Issues / #2354

#2354 Intel arc alchemist A770 xe driver report on 0.1.44

open · @flare04 · 0 commentaires · Sur GitHub

BenchmarksSetup & installServer & APIDocumentation

Description

<h2 style="box-sizing: border-box; margin: 22px 0px 10px; padding: 0px 0px 4px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; font-weight: 700; color: rgb(255, 255, 255); line-height: 1.3; font-size: 20px; border-bottom: 1px solid rgb(42, 42, 69); font-family: -apple-system, &quot;system-ui&quot;, &quot;Segoe UI&quot;, Inter, system-ui, sans-serif; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; letter-spacing: normal; orphans: 2; text-align: start; text-indent: 0px; text-transform: none; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; white-space: normal; background-color: rgb(13, 13, 26); text-decoration-thickness: initial; text-decoration-style: initial; text-decoration-color: initial;">The summary, in brief</h2><p style="box-sizing: border-box; margin: 0px 0px 10px; padding: 0px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; color: rgb(255, 248, 220); font-family: -apple-system, &quot;system-ui&quot;, &quot;Segoe UI&quot;, Inter, system-ui, sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: 2; text-align: start; text-indent: 0px; text-transform: none; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; white-space: normal; background-color: rgb(13, 13, 26); text-decoration-thickness: initial; text-decoration-style: initial; text-decoration-color: initial;"><strong style="box-sizing: border-box; margin: 0px; padding: 0px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; color: rgb(255, 255, 255); font-weight: 600;">Experience:</strong> v0.1.44 (tag <code style="box-sizing: border-box; margin: 0px; padding: 1px 5px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; font-family: ui-monospace, SFMono-Regular, &quot;SF Mono&quot;, Menlo, Consolas, &quot;Liberation Mono&quot;, monospace; font-size: 12.5px; background: none 0% 0% / auto repeat scroll padding-box border-box rgba(0, 0, 0, 0.35); border-radius: 4px; color: rgb(255, 255, 255);">ed4773eb</code>, 376 commits past v0.1.42) builds clean on the A770 — 129/129 targets, 42.9 MB — and runs on our serve setup with the config differing <strong style="box-sizing: border-box; margin: 0px; padding: 0px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; color: rgb(255, 255, 255); font-weight: 600;">only</strong> in <code style="box-sizing: border-box; margin: 0px; padding: 1px 5px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; font-family: ui-monospace, SFMono-Regular, &quot;SF Mono&quot;, Menlo, Consolas, &quot;Liberation Mono&quot;, monospace; font-size: 12.5px; background: none 0% 0% / auto repeat scroll padding-box border-box rgba(0, 0, 0, 0.35); border-radius: 4px; color: rgb(255, 255, 255);">exe</code>/<code style="box-sizing: border-box; margin: 0px; padding: 1px 5px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; font-family: ui-monospace, SFMono-Regular, &quot;SF Mono&quot;, Menlo, Consolas, &quot;Liberation Mono&quot;, monospace; font-size: 12.5px; background: none 0% 0% / auto repeat scroll padding-box border-box rgba(0, 0, 0, 0.35); border-radius: 4px; color: rgb(255, 255, 255);">cwd</code>. It's been live since today via a <code style="box-sizing: border-box; margin: 0px; padding: 1px 5px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; font-family: ui-monospace, SFMono-Regular, &quot;SF Mono&quot;, Menlo, Consolas, &quot;Liberation Mono&quot;, monospace; font-size: 12.5px; background: none 0% 0% / auto repeat scroll padding-box border-box rgba(0, 0, 0, 0.35); border-radius: 4px; color: rgb(255, 255, 255);">v0144.conf</code> drop-in that resets <code style="box-sizing: border-box; margin: 0px; padding: 1px 5px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; font-family: ui-monospace, SFMono-Regular, &quot;SF Mono&quot;, Menlo, Consolas, &quot;Liberation Mono&quot;, monospace; font-size: 12.5px; background: none 0% 0% / auto repeat scroll padding-box border-box rgba(0, 0, 0, 0.35); border-radius: 4px; color: rgb(255, 255, 255);">ExecStart=</code> then wins over <code style="box-sizing: border-box; margin: 0px; padding: 1px 5px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; font-family: ui-monospace, SFMono-Regular, &quot;SF