tt-awesome — Official Tenstorrent Releases Latest stable releases from Tenstorrent's own repositories https://tenstorrent.github.io/tt-awesome/feeds/releases-official.xml Tenstorrenthttps://tenstorrent.github.io/tt-awesome/ 2026-10-22T18:56:32Z https://github.com/tenstorrent/tt-smi/releases/tag/v6.8.0 tt-smi v6.8.0 2026-10-22T18:56:32Z TTDevice.get_pci_device in the UMD backend (#265), cleaning up the hardware abstraction layer and reducing coupling to internal APIs—a change that should make the tool more resilient to future UMD shifts. The versions display box also got a visual refresh to improve readability at a glance.]]> tt-smi v6.8.0 removes direct use of TTDevice.get_pci_device in the UMD backend (#265), cleaning up the hardware abstraction layer and reducing coupling to internal APIs—a change that should make the tool more resilient to future UMD shifts. The versions display box also got a visual refresh to improve readability at a glance.

Links: Repository · v6.8.0

official · added 2026-10-22

]]> https://github.com/tenstorrent/luwen/releases/tag/bh-mod-v2.4.0 luwen bh-mod-v2.4.0 2026-10-22T14:03:34Z Luwen now lets you override max_tdp_limit, giving you finer control over thermal and power constraints on Blackhole hardware—useful when you need to push past default limits in testing or when running in environments where the stock setting is overly conservative.]]> Luwen now lets you override max_tdp_limit, giving you finer control over thermal and power constraints on Blackhole hardware—useful when you need to push past default limits in testing or when running in environments where the stock setting is overly conservative.

Also in this run:

  • bh-mod-v2.3.0 (2026-10-21) — Pyluwen can now be built and packaged for RISC-V 64-bit hosts, expanding the range of development environments where you can work with Blackhole hardware—useful if you're running on non-x86 infrastructure. The release also patches a cable fault detection path in the BH mod layer, tightening error handling for physical connectivity issues.

Links: Repository · bh-mod-v2.4.0

official · added 2026-10-22

]]> https://github.com/tenstorrent/sfpi/releases/tag/7.89.0-macros SFPI 7.89.0-macros 2026-10-21T19:29:06Z SFPI released 7.89.0-macros. Repository: https://github.com/tenstorrent/sfpi

Also in this run:

  • 7.89.0 (2026-10-19) — SFPI now supports sfpi::mask to generate comparison masks directly—letting you construct all-ones or all-zeroes results from any comparison operation without intermediate steps, which streamlines conditional logic in compute kernels. The release also tightens floating-point negation handling, making sign flips more predictable across different operand types.

Links: Repository · 7.89.0-macros

official · added 2026-10-21

]]> https://github.com/tenstorrent/ttnn-visualizer/releases/tag/v0.108.0 ttnn-visualizer v0.108.0 2026-10-21T19:25:28Z performance-report generation now closes its figures to avoid memory bloat. A few reliability fixes round out the release—report links no longer linger across runs, top_ops and diff_reports can read signpost ranges for finer-grained analysis, and the Buffer Summary's late-deallocation toggle displays accurately.]]> The visualizer now includes SoC descriptors for Quasar-32 and Grendel, bringing hardware support in line with recent Tenstorrent silicon—a must-have for teams profiling kernels on these devices. Beyond that, the interface got sharper: slowest-operations rows now link to their ops on the graph and auto-expand folds to show them, the side column resizes intelligently when viewing bottlenecks in isolation, and performance-report generation now closes its figures to avoid memory bloat. A few reliability fixes round out the release—report links no longer linger across runs, top_ops and diff_reports can read signpost ranges for finer-grained analysis, and the Buffer Summary's late-deallocation toggle displays accurately.

Links: Repository · v0.108.0

official · added 2026-10-21

]]> https://github.com/tenstorrent/tt-cli/releases/tag/v1.2.0 tt-cli v1.2.0 2026-10-21T17:50:08Z tt model serve command is now available alongside tt serve, and tt launch hermes brings agent connectivity for model serving—both small quality-of-life wins for developers building inference workflows. More importantly, a security fix redacts secrets from logs and crash reports, and a file-permission fix resolves weight-loading issues with restrictive umasks. The release also expands hardware support across the p100 and p300 lines for models like Ornith 1.5 and Gemma 4, ships tt-inference-server v0.25.0 under the hood, and makes bundle info human-readable by default.]]> The tt model serve command is now available alongside tt serve, and tt launch hermes brings agent connectivity for model serving—both small quality-of-life wins for developers building inference workflows. More importantly, a security fix redacts secrets from logs and crash reports, and a file-permission fix resolves weight-loading issues with restrictive umasks. The release also expands hardware support across the p100 and p300 lines for models like Ornith 1.5 and Gemma 4, ships tt-inference-server v0.25.0 under the hood, and makes bundle info human-readable by default.

