estimate | 1 · single | short | no data — contribute | no data — contribute | 94.0 GB (estimate) | 94.0 GB (estimate) | 96 GB kit (estimate) | Calculator Labeled Q3 estimate from the calculator + UD-Q3_K_XL listing. Not a local desktop speed run. |
| UD-Q3_K_XL estimate | 1 · single | 8K | no data — contribute | no data — contribute | 94.0 GB (estimate) | 94.0 GB (estimate) | 96 GB kit (estimate) | Calculator Labeled Q3 estimate from the calculator + UD-Q3_K_XL listing. Not a local desktop speed run. |
| UD-Q3_K_XL estimate | 1 · single | 32K | no data — contribute | no data — contribute | 94.0 GB (estimate) | 94.0 GB (estimate) | 96 GB kit (estimate) | Calculator Labeled Q3 estimate from the calculator + UD-Q3_K_XL listing. Not a local desktop speed run. |
| UD-Q3_K_XL estimate | 1 · single | 100K | no data — contribute | no data — contribute | 94.0 GB (estimate) | 94.0 GB (estimate) | 96 GB kit (estimate) | Calculator Labeled Q3 estimate from the calculator + UD-Q3_K_XL listing. Not a local desktop speed run. |
| UD-Q3_K_XL estimate | 1 · single | 128K | no data — contribute | no data — contribute | 94.0 GB (estimate) | 94.0 GB (estimate) | 96 GB kit (estimate) | Calculator Labeled Q3 estimate from the calculator + UD-Q3_K_XL listing. Not a local desktop speed run. |
| UD-Q3_K_XL estimate | 1 · single | 256K | no data — contribute | no data — contribute | 94.0 GB (estimate) | 94.0 GB (estimate) | 96 GB kit (estimate) | Calculator Labeled Q3 estimate from the calculator + UD-Q3_K_XL listing. Not a local desktop speed run. |
| UD-Q3_K_XL estimate | 2 · split | short | no data — contribute | no data — contribute | 47.0 GB (estimate) | 94.0 GB (estimate) | 64 GB kit (estimate) | Calculator Labeled Q3 estimate from the calculator + UD-Q3_K_XL listing. Not a local desktop speed run. |
| UD-Q3_K_XL estimate | 2 · split | 8K | no data — contribute | no data — contribute | 47.0 GB (estimate) | 94.0 GB (estimate) | 64 GB kit (estimate) | Calculator Labeled Q3 estimate from the calculator + UD-Q3_K_XL listing. Not a local desktop speed run. |
| UD-Q3_K_XL estimate | 2 · split | 32K | no data — contribute | no data — contribute | 47.0 GB (estimate) | 94.0 GB (estimate) | 64 GB kit (estimate) | Calculator Labeled Q3 estimate from the calculator + UD-Q3_K_XL listing. Not a local desktop speed run. |
| UD-Q3_K_XL estimate | 2 · split | 100K | no data — contribute | no data — contribute | 47.0 GB (estimate) | 94.0 GB (estimate) | 64 GB kit (estimate) | Calculator Labeled Q3 estimate from the calculator + UD-Q3_K_XL listing. Not a local desktop speed run. |
| UD-Q3_K_XL estimate | 2 · split | 128K | no data — contribute | no data — contribute | 47.0 GB (estimate) | 94.0 GB (estimate) | 64 GB kit (estimate) | Calculator Labeled Q3 estimate from the calculator + UD-Q3_K_XL listing. Not a local desktop speed run. |
| UD-Q3_K_XL estimate | 2 · split | 256K | no data — contribute | no data — contribute | 47.0 GB (estimate) | 94.0 GB (estimate) | 64 GB kit (estimate) | Calculator Labeled Q3 estimate from the calculator + UD-Q3_K_XL listing. Not a local desktop speed run. |
llama.cpp first. Strata only where this site already documents it. Flags marked on the recipe stay shape-only. Recommended context is 256K.
If the app is not installed, packages are not published yet. There is no public download URL. Read the Stack Lab install notes . If Stack Lab is already running on this machine, open the local app .
llama.cpp · UD-Q6_K_XL · 262K
llama.cpp · Vulkan flags not recorded
# Shape only — llama.cpp Q6, 8 threads on the same Flash file. Exact flags TODO.
./build-vk/bin/llama-server \
-m ~/models/Qwen3.8-Flash-Next-UD-Q6_K_XL.gguf \
-t 8 \
-c 262144 \
--host 127.0.0.1 --port 8080
Strata · UD-Q6_K_XL · 262K
Strata · HIP flags not recorded
# Shape only — exact Strata flags TODO. Recipe strata-flash-q6, no MTP.
# This site’s convention: 127.0.0.1:8083 so llama-server can keep :8080.
curl -s http://127.0.0.1:8083/v1/models
Calculator VRAM at the shopper quant, 1 GPU and 2 GPUs (split). The R9700 35B one-card badge is the measured card; other SKUs stay estimates. Speed is not invented here. Offload rows still show both RAM labels on this page.
