News, Tips & Reviews
B&H Gift Cards
The Professional's Source Since 1973
Ask Our Experts800-606-6969
Hello, Log InAccount & Orders
My Cart
Home

Intel Arc Pro B70 Graphics Card

BH #INAPB7032GB • MFR #33P01IB0BB
7 reviews
Print
Intel Arc Pro B70 Graphics Card
Key Features
  • 2800 MHz Boost Clock Speed
  • 32 Xe Cores | 256 XMX AI Engines
  • Xe2 Architecture | 32 RT Units
  • 32GB of ECC GDDR6 VRAM
Built using the Xe2 microarchitecture, the Intel Arc Pro B70 Workstation Graphics Card is designed to deliver responsive visuals with ray tracing while accelerating AI workloads.
More Details
Top Seller
In Stock
No Store Display
$999.99
$1,079.00
Save $79.01
Limited supply at this price
+
Save the Tax‡
or Pay $84/mo. suggested payments for 12 Mos.§ when you use the Payboo credit card. Learn More
Add to Cart
Free Expedited Shipping
Protect Your Gear
Add a protection plan from $69.99
See Options
Ask an Expert
800-606-6969

Intel B70 Overview

Built using the Xe2 microarchitecture, the Intel Arc Pro B70 Workstation Graphics Card is designed to deliver responsive visuals with ray tracing while accelerating AI workloads.

In addition to featuring 32GB of ECC GDDR6 VRAM with a 256-bit memory interface, this GPU features 32 Xe cores with a 2800 MHz boost clock speed. With 256 XMX AI Engines, the Arc Pro B70 delivers up to 367 TOPS (Trillions of Operations per Second) to upscale graphics, optimize gaming performance, and accelerate generative AI content creation. The 32 ray tracing units provide realistic lighting and reflections with compatible games.

Xe2 Architecture
Based on Xe2 architecture, this GPU features Intel Xe Matrix Extensions (XMX) AI cores and advanced ray tracing units, bringing high-performance capabilities to creators, developers, and engineers.
XeSS 2
Intel's Xe Super Sampling (XeSS) is updated with three key technologies: XeSS Super Resolution, XeSS Frame Generation, and Xe Low Latency.

XeSS Super Resolution offers AI-based upscaling for over 150 supported games. AI-powered XeSS Frame Generation delivers a more fluid visual experience by adding interpolated frames using optical flow and motion vector reprojection. Lastly, Xe Low Latency integrates with the game engine to provide faster responsiveness. With all three technologies active, XeSS 2 can increase the fps output by up to 3.9x compared to the previous generation, delivering high performance with demanding AAA games.

Optimized for AEC
This GPU is optimized for AEC (architecture, engineering, construction) and inference workstations, providing stability and performance through a wide range of independent software vendors (ISV) certifications.
OS Compatibility
This GPU is compatible with consumer and pro drivers on Windows. On Linux, this graphics card supports a containerized software stack to simplify AI deployments and will be progressively upgraded with features and optimizations.
UPC: 735858551984

Intel B70 Specs

Key Specs
GPU Model
Intel Arc Pro B70
Stream Processors
32
Dedicated AI Cores
Yes: 367 TOPS
Interface
PCI Express 5.0 x16
Video RAM
32 GB ECC GDDR6
Video Output
4x DisplayPort 2.1
Power Connection
1x 8-Pin PCI Connector
GPU Width
Dual-Slot
GPU
GPU Model
Intel Arc Pro B70
Stream Processors
32
Clock Speed
Boost Mode: 2800 MHz
Base Mode: 2280 MHz
Dedicated AI Cores
Yes: 367 TOPS
Interface
PCI Express 5.0 x16
Supported APIs
DirectX: 12
Vulkan
OpenGL
OpenCL
Memory
Video RAM
32 GB ECC GDDR6
Memory Interface Width
256-Bit
Memory Bandwidth
608 GB/s
Display Support
Video Output
4x DisplayPort 2.1
Power Requirements
Max Power Consumption
230 W
Power Connection
1x 8-Pin PCI Connector
Compatibility
Expansion Slot Compatibility
Full Height
Height
3.9" / 99.1 mm
Length
10.5" / 266.7 mm
Width
Dual-Slot
OS Compatibility
Windows
General
Cooler Type
Blower-Style Fan
Packaging Info
Package Weight
3.165 lb
Box Dimensions (LxWxH)
15.3 x 8 x 4.1"

Intel B70 Reviews

Great card for AI work!