Links: Repository · v1.2.0

official · added 2026-10-21

]]> https://github.com/tenstorrent/ttsim/releases/tag/v1.13.0 ttsim v1.13.0 2026-10-21T11:54:58Z 1.1–1.4× faster on Wormhole and Blackhole, a win for test turnaround on hardware-heavy workloads. Beyond that speedup, this release lands a raft of chip-specific fixes—corrected packer behavior for narrowing conversions and int8/int32 edge cases on BH, UNPACR counter fixes on WH/BH, threshold ReLU precision tightening, and several QSR MVMUL and format-conversion corrections—alongside new register support (unpacker 1 offsets, PACR Z stride) and better validation for semaphore and MXFP4 operations across the board.]]> Idle Tensix tile simulation is now 1.1–1.4× faster on Wormhole and Blackhole, a win for test turnaround on hardware-heavy workloads. Beyond that speedup, this release lands a raft of chip-specific fixes—corrected packer behavior for narrowing conversions and int8/int32 edge cases on BH, UNPACR counter fixes on WH/BH, threshold ReLU precision tightening, and several QSR MVMUL and format-conversion corrections—alongside new register support (unpacker 1 offsets, PACR Z stride) and better validation for semaphore and MXFP4 operations across the board.

Also in this run:

  • v1.12.1 (2026-10-19) — ttsim v1.12.1 corrects PCIe topology for the x32 configurations on Wormhole and Blackhole, which matters if you're simulating multi-chip setups, and adds packer support for bf16→fp32 conversion on BH to expand data-type flexibility in your kernels. The release also tightens validation checks across throttle modes and packer edge cases on Blackhole, catching subtle correctness issues that could otherwise slip through simulation.
  • v1.12.0 (2026-10-19) — Floating-point environment isolation across WH/BH and a sweep of numerical fixes make this release a solid stability win: libttsim no longer touches the host's FPU state, and several long-tail rounding and packing bugs with fp16, fp8, and bf16 are now corrected. Beyond that, the release rounds out instruction and format support—BFP2 MVMUL and GAPOOL operations, negative ReLU thresholds, packer Read_raw for bf16/fp16, and mid-word BFP4/BFP2 unpacks all land on WH/BH—while QSR gains IEEE-compliant RV32F with Zfh extensions and ttsim-riscv64 picks up debugging conveniences like --disk-persist and --stop-on-uart-text.
  • v1.11.5 (2026-10-17) — QSR simulation now supports tt.rocc.noc_fence and global semaphore initialization, filling gaps that made certain workloads difficult to test on the simulator — particularly useful if you're developing compute kernels that rely on synchronization primitives. The release also expands PCIe endpoint coverage and RV32 FENCE modes, rounds out tile counter support, and fixes PCIe BAR attributes that could cause subtle mismatches between simulation and hardware behavior.
  • v1.11.4 (2026-10-15) — QSR (Tenstorrent's functional simulator) levels up with substantially expanded instruction coverage—RV64 cache maintenance, SFPLUT, enhanced semaphore and NOC atomics support, plus new PCIe and Ethernet register stubs—making it a more faithful testbed for complex workloads running on real hardware. The fix to Tensix semaphore reset values and tightened stall logic for SFP_STOCH_RND and SFPMUL24 addresses real correctness issues, while bumping the MOP inner_loop_len limit across all chips opens the door to denser, more efficient instruction sequences. Better error reporting should also make debugging simulation mismatches less of a detective job.