Mono&quot;, Menlo, Consolas, &quot;Liberation Mono&quot;, monospace; font-size: 12.5px; background: none 0% 0% / auto repeat scroll padding-box border-box rgba(0, 0, 0, 0.35); border-radius: 4px; color: rgb(255, 255, 255);">v0142.conf</code>; rollback is <code style="box-sizing: border-box; margin: 0px; padding: 1px 5px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; font-family: ui-monospace, SFMono-Regular, &quot;SF Mono&quot;, Menlo, Consolas, &quot;Liberation Mono&quot;, monospace; font-size: 12.5px; background: none 0% 0% / auto repeat scroll padding-box border-box rgba(0, 0, 0, 0.35); border-radius: 4px; color: rgb(255, 255, 255);">rm</code> that file + reload + waited restart.</p><p style="box-sizing: border-box; margin: 0px 0px 10px; padding: 0px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; color: rgb(255, 248, 220); font-family: -apple-system, &quot;system-ui&quot;, &quot;Segoe UI&quot;, Inter, system-ui, sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: 2; text-align: start; text-indent: 0px; text-transform: none; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; white-space: normal; background-color: rgb(13, 13, 26); text-decoration-thickness: initial; text-decoration-style: initial; text-decoration-color: initial;"><strong style="box-sizing: border-box; margin: 0px; padding: 0px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; color: rgb(255, 255, 255); font-weight: 600;">Measured, paired on one card</strong> (unique nonce per request, service stopped for the run, throughput from the engine's own log):</p><input type="search" class="markdown-table-filter" placeholder="Filter table" aria-label="Filter table" autocomplete="off" spellcheck="false" style="box-sizing: border-box; margin: 8px 0px 4px; padding: 5px 8px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; font-family: -apple-system, &quot;system-ui&quot;, &quot;Segoe UI&quot;, Inter, system-ui, sans-serif; transition-property: none !important; transition-duration: 0s !important; transition-timing-function: ease; transition-delay: 0s !important; display: block; width: min(260px, 100%); border: 1px solid rgba(255, 255, 255, 0.14); border-radius: 6px; background: none 0% 0% / auto repeat scroll padding-box border-box rgba(255, 255, 255, 0.04); color: rgb(255, 248, 220); font-style: inherit; font-variant: inherit; font-weight: inherit; font-stretch: inherit; line-height: inherit; font-optical-sizing: inherit; font-size-adjust: inherit; font-kerning: inherit; font-feature-settings: inherit; font-variation-settings: inherit; font-language-override: inherit; font-size: 12px; orphans: 2; widows: 2; -webkit-text-stroke-width: 0px; white-space: normal; text-decoration-thickness: initial; text-decoration-style: initial; text-decoration-color: initial;">

```
prompt | arm | prefill | decode

-- | -- | -- | --

1,893 tok, n=4 | v0.1.42 | 162.85 | 22.95
               
[strata-v0144-report.md](https://github.com/user-attachments/files/33414550/strata-v0144-report.md)

| v0.1.44 | 181.3 (+11%) | 26.4 (+15%)

30,632 tok, n=2 | v0.1.42 | 188.75 | 26.5
                | v0.1.44 | 208.6 (+10.5%) | 28.1 (+6.0%)
```

<p style="box-sizing: border-box; margin: 0px 0px 10px; padding: 0px; scrollbar-width: thin; scrollbar-color: rgba(255, 255, 255, 0.12) transparent; transition-property: none !important; transition-duration: 0s !important; transition-delay: 0s !important; color: rgb(255, 248, 220); font-family: -apple-system, &quot;system-ui&quot;, &quot;Segoe UI&quot;, Inter, system-ui, sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: 2; text-align: start; text-indent: 0px; text-transform: none; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; white-space: normal; background-color: rgb(13, 13, 26); text-decoration-thickness: initial; text-decoration-style: initial; text-decoration-color: initial;">Our patches are non-perf (header detection, a startup diagnostic), so the gain is upstream's own — the Alchemist GDN v6 + GR kernel work. Two honest caveats are in the doc: the short-prompt prefill reading isn't a ceiling (it's cold-expert-read dominated), and upstream's "478→511 tok/s @ 32K" doesn't reproduce as a tok/s figure here (~188–208 both arms).</p>

More details in the attachment

Sur le site

Liens install, modèles, releases.