By Elliott
Rated 5 out of 5
Date: 2026-07-05

DO NOT USE REGULAR VERSION OF COMFY UI. Use ComfyUI through intel AI Playground. That version uses python 3.12 and has all the good intel PyTorch files. Regular ComfyUI will throw nVidia errors even though you are not running an nVidia card (its almost as though nVida is trying to stop people from using intel cards... WEIRD). This card is definitely not as plug and play friendly as nVidia or AMD. But its so much cheaper that its worth it. The intel software has improved so much over the last year it's crazy. Gemini told me that I should "use the latest version of ComfyUI with my B70", which was a total lie. The thermal management and fan speed logic on the card is perfect. The intel software that manages it is very good now (better than AMD). There will definitely be some intel specific things you will have to learn regarding AI workloads but it's worth it in my opinion. If your are not tech savvy and don't know how to tinker with things then definitely don't buy this card. nVidia and AMD are gonna have to lower their prices now, which is great for everyone :)

Its ok BUT yo will have to work with it, ALSO NOT WORTH THE PRICE INCREASE!!!

By J
Rated 4 out of 5
Date: 2026-05-02

As you can see from many of these other comments these people we're having a difficult time setting up this card and all that is true... You will not have a fun time. I'll go ahead and let you know if you're running this card on Windows the fastest I found for single inference speeds is Lama CPP with the Vulkan backend.... If you are running this on Linux SYCL is going to probably give you the faster prompt processing speeds. But to get there you are going to have to climb a mountain. You're going to have to pull from the most recent gets you're going to have to do coding you're going to have to figure all this stuff out on your own No ones going to hold your hand because the stack is immature. Know to before you buy this card and ask yourself if it's worth it. If they keep raising the price on this card I guarantee you this card will not be worth it anymore. I purchased this card for $949 and I'm seeing retailers push it over $1000 that's getting awfully close to the R9700.... If you're the person that's gonna buy 4 of these cards that's almost what a Nvidia DGX spark cost and I'm telling you now... prompt processing is going to be one of your biggest bottlenecks. The DGX Spark is awesome at prompt processing. At the above $1000 price It's already not worth it for ppl buying 4 at once.... Just go ahead and buy A DGX Spark equivalent at that point. This card takes a lot of work to set up It is not going to work with things like Turbo quant... It has a lot of other stuff that is limited to it that even AMD doesn't have currently and the fact that they're already jacking this price up is almost criminal right now. The cards stack is immature In its current state it is not offering an easy user experience at all. Don't think you're buying this card and running any 70B language models on it If you get two of them you're not they don't have anything like NV Link so you get the benefits of tensor parallelism, now you can put them side by side and they do skip the CPU to talk to one another but that's still going to be over PCI So that's still gonna be a huge bandwidth bottleneck. The purpose I created my rig for was to run 30B models and they run a little slowly right now because the whole stack is not optimized, like I said you're not running a 70B model on two cards.... Not with any expectation of real time prompt processing. If the price keeps going up this card is just not going to be worth it BH Photo and Newegg have already increased this card by about 150 to $200... It was the low price point that made this card attractive and the companies and the scalpers are going to take that away from this card and it's going to go straight into the gutter as one of the forgotten things Intel has done again. If Intel was smart They would do fixed price marketing What I'm trying to tell you is do not get hyped up about this card.... It's a card that gives you a window into running local home AI but it is not a card worth the hype It is a lot of work.

See any errors on this page?
Whatever You Need, We've Got It