Links: Repository · v1.13.0

official · added 2026-10-21

]]> https://github.com/tenstorrent/tt-studio/releases/tag/v2.12.0 TT-Studio v2.12.0 2026-10-16T19:57:01Z pipx install tt-studio without cloning the repo—configuration and model state live in ~/.tt-studio and persist across upgrades—making it far simpler to keep the toolchain current. More consequentially, the platform now surfaces 51 community models (Gemma, Qwen, Mistral, and others) via tt-model-manager alongside inference-server offerings, all deployable from the same interface and compatible with chat, Voice Agent, and Open WebUI. The model catalog now auto-syncs to match your installed TT Inference Server version at startup, keeping what you can deploy in sync with what the server supports; a slew of fixes address deployment reliability—proper P300 card resets between models, cleaned-up chat parameter defaults for reasoning models, and smoother Open WebUI onboarding—plus infrastructure hardening (Docker Control now runs containerized with no host socket exposure) and frontend/backend upgrades to React 19 and Django 5.2 LTS.]]> You can now install TT Studio with pipx install tt-studio without cloning the repo—configuration and model state live in ~/.tt-studio and persist across upgrades—making it far simpler to keep the toolchain current. More consequentially, the platform now surfaces 51 community models (Gemma, Qwen, Mistral, and others) via tt-model-manager alongside inference-server offerings, all deployable from the same interface and compatible with chat, Voice Agent, and Open WebUI. The model catalog now auto-syncs to match your installed TT Inference Server version at startup, keeping what you can deploy in sync with what the server supports; a slew of fixes address deployment reliability—proper P300 card resets between models, cleaned-up chat parameter defaults for reasoning models, and smoother Open WebUI onboarding—plus infrastructure hardening (Docker Control now runs containerized with no host socket exposure) and frontend/backend upgrades to React 19 and Django 5.2 LTS.

Links: Repository · v2.12.0

official · added 2026-10-16

]]> https://github.com/tenstorrent/tt-toplike/releases/tag/v0.13.17 tt-toplike v0.13.17 2026-10-16T17:58:26Z /proc access, addressing a community-reported problem that made production deployments harder to monitor. The release also polishes the inference server visualization with refined activity lanes and better handling of servers missing /health endpoints, particularly helpful if you're running models through tt-model-manager.]]> tt-toplike 0.13.17 quiets a noisy auditd issue that could flood logs on systems with restricted /proc access, addressing a community-reported problem that made production deployments harder to monitor. The release also polishes the inference server visualization with refined activity lanes and better handling of servers missing /health endpoints, particularly helpful if you're running models through tt-model-manager.

Links: Repository · v0.13.17

official · added 2026-10-16

]]> https://github.com/tenstorrent/tt-local-generator/releases/tag/v0.104.4 tt-local-generator v0.104.4 2026-10-15T22:03:43Z tt-metal installation by falling back to a lightweight served version, which should streamline setup for developers just wanting to generate videos without building the whole stack.]]> LoRA support now works with AnimateDiff generations, letting you apply custom fine-tuned adapters to video synthesis—previously this combination would fail. On the infrastructure side, AnimateDiff can now run without a full tt-metal installation by falling back to a lightweight served version, which should streamline setup for developers just wanting to generate videos without building the whole stack.

Links: Repository · v0.104.4

official · added 2026-10-15

]]> https://github.com/tenstorrent/tt-animatediff/releases/tag/v0.13.2 tt-animatediff v0.13.2 2026-10-15T20:20:53Z pixel, pixel-fast, and lcm adapters without redeployment — plus live frame previews scoped per client and experimental motion adapter support give you finer control over generation. The CLI gains --lora to invoke LCM variants at 4 steps, and the container image (episod/tt-animatediff-container-v5.1) has been rebuilt and verified against its manifest after the previous upload was incomplete; if you have an older cached copy, run tt-model serve with --refresh. The server also supports prompt schedules, sampler selection (euler/Lightning), and chain save/load with alpha blending, all gated by capability — see the README changelog for the full feature set.]]> LoRA variants are now served through the operator catalog, letting you swap between pixel, pixel-fast, and lcm adapters without redeployment — plus live frame previews scoped per client and experimental motion adapter support give you finer control over generation. The CLI gains --lora to invoke LCM variants at 4 steps, and the container image (episod/tt-animatediff-container-v5.1) has been rebuilt and verified against its manifest after the previous upload was incomplete; if you have an older cached copy, run tt-model serve with --refresh. The server also supports prompt schedules, sampler selection (euler/Lightning), and chain save/load with alpha blending, all gated by capability — see the README changelog for the full feature set.

