Speedster®7t FPGAs are optimized for high-bandwidth workloads and
eliminate the performance bottlenecks associated with traditional FPGAs.
The
Speedster7t FPGA family is optimized for high-bandwidth workloads and
eliminates the performance bottlenecks associated with traditional
FPGAs. Built on the TSMC 7nm FinFET process, Speedster7t FPGAs feature a
revolutionary new 2D network-on-chip (2D NoC), an array of new machine
learning processors (MLPs) optimized for high-bandwidth and artificial
intelligence/machine learning (AI/ML) workloads, high-bandwidth GDDR6
interfaces, 400G Ethernet and PCI Express Gen5 ports. The 2D NoC
connects all of the interfaces to over 80 access points in the FPGA
fabric to deliver ASIC-level performance while retaining the full
programmability of FPGAs. Get started today with the VectorPath
accelerator card, featuring the Speedster7t FPGA.
|
2D Network on Chip (2D NoC)
|
High-Speed Data Interfaces
|
Machine Learning Processors (MLPs)
|
-
>20 Tbps bandwidth
-
Interconnects, I/O, memory, internal functional blocks and FPGA fabric.
-
Reduces routing congestion vs. traditional FPGAs by over 40%
-
Simplifies FPGA design
|
-
PCIe Gen5
-
112 Gbps SerDes channels
-
400G Ethernet
-
GDDR6 memory
-
Up to eight controllers
-
Up to 4 Tbps bandwidth
|
-
Native support for integer, floating point, bfloat16 and block floating point
-
Speeds up to 750 MHz with co-located block RAM
-
Fully fracturable integer multiplier/accumulator
|
|
|
|
|
Speedster7t Product Table
|
Feature
|
AC7t700(1)
|
AC7t800
|
AC7t1400(1)
|
AC7t1500
|
|
Logic Density
|
|
System Logic Cells (SLCs)
|
730k
|
730k
|
1,546k
|
1,546k
|
|
6-Input Look-Up Tables (LUTs)
|
326k
|
326k
|
692k
|
692k
|
|
|
|
32-bit (INT)(3)
|
864
|
864
|
2,560
|
2,560
|
|
16-bit (FP or TensorFlow bFloat)(4)
|
1,728
|
1,728
|
5,120
|
5,120
|
|
24-bit (FP)
|
1,728
|
1,728
|
5,120
|
5,120
|
|
16-bit (INT)
|
3,456
|
3,456
|
10,240
|
10,240
|
|
8-bit/6-bit (INT or FP)
|
13,824
|
13,824
|
40,960
|
40,960
|
|
4-bit (INT or Block FP)
|
27,648
|
27,648
|
81,920
|
81,920
|
|
On-Chip Memory
|
|
Total Memory
|
86 Mb
|
86 Mb
|
195 Mb
|
195 Mb
|
|
Local RAM (LRAM) Blocks (2.3 kb)
|
864
|
864
|
2,560
|
2,560
|
|
Block RAM (BRAM) Blocks (72 kb)
|
1,152
|
1,152
|
2,560
|
2,560
|
|
High-Performance Interfaces
|
|
112 Gbps SerDes
|
24
|
24
|
32
|
32
|
|
DDR4/5 Memory Channels
|
1 (DDR4/DDR5)
|
1 (DDR4/DDR5)
|
1 (DDR4)
|
1 (DDR4)
|
|
GDDR6 Memory Channels
|
6 (1.5 Tbps)
|
6 (1.5 Tbps)
|
16 (4 Tbps)
|
16 (4 Tbps)
|
|
PCIe (Gen5)
|
1 (×8)
|
1 (×16)
|
1 (×8)
|
2 (×8 and ×16)
|
|
Ethernet (10 Gbps to 400 Gbps)
|
8 lanes
400G (max)
|
8 lanes
2 × 400G (max)
|
16 lanes
400G (max)
|
16 lanes
2 × 400G (max)
|
|
Security
|
|
Inline Cryptography (25G)
|
Hard IP
|
Hard IP
|
None
|
Soft IP
|
|
Performance
|
|
2D NoC bandwidth (Tbps)
|
12
|
12
|
21
|
21
|
|
ML TOps: INT-8 or bFloat-16(5)
|
20.5
|
20.5
|
61
|
61
|
|
Packaging
|
|
42.5 × 42.5 mm BGA
|
1,677 balls
|
1,677 balls
|
-
|
-
|
|
52.5 × 52.5 mm BGA
|
-
|
-
|
2,597 balls
|
2,597 balls
|
|
Table Notes
-
AC7t700 and AC7t1400 SerDes total device bandwidth limited to 500 Gbps
-
Resources per number format
-
INT - Integer
-
FP - Floating point
-
ML - Machine learning
Next-Generation FPGAs from Achronix
The next generation of Achronix FPGA technology brings many innovations to the FPGA industry including:
-
Power and performance optimized 2D NoC technology providing fast data transport across the entire FPGA and die-to-die interconnects
-
Enhanced DSP and ML processor arrays offering low-power, high-performance compute capabilities
-
New, state-of-the-art, security and encryption features to enable fully secure programming and communication protocols
-
Support for the highest performance SerDes (up to 112 Gbps), Ethernet (2 x 400 GE) and PCIe (Gen5)
-
Support for the highest bandwidth external memory interfaces (GDDR6 and DDR5)
Request More Information
Our experts are happy to advise you on how Achronix can help with your toughest design challenges.
|
|
