Article: ISSCC 2005: The CELL Microprocessor
By: Ping-Che Chen (pcchen.delete@this.shts.seed.net.tw), February 12, 2005 1:43 am
Room: Moderated Discussions
Ricardo Bugalho (ricardo.b@zmail.pt) on 2/11/05 wrote:
---------------------------
>IIRC, IEEE 854 generalizes IEEE 754 for other bases than binary. So, while it's
>problably a typo, it's still correct. :)
>
Ah yes, 854 includes both base 2 and base 10. However, it does not specify any specific format for numbers (such as the number of bits, exponent bias, etc). Only 754 has specific formats. I hope this does not mean that CELL's DP is not in the same format as normal 754 DP ;)
If CELL has about 25 ~ 30 GFLOPS capability in DP, it would be much more balanced in DP mode than in SP mode, since it has about 1 byte memory bandwidth for each FLOP, roughly the same as many contemporary RISC CPU.
---------------------------
>IIRC, IEEE 854 generalizes IEEE 754 for other bases than binary. So, while it's
>problably a typo, it's still correct. :)
>
Ah yes, 854 includes both base 2 and base 10. However, it does not specify any specific format for numbers (such as the number of bits, exponent bias, etc). Only 754 has specific formats. I hope this does not mean that CELL's DP is not in the same format as normal 754 DP ;)
If CELL has about 25 ~ 30 GFLOPS capability in DP, it would be much more balanced in DP mode than in SP mode, since it has about 1 byte memory bandwidth for each FLOP, roughly the same as many contemporary RISC CPU.
| Topic | Posted By | Date |
|---|---|---|
| Cell article is live | David Kanter | 2005/02/11 07:00 AM |
| Cell article is live | Ping-Che Chen | 2005/02/11 11:50 AM |
| Cell article is live | Ricardo Bugalho | 2005/02/11 01:13 PM |
| Cell article is live | Ping-Che Chen | 2005/02/12 01:43 AM |
| awesome read, very well done, David | mas | 2005/02/12 10:30 AM |
| awesome read, very well done, David | David Wang | 2005/02/12 10:56 AM |