Links: Repository · v0.13.2

official · added 2026-10-15

]]> https://github.com/tenstorrent/tt-umd/releases/tag/v0.9.14 tt-umd v0.9.14 2026-10-15T15:19:30Z matching the ISA requirement that NoC requests to MMIO be exactly one register wide — a fix needed by tt-exalens and picked up in this bump. The release also brings IP partitioning support for RTL simulator builds (multiple devices per simulator run via ip_layout.yaml), Keraunos device support, improved warm reset recovery on BH QuietBox, and refactored utilities (MutexInterface, RobustMutex, base error types, and SemVer) into a new tt-umd-utils component for reuse across the ecosystem. Wheels for Python 3.9–3.13 on x86_64 and aarch64 are on PyPI.]]> MMIO register access now splits multi-register reads and writes into single 32-bit operations on Remote, JTAG, and TTSim protocols, matching the ISA requirement that NoC requests to MMIO be exactly one register wide — a fix needed by tt-exalens and picked up in this bump. The release also brings IP partitioning support for RTL simulator builds (multiple devices per simulator run via ip_layout.yaml), Keraunos device support, improved warm reset recovery on BH QuietBox, and refactored utilities (MutexInterface, RobustMutex, base error types, and SemVer) into a new tt-umd-utils component for reuse across the ecosystem. Wheels for Python 3.9–3.13 on x86_64 and aarch64 are on PyPI.

Links: Repository · v0.9.14

official · added 2026-10-15

]]> https://github.com/tenstorrent/tt-inference-server/releases/tag/v0.25.0 tt-inference-server v0.25.0 2026-10-15T11:39:48Z This release brings expanded model coverage across Tenstorrent's hardware lineup, particularly for Blackhole Galaxy: Qwen3-32B, Gemma-4 (both 31B and 12B instruction-tuned variants), and the large-scale GPT-OSS-120B are now supported, while Qwen3.8-27B and a video diffusion model (Wan2.2-T2V-A14B) gain new platform homes. LoRA fine-tuning for Llama-3.1-8B-Instruct has also been uplifted on P150, extending training capabilities to that hardware. Most models reach COMPLETE status, though GPT-OSS-120B and the LoRA trainer remain experimental—worth trying if you're pushing scale or customization boundaries, but expect rough edges.

Links: Repository · v0.25.0

official · added 2026-10-15

]]> https://github.com/tenstorrent/tt-metal/releases/tag/v0.80.0 tt-metal v0.80.0 2026-10-14T12:25:50Z Llama QB2 prefill crashes are fixed, a 31B Gemma-4 hang in sampling is resolved, and Gemma-4 gets tuned prefill warmup and sampling thresholds to improve inference quality. Also, new model support lands with Gemma-4 31B QB2.]]> This release tackles stability and scale across the model zoo: Llama QB2 prefill crashes are fixed, a 31B Gemma-4 hang in sampling is resolved, and Gemma-4 gets tuned prefill warmup and sampling thresholds to improve inference quality. Also, new model support lands with Gemma-4 31B QB2.

Links: Repository · v0.80.0

official · added 2026-10-14

]]> https://github.com/tenstorrent/tt-system-firmware/releases/tag/v19.16.0 tt-system-firmware v19.16.0 2026-10-10T13:49:17Z GDDR CA latch-and-retest functionality that hardens initialization—especially useful if you're pushing Blackhole hardware in production. Ethernet training is now more resilient to link-partner resets, PCIe EQ presets can vary by platform, and the boot firmware table layout got redesigned for safer updates; see the migration guide if you're upgrading from 19.15.0.]]> Significant reliability bumps across the stack in this release. The headline is GDDR CA latch-and-retest functionality that hardens initialization—especially useful if you're pushing Blackhole hardware in production. Ethernet training is now more resilient to link-partner resets, PCIe EQ presets can vary by platform, and the boot firmware table layout got redesigned for safer updates; see the migration guide if you're upgrading from 19.15.0.

Links: Repository · v19.16.0

official · added 2026-10-10

]]> https://github.com/tenstorrent/tt-installer/releases/tag/v3.6.2 tt-installer v3.6.2 2026-10-09T17:43:08Z tt-cli now installs as an isolated uv tool pinned to each release, eliminating version conflicts and simplifying dependency management across your Tenstorrent setup. The installer also now supports Ubuntu 26.04 in its test matrix and refines shell path handling, ensuring smoother onboarding on newer Ubuntu LTS releases.]]> tt-cli now installs as an isolated uv tool pinned to each release, eliminating version conflicts and simplifying dependency management across your Tenstorrent setup. The installer also now supports Ubuntu 26.04 in its test matrix and refines shell path handling, ensuring smoother onboarding on newer Ubuntu LTS releases.