| GPU | 1 GPU | 2 GPU |
|---|---|---|
| Can R9700 run Qwen 3.8 Flash? R9700 sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can PRO W7900 run Qwen 3.8 Flash? PRO W7900 sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can PRO W7800 run Qwen 3.8 Flash? PRO W7800 sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can PRO W7700 run Qwen 3.8 Flash? PRO W7700 sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can PRO W6800 run Qwen 3.8 Flash? PRO W6800 sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 7900 XTX run Qwen 3.8 Flash? RX 7900 XTX sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 7900 XT run Qwen 3.8 Flash? RX 7900 XT sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 9070 XT run Qwen 3.8 Flash? RX 9070 XT sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 9070 run Qwen 3.8 Flash? RX 9070 sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 9070 GRE run Qwen 3.8 Flash? RX 9070 GRE sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 9060 XT run Qwen 3.8 Flash? RX 9060 XT sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 7900 GRE run Qwen 3.8 Flash? RX 7900 GRE sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 7800 XT run Qwen 3.8 Flash? RX 7800 XT sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 7700 XT run Qwen 3.8 Flash? RX 7700 XT sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 7600 XT run Qwen 3.8 Flash? RX 7600 XT sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 7600 run Qwen 3.8 Flash? RX 7600 sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 6800 XT run Qwen 3.8 Flash? RX 6800 XT sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 6950 XT run Qwen 3.8 Flash? RX 6950 XT sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can RX 6700 XT run Qwen 3.8 Flash? RX 6700 XT sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
| Can Strix Halo run Qwen 3.8 Flash? Strix Halo sheet · Estimate | Does not fit this GPU 167.9 GB estimate | Does not fit this GPU 167.9 GB estimate |
Head-to-head pages use the same sizes, cited context, and calculator fit already on the model pages. Q3 is a labeled estimate. No invented prices. Qwen 3.8 Flash against the other cited /models pages:
Also: Gemma 3 27B · gpt-oss 20B · Llama 3.3 70B · Qwen3.6 27B · Qwen3.8 27B
| Model | Setup | Backend | Quant | Ctx | Decode | Evidence |
|---|---|---|---|---|---|---|
| Qwen 3.8 Flash UD-Q6_K_XL (folder) | 2× r9700 | llama.cpp vulkan | UD-Q6_K_XL | 262K | 16 | Tested on our rig |
| Qwen 3.8 Flash UD-Q6_K_XL (folder) | 2× r9700 | Strata hip | UD-Q6_K_XL | 262K | 32.6 | Tested on our rig |
Measured cells come from verified amdai-bench/run@1 cards credited to the submitter.
amdai-bench/run@1 · single · 256K
Strata v0.1.42 · 2 GPUs · 262,144 tokens
Credited to AMDai @amdai · local-desktop verified submitter
2026-10 (date not recorded)
Measured on our rig, 2026-10. Prior first-party Flash Q6: Strata ~32.6 tok/s. Card count was not published; recorded on the two-R9700 host. A different run from the 32.8 / 31.8 two-card points.
Credited to AMDai @amdai · local-desktop verified submitter
2026-10 (date not recorded)
Measured on our rig, 2026-10. Prior first-party Flash Q6: llama.cpp ~16 tok/s versus Strata ~32.6. Card count was not published; recorded on the two-R9700 host.
Credited to AMDai @amdai · local-desktop verified submitter
2026-10 (date not recorded)
Measured on our rig, 2026-10. Two-card Flash Q6 at 262K load. Strata 32.8 tok/s at 8K. DDR5-5200.
Credited to AMDai @amdai · local-desktop verified submitter
2026-10 (date not recorded)
Measured on our rig, 2026-10. Two-card Flash Q6, Strata, DDR5-3600, 32.0 tok/s at 8K. Pair with 32.8 at DDR5-5200.
amdai-bench/run@1 · single · 32K
Strata v0.1.42 · 2 GPUs · 32,768 tokens
Credited to AMDai @amdai · local-desktop verified submitter
2026-10 (date not recorded)
Measured on our rig, 2026-10. Two-card Flash Q6 at 262K load. Strata 32.8 tok/s at 32K.
amdai-bench/run@1 · single · 100K
Strata v0.1.42 · 2 GPUs · 100,000 tokens
Credited to AMDai @amdai · local-desktop verified submitter
2026-10 (date not recorded)
Measured on our rig, 2026-10. Two-card Flash Q6 at 262K load. Strata 31.8 tok/s at 100K.
Credited to AMDai @amdai · local-desktop verified submitter
2026-10 (date not recorded)
Measured on our rig, 2026-10. Two-card Flash Q6, llama.cpp, DDR5-5200, 18.5 tok/s at 8K.
Credited to AMDai @amdai · local-desktop verified submitter
2026-10 (date not recorded)
Measured on our rig, 2026-10. Two-card Flash Q6, llama.cpp, DDR5-5200, 18.2 tok/s at 32K.
Credited to AMDai @amdai · local-desktop verified submitter
2026-10 (date not recorded)
Measured on our rig, 2026-10. Two-card Flash Q6, llama.cpp, DDR5-5200, 17.9 tok/s at 100K.
Credited to AMDai @amdai · local-desktop verified submitter
2026-10 (date not recorded)
Measured on our rig, 2026-10. Two-card Flash Q6, llama.cpp, DDR5-3600, 15.9 tok/s at 8K.
Credited to AMDai @amdai · local-desktop verified submitter
2026-10 (date not recorded)
Measured on our rig, 2026-10. Two-card Flash Q6, llama.cpp, DDR5-3600, 15.6 tok/s at 32K.
Credited to AMDai @amdai · local-desktop verified submitter
2026-10 (date not recorded)
Measured on our rig, 2026-10. Two-card Flash Q6, llama.cpp, DDR5-3600, 15.3 tok/s at 100K. Strata prefill drops 9–20% at DDR5-5200 versus 3600 — unexplained; no first-party prefill tok/s published.