From WikiChip
Difference between revisions of "intrinsity/fastmath"
(→Documents) |
|||
Line 84: | Line 84: | ||
* [[:File:Intrinsity Symbol Rate Processing.pdf|Symbol Rate Processing]] | * [[:File:Intrinsity Symbol Rate Processing.pdf|Symbol Rate Processing]] | ||
* [[:File:Intrinsity Timing of Single Threaded Application.pdf|Timing of Single Threaded Application]] | * [[:File:Intrinsity Timing of Single Threaded Application.pdf|Timing of Single Threaded Application]] | ||
+ | |||
+ | === Others === | ||
+ | * [[:File:FastMATH.ppt|FastMATH Presentation]] |
Latest revision as of 17:41, 3 July 2016
FastMATH was a family of matrix and vector math processors with an on-die RISC CPUs introduced by Intrinsity. The chips were developed using Intrinsity's own proprietary Fast14 technology.
Contents
Architecture[edit]
- Main article: FastMATH Microarchitecture
FastMATH was a series of microprocessors developed by Intrinsity using Fast14 technology - i.e. processors designed using custom dynamic domino logic. These chips incorporate the FastMIPS core along with a custom high-performance matrix and vector math coprocessor.
Matrix and Vector Math Processing Unit[edit]
The unit is designed as a single-instruction, multiple-data (SIMD) architecture capable of oeprating on 4x4 arrays of 32-bit values. Operates support fixed-point matrix, vector, and scalar data types with dedicated local register file. Data arrays are directly fetched from L2 cache.
- Zero-cycle latency, two-cycle throughput
- 64 GOPS (peak) at 2 GHz
- 551,000 radix-4 1024-point 16-bit FFTs/sec at 2 GHz
- 32 GMACs/sec at 2 GHz
![]() |
This section requires expansion; you can help adding the missing info. |
Members[edit]
FastMATH Processors | |||||||||
---|---|---|---|---|---|---|---|---|---|
Model | µarch | Process | Launched | Frequency | Power | Memory | VCORE | ||
FastMATH-1.5 | FashMATH | 130 nm 0.13 μm 1.3e-4 mm | 2002 | 1.5 GHz 1,500 MHz 1,500,000 kHz | 13.5 W 13,500 mW 0.0181 hp 0.0135 kW | 1,024 MiB 1,048,576 KiB 1,073,741,824 B 1 GiB 9.765625e-4 TiB | 1 V 10 dV 100 cV 1,000 mV | ||
FastMATH-2 | FashMATH | 130 nm 0.13 μm 1.3e-4 mm | 2002 | 2 GHz 2,000 MHz 2,000,000 kHz | 15 W 15,000 mW 0.0201 hp 0.015 kW | 1,024 MiB 1,048,576 KiB 1,073,741,824 B 1 GiB 9.765625e-4 TiB | 1 V 10 dV 100 cV 1,000 mV | ||
FastMATH-3 | FashMATH | 130 nm 0.13 μm 1.3e-4 mm | 3 GHz 3,000 MHz 3,000,000 kHz | 1,024 MiB 1,048,576 KiB 1,073,741,824 B 1 GiB 9.765625e-4 TiB | 1.25 V 12.5 dV 125 cV 1,250 mV | ||||
FastMATH-LP | FashMATH | 130 nm 0.13 μm 1.3e-4 mm | 2003 | 1 GHz 1,000 MHz 1,000,000 kHz | 6 W 6,000 mW 0.00805 hp 0.006 kW | 1,024 MiB 1,048,576 KiB 1,073,741,824 B 1 GiB 9.765625e-4 TiB | 0.85 V 8.5 dV 85 cV 850 mV | ||
Count: 4 |
Documents[edit]
Manuals[edit]
White Paper[edit]
- Advanced Processing Techniques
- Chip Rate Processing
- Computed Tomography
- FixedPoint
- Hard Copy Imaging
- RACH Preamble Detection
- Symbol Rate Processing
- Timing of Single Threaded Application
Others[edit]
Facts about "FastMATH - Intrinsity"
designer | Intrinsity + |
first announced | 2000 + |
first launched | 2002 + |
full page name | intrinsity/fastmath + |
instance of | microprocessor family + |
main designer | Intrinsity + |
manufacturer | TSMC + |
name | FastMATH + |
package | CBGA-670 + |
process | 130 nm (0.13 μm, 1.3e-4 mm) + |
technology | CMOS + |
word size | 32 bit (4 octets, 8 nibbles) + |