Links: Repository · v3.6.2

official · added 2026-10-09

]]> https://github.com/tenstorrent/tt-flash/releases/tag/v4.1.0 tt-flash v4.1.0 2026-10-02T19:04:14Z pyluwen to a stable 0.10.0 release, and updates Galaxy variant naming to GALAXY_CF for consistency.]]> tt-flash v4.1.0 fixes a critical hang that could occur when flashing Galaxy (114) devices, and now allows flashing to proceed even when GDDR training or BIST fails—useful when those checks flag non-fatal issues but you still need to get firmware on the chip. Under the hood, this release hardens the multiprocessing pool infrastructure with better fork sequencing and stdout handling to prevent deadlocks, pins pyluwen to a stable 0.10.0 release, and updates Galaxy variant naming to GALAXY_CF for consistency.

Links: Repository · v4.1.0

official · added 2026-10-02

]]> https://github.com/tenstorrent/tt-exalens/releases/tag/v0.4.1 tt-exalens v0.4.1 2026-09-28T13:19:21Z Device timeout errors now propagate correctly to native code, fixing a debugging blind spot where exceptions were being swallowed during hardware communication failures. The release also adds Quasar support across several fronts—debug bus signal visibility, watchpoint PC address testing, and rocket core program counter readout when the system management network is unavailable—while making the debugger smarter about hiding unsupported commands on incompatible architectures. A CMake fix pins ELFIO to a working commit and simulation connections now work with the latest UMD API.

Links: Repository · v0.4.1

official · added 2026-09-28

]]> https://github.com/tenstorrent/whisper/releases/tag/r1.864 whisper r1.864 2026-09-24T19:46:01Z whisper released r1.864. Repository: https://github.com/tenstorrent/whisper

Links: Repository · r1.864

official · added 2026-09-24

]]> https://github.com/tenstorrent/tt-operator/releases/tag/v0.3.0 tt-operator v0.3.0 2026-09-16T16:11:43Z tt-dra-driver on anything below k8s 1.33, the Helm chart fails fast with a clear error rather than letting a broken deployment proceed. This release also ships bumped versions of the driver manager, fabric manager, and DRA driver, and adds a per-release component version matrix to the docs so you can track exactly what's in each build.]]> The operator now catches Kubernetes version mismatches early: if you try to enable tt-dra-driver on anything below k8s 1.33, the Helm chart fails fast with a clear error rather than letting a broken deployment proceed. This release also ships bumped versions of the driver manager, fabric manager, and DRA driver, and adds a per-release component version matrix to the docs so you can track exactly what's in each build.

Links: Repository · v0.3.0

official · added 2026-09-16

]]> https://github.com/tenstorrent/tt-vscode-toolkit/releases/tag/v0.1.31 tt-vscode-toolkit v0.1.31 2026-09-09T22:09:46Z This release primarily fixes errors and modernizes lessons about the TT-QuietBox 2 and working with ttsim.

Links: Repository · v0.1.31

official · added 2026-09-09

]]> https://github.com/tenstorrent/tt-kmd/releases/tag/ttkmd-2.11.0 tt-kmd ttkmd-2.11.0 2026-08-14T18:31:59Z TENSTORREAM_IOCTL_SMC_MSG interface—letting applications post requests and poll responses asynchronously via separate non-blocking calls—and forwards Blackhole firmware logs directly to the kernel log with configurable level filtering. More importantly, it squashes several nasty concurrency bugs: a deadlock between TLB allocation and process memory locking, a race where TLB windows could be reassigned mid-configuration, and device-removal races that could leave in-flight open() calls accessing freed state. The driver now also validates telemetry tables more strictly to catch corrupted firmware data and reports actual ARC errors from power-state changes instead of masking them with -EINVAL.]]> This release brings userspace access to Blackhole and Wormhole system-management firmware through a new TENSTORREAM_IOCTL_SMC_MSG interface—letting applications post requests and poll responses asynchronously via separate non-blocking calls—and forwards Blackhole firmware logs directly to the kernel log with configurable level filtering. More importantly, it squashes several nasty concurrency bugs: a deadlock between TLB allocation and process memory locking, a race where TLB windows could be reassigned mid-configuration, and device-removal races that could leave in-flight open() calls accessing freed state. The driver now also validates telemetry tables more strictly to catch corrupted firmware data and reports actual ARC errors from power-state changes instead of masking them with -EINVAL.

