Tenstorrent Blackhole
Tenstorrent
17
Open RISC-V AI hardware
- Version
- Blackhole
- Cost
- from $999/unit
- Model
- Tensix cores
Intel Gaudi 3
Intel · #6 most active of 18 in Chips & Compute
Gaudi 3 targets price-performance against Nvidia parts with open Ethernet scale-out and an oneAPI/PyTorch software path.
100
Trust score · Verified
2 data points checked · 2026-09-13
Current version
1.24.1
Entry cost
Pricing not published. Available via Intel Tiber AI Cloud or OEM platforms.
Changes / 30d
5
The short answers · verified September 16, 2026
- What is the latest version of Intel Gaudi 3?
- The current shipped version of Intel Gaudi 3 is 1.24.1, as of September 16, 2026.
- How much does Intel Gaudi 3 cost?
- Intel Gaudi 3 has no published flat entry price. Pricing not published. Available via Intel Tiber AI Cloud or OEM platforms.
- What AI model does Intel Gaudi 3 use?
- Intel Gaudi 3 runs primarily on Accelerator.
Source: Intel — docs.habana.ai · Reusable under CC BY 4.0 — cite Tomorrow
Capabilities
- 128GB HBM2e
- Open Ethernet scale-out
- PyTorch native
- oneAPI toolchain
- Price-performance focus
- PyTorch Support
- vLLM Integration
- DeepSpeed Training
- SGLang Support
- FP8 Training
- Quantization (UINT4, NF4)
- Hugging Face Optimum
- Triton Inference Server
- TPC Programming
- HCCL (Habana Collective Communications Library)
- GPU Migration Toolkit
- DeepSpeed Support
- FP8 Training & Inference
- Hugging Face Optimum Integration
- Distributed Training (HCCL)
- Quantized Inference (UINT4/NF4)
- Triton & TorchServe Model Serving
- PyTorch Training
- DeepSpeed
- vLLM Plugin
- SGLang
- HPU Graphs
- Inference with Quantization
- HCCL
- Inference
- vLLM Support
- Model Serving
- Distributed Training
- Havana Collective Communications Library (HCCL)
- Transformer Engine
- vLLM Inference
- SGLang Inference
- FP8 / INT4 Quantization
- HCCL Collective Communications
- Kubernetes Orchestration
- PyTorch
- vLLM
- INT4 Inference
- TorchServe
- Quantization (FP8, INT4, NF4)
- H HCCL
- Quantization
- PyTorch training
- FP8 training
- vLLM inference
- DeepSpeed support
- SGLang support
- HPU graphs
- DeepSpeed Integration
- INT4 Quantization
- Quantization (FP8/UINT4/NF4)
- FP8/INT4 Quantization
- HCCL scale-out
- Inference Serving
- Quantization (INT4/NF4)
- Inference using vLLM
- Inference using SGLang
- HCCL Scale-out
- Intel Gaudi Software Suite
Pricing
OEM/cloud
quoted
Free
Entry price over time
Not enough pricing history yet — we start charting from the second observation.
Geek mode
Models under the hood
- disclosed
Accelerator
Intel
- disclosed
Intel Gaudi 2
Intel
- disclosed
Intel Gaudi 3
Intel
Context window
n/a
Public API
yes
Multi-model routing
no
Stack signals
- vllm
- sglang
- pytorch
- deepspeed
Shared model stack
Other tracked products running on the same foundation models — a quick read on how much of the catalog moves when one of these models changes.
- AcceleratorIntel · 1 product
- Intel Gaudi 2Intel · 1 product
- Intel Gaudi 3Intel · 1 product
green disclosed · amber inferred · grey unattributed. Aliases are folded into one model; provider concentration counts only models with an established vendor.
Reported scale
Reference data. Each figure is whatever the source actually said — weekly users, downloads, revenue run-rate — with its own definition and date. These are not comparable between tools and are never used to rank anything.
No public usage figure on record for this product.
Public market scorecard
Intel investor relationsIntel trades as NASDAQ: INTC. Figures below are as reported for Q2 2026 on Jul 23, 2026; market levels are the close on Aug 6, 2026 and are a reference marker, not a live quote.
