TypeScript compiled to CUDA
landscape 1200×630square 1080×1080portrait 1080×1350story 1080×1920HD 16:9 1920×1080
Brand kit · public beta
Logos, images & share kit.
Writing about Symmetra, or sharing it with someone who should see it? Everything below is free to use for that. Every image is sized for its platform and carries the same message: TypeScript compiled to CUDA.
Ready-made images
Seven campaign images, each in five sizes: landscape for link posts, square and portrait for feeds, story for 9:16 formats and 16:9 HD for video thumbnails and slides.
From .ts to GPU in one command
landscape 1200×630square 1080×1080portrait 1080×1350story 1080×1920HD 16:9 1920×1080
Measured. Not promised.
landscape 1200×630square 1080×1080portrait 1080×1350story 1080×1920HD 16:9 1920×1080
Your API is a Turing machine
landscape 1200×630square 1080×1080portrait 1080×1350story 1080×1920HD 16:9 1920×1080
Run real Turing machines free
landscape 1200×630square 1080×1080portrait 1080×1350story 1080×1920HD 16:9 1920×1080
Same syntax. Different silicon.
landscape 1200×630square 1080×1080portrait 1080×1350story 1080×1920HD 16:9 1920×1080
The public beta is open
landscape 1200×630square 1080×1080portrait 1080×1350story 1080×1920HD 16:9 1920×1080
Banners & profile images
Cover images are laid out with the text clear of the area where each platform overlays the profile picture.
Logos & colours
Use the logo as supplied on a dark background. Do not recolour, stretch or add effects, and leave clear space equal to the height of the hexagon.
Violet
#7c6fff
Mint
#5af0c4
Sky
#8bc6ff
Ember
#ff8b61
Night
#04070b
Panel
#0c1923
Type: Exo 2 for text, IBM Plex Mono for code, Hemi Head Bold Italic for display headlines.
Boilerplate & facts
Copy as is. Symmetra is a public beta, so please keep the beta framing and the test conditions whenever you quote a figure.
One line
Symmetra is a GPU-native runtime that compiles TypeScript route handlers to CUDA and runs them on NVIDIA GPUs.
Short description
Symmetra is a public-beta runtime that compiles TypeScript route handlers to CUDA and executes them on the GPU, with no VM and no garbage collector. Handlers run as Turing-machine programs on a persistent GPU kernel, while the CPU only handles network I/O. It supports NVIDIA RTX 20-series and newer GPUs with CUDA 12.x or 13.x, and three engines are free.
Measured figure (quote with its conditions)
624K requests per second on a trivial handler, with a p50 latency of 304 µs, on a single NVIDIA L4 (Google Cloud) using 32 of its 58 streaming multiprocessors (55% GPU capacity). Public beta measurement under those conditions, not a guarantee.
Please
- Say “public beta”. It is not yet recommended for production-critical workloads.
- Quote numbers together with the GPU and conditions. Do not extrapolate them.
- Link to symmscript.com.
Press or partnership questions: symmscript.com/support.