Links: Repository · ttkmd-2.11.0

official · added 2026-08-14

]]> https://github.com/tenstorrent/tt-burnin/releases/tag/v0.4.4 tt-burnin v0.4.4 2026-08-14T17:13:48Z This release pulls in pyluwen 0.9.0, which brings improvements to the underlying device communication layer that tt-burnin relies on for stress testing. If you're running burnin tests on your Tenstorrent hardware, the updated pyluwen dependency should give you better stability and compatibility with the latest device firmware.

Links: Repository · v0.4.4

official · added 2026-08-14

]]> https://github.com/tenstorrent/tt-forge/releases/tag/1.4.0 tt-forge 1.4.0 2026-07-30T11:17:16Z supported model matrix, while internal tooling improvements (including a move of AI bringup scripts to a dedicated repo) clean up the monorepo and streamline testing workflows. The release includes updated PyTorch and dependency commits across tt-xla, tt-mlir, and tt-metal, ensuring compatibility with recent compiler and hardware stacks.]]> tt-forge 1.4.0 brings a comprehensive refresh of model coverage and infrastructure, now supporting everything from Llama 3.2 Vision to Qwen 3 and Falcon 3 across both n150 and p150 hardware. The most significant change is the expanded LLM lineup—you can now compile and deploy over 40 language model variants including the latest reasoning-focused OLMo3 models and multimodal systems like Pixtral—making it practical to benchmark performance across a much wider range of production targets. On the non-LLM side, embedding models, vision transformers, and diffusion backends round out the supported model matrix, while internal tooling improvements (including a move of AI bringup scripts to a dedicated repo) clean up the monorepo and streamline testing workflows. The release includes updated PyTorch and dependency commits across tt-xla, tt-mlir, and tt-metal, ensuring compatibility with recent compiler and hardware stacks.

Links: Repository · 1.4.0

official · added 2026-07-30

]]> https://github.com/tenstorrent/tt-xla/releases/tag/1.4.0 tt-xla 1.4.0 2026-07-30T11:06:22Z Chunked prefill with decoupled buckets and runtime scheduling, plus multi-parallelism support across generative and pooling models, unlock more efficient vLLM inference on Tenstorrent hardware—letting you scale from single-device to data-parallel and tensor-parallel deployments without manual graph fragmentation. Beyond vLLM, you'll find new e2e generative model pipelines (Flux, Flux.2, Janus-Pro, Z-Image, GLM-Image), MLA attention in vLLM, mixed-precision quantization heuristics, improved SDPA handling for bool masks behind an opt-in flag, and fixes to device-count mismatches and layer-norm dtype issues under AOTAutograd, all tested against a growing roster of LLMs and image-generation models across n150, p150, and n300 hardware.

Links: Repository · 1.4.0

official · added 2026-07-30

]]> https://github.com/tenstorrent/tt-topology/releases/tag/v1.2.20 tt-topology v1.2.20 2026-07-23T15:35:12Z pyluwen dependency to 0.8.5, bringing in improvements and fixes from that library that support tt-topology's hardware introspection and topology validation work. If you're using tt-topology to characterize or configure your Tenstorrent mesh, this update ensures you're working with the latest device communication layer.]]> This release bumps the underlying pyluwen dependency to 0.8.5, bringing in improvements and fixes from that library that support tt-topology's hardware introspection and topology validation work. If you're using tt-topology to characterize or configure your Tenstorrent mesh, this update ensures you're working with the latest device communication layer.

Links: Repository · v1.2.20

official · added 2026-07-23

]]> https://github.com/tenstorrent-riscv-software/tt-bh-linux/releases/tag/v0.11 tt-bh-linux v0.11 2026-04-13T15:10:59Z This release upgrades to Linux Kernel 7.0 and adds device tree support for performance monitoring, letting you track system metrics directly from userspace. The kernel now ships with a patch that eliminates spurious swiotlb errors during boot, cleaning up initialization logs and making debugging easier. Several fixes to the console tool and build system round out the update.