Q2 2026 revenue
~$12.5B
flat YoY
- Foundry
- still loss-making
- 18A ramp is the swing factor
- Client computing
- ~$7B
Also in Chips & Compute
All Chips & ComputeGoogle TPU IronwoodGoogle
12
Inference-first TPU pods
- Version
- Ironwood (7th generation)
- Cost
- from $5.4/unit
- Model
- Gemini
Cerebras WSE-3
Cerebras
9
Wafer-scale inference and training
- Version
- WSE-3
- Cost
- Free tier / developer tier pricing available (contact/partner integrations)
- Model
- Codex-Spark
NVIDIA Blackwell
NVIDIA
9
GB200/B200 rack-scale AI systems
- Version
- Blackwell
- Cost
- Pricing is not publicly listed (enterprise hardware architecture/system)
- Model
- Blackwell
5
Cloud-native training silicon
- Version
- Trainium2
- Cost
- Pricing not published; available via AWS EC2 instance pricing.
- Model
- Neuron compiler
SambaNova SN40L
SambaNova
5
Dataflow chips for model serving
- Version
- SN40L
- Cost
- Free tier available. Paid plans include Developer (Pay-as-you-go) and Enterprise (Subscription-based).
- Model
- DeepSeek-V3.1
Change history
- version
- version
- capability
New capabilities: Intel Gaudi Software Suite
63 tracked64 tracked · +Intel Gaudi Software Suite
source - version
- version
- version
- capability
New capabilities: Inference using vLLM, Inference using SGLang, HCCL Scale-out
60 tracked63 tracked · +Inference using vLLM, Inference using SGLang, HCCL Scale-out
source - version
- version
- capability
New capabilities: Inference Serving, Quantization (INT4/NF4)
58 tracked60 tracked · +Inference Serving, Quantization (INT4/NF4)
source - capability
New capabilities: FP8/INT4 Quantization, HCCL scale-out
56 tracked58 tracked · +FP8/INT4 Quantization, HCCL scale-out
source - version
- version
- version
- capability
New capabilities: Quantization (FP8/UINT4/NF4)
55 tracked56 tracked · +Quantization (FP8/UINT4/NF4)
source - version
- version
- capability
New capabilities: DeepSpeed Integration, INT4 Quantization
53 tracked55 tracked · +DeepSpeed Integration, INT4 Quantization
source - capability
New capabilities: PyTorch training, FP8 training, vLLM inference
47 tracked53 tracked · +PyTorch training, FP8 training, vLLM inference, DeepSpeed support, SGLang support, HPU graphs
source - capability
- model
Intel Gaudi 3 added Intel Gaudi 2, Intel Gaudi 3 to its model stack
AcceleratorAccelerator, Intel Gaudi 2, Intel Gaudi 3
source - capability
New capabilities: Quantization (FP8, INT4, NF4), H HCCL
44 tracked46 tracked · +Quantization (FP8, INT4, NF4), H HCCL
source - version
- capability
- version
- version
- capability
- version
- capability
- capability
New capabilities: FP8 / INT4 Quantization, HCCL Collective Communications, Kubernetes Orchestration
37 tracked40 tracked · +FP8 / INT4 Quantization, HCCL Collective Communications, Kubernetes Orchestration
source - capability
New capabilities: vLLM Inference, SGLang Inference
35 tracked37 tracked · +vLLM Inference, SGLang Inference
source - version
- capability
- version
- capability
- capability
- capability
- version
- capability
- version
Intel Gaudi 3 compared
Straight head-to-head pages against the busiest products in Chips & Compute.
How to cite this page
Free to cite and reuse under CC BY 4.0. Permalink: https://tomorrow.aliensquad.ai/tools/intel-gaudi-3
APA
Tomorrow. (2026). Intel Gaudi 3 — version, pricing and model stack [Data set entry]. AlienSquad. Retrieved 2026-09-17, from https://tomorrow.aliensquad.ai/tools/intel-gaudi-3
BibTeX
@misc{tomorrow-tools-intel-gaudi-3,
author = {{Tomorrow}},
title = {Intel Gaudi 3 — version, pricing and model stack},
year = {2026},
publisher = {AlienSquad},
howpublished = {\url{https://tomorrow.aliensquad.ai/tools/intel-gaudi-3}},
note = {Accessed: 2026-09-17}
}