Links: Repository · v0.11

official · added 2026-04-13

]]> https://github.com/tenstorrent/tt-firmware/releases/tag/v19.6.0 tt-firmware v19.6.0 2026-02-20T16:53:34Z This release brings memory and stability improvements across Tenstorrent's platform lineup, with Wormhole now supporting Samsung GDDR6 and both platforms implementing security wipes during initialization. On Blackhole, the p300 gains critical fixes for reboot reliability and expanded board support with Galaxy revC, while p300c power limits have been tuned (550W board power, 125W TDP, 88°C GDDR thermals). The telemetry layer also gets new visibility into AICLK arbitration decisions with granular tracing, helping developers debug frequency scaling behavior more effectively.

Links: Repository · v19.6.0

official · added 2026-02-20

]]> https://github.com/tenstorrent/tt-tools-common/releases/tag/v1.6.0 tt-tools-common v1.6.0 2025-12-23T21:02:08Z python3-tt-tools-common in the Tenstorrent PPA, so the shared plumbing under tt-smi, tt-flash, and friends installs cleanly via apt.]]> A packaging-focused cut: beyond the version bump, this build generates proper Ubuntu and Debian changelogs — it's the version now shipping as python3-tt-tools-common in the Tenstorrent PPA, so the shared plumbing under tt-smi, tt-flash, and friends installs cleanly via apt.

Also in this run:

  • v1.5.1 (2025-12-23) — Chip resets get more configurable with a new secondary_bus_reset option in the reset sequence (#60) — useful when the default secondary-bus step doesn't suit your platform. The full_lds_reset path also picked up cleaner arguments and much more detailed docstrings, so anyone scripting resets against this library has a clearer contract to work with.

Links: Repository · v1.6.0

official · added 2025-12-23

]]> https://github.com/tenstorrent/tt-system-tools/releases/tag/v1.4.1 tt-system-tools v1.4.1 2025-12-08T17:23:48Z tt-system-tools released v1.4.1. Repository: https://github.com/tenstorrent/tt-system-tools

Links: Repository · v1.4.1

official · added 2025-12-08

]]> https://github.com/tenstorrent/riescue/releases/tag/v1.7.0 RiESCUE v1.7.0 2025-12-03T19:29:44Z test_setup now runs once during scheduler initialization rather than on every loop iteration. Bug fixes address deterministic CSR ordering, privilege mode handling in virtualized environments, and stack allocation for algorithm tests, while new documentation with Mermaid flowcharts and API details should help developers understand and debug runtime behavior more easily.]]> The loader and scheduler in RiescueD have been refactored with explicit interfaces that make their control flow clearer and reduce redundant checks—test_setup now runs once during scheduler initialization rather than on every loop iteration. Bug fixes address deterministic CSR ordering, privilege mode handling in virtualized environments, and stack allocation for algorithm tests, while new documentation with Mermaid flowcharts and API details should help developers understand and debug runtime behavior more easily.

Links: Repository · v1.7.0

official · added 2025-12-03

]]> https://github.com/tenstorrent/tt-torch/releases/tag/0.4.0 tt-torch 0.4.0 2025-09-29T22:23:47Z tt-torch released 0.4.0. Repository: https://github.com/tenstorrent/tt-torch

Links: Repository · 0.4.0

official · added 2025-09-29

]]> https://github.com/tenstorrent/tt-polaris/releases/tag/pre_perfmodel_merge polaris pre_perfmodel_merge 2025-09-19T18:16:27Z polaris released pre_perfmodel_merge. Repository: https://github.com/tenstorrent/tt-polaris

Links: Repository · pre_perfmodel_merge

official · added 2025-09-19

]]> https://github.com/tenstorrent/riscv_arch_tests/releases/tag/v0.2.0%2Baligned-access riscv_arch_tests v0.2.0+aligned-access 2025-01-23T17:16:16Z This release brings byte-alignment compliance testing to the RISC-V architecture test suite, ensuring that RV IMFA tests across all privilege levels (Machine, Supervisor, and User) and configurations (bare metal and paging disabled) properly enforce alignment constraints. The CI pipeline now validates that these tests correctly reject misaligned memory accesses, catching potential issues where implementations might incorrectly permit unaligned data or instruction fetches that the spec forbids.

Links: Repository · v0.2.0+aligned-access

official · added 2025-01-23

]]>