New L Series

Technical discussions around xCORE processors (e.g. xcore-200 & xcore.ai).
User avatar
JohnWilson
Active Member
Posts: 38
Joined: Fri Oct 28, 2011 10:53 pm
Contact:

Post by JohnWilson »

mikronauts wrote:where threads per core < 8 some threads just appear to be disabled; and I would not be surprised at all if the limitation is based purely on the processor model number selected in the IDE perhaps enforced by checking a processor-internal fuse/rom setting.
Whatever else it is, it can't be as simple as an IDE hack, since the "getr rX,XS1_RES_TYPE_THREAD" instruction has to be in on it. I'm fantasizing that they've added more execution-unit slots so that you get *six* instructions every 8/10 nsec instead of only four, but even if it's only a crippled version of the regular quad-issue core I'm sure the point is that it's priced accordingly so you only pay for what you use and XMOS can reach more customers. It'll be a little surprising if that flies (it seems to me the economics would pull in opposite directions) but, whatever!

I also hope this doesn't distract them from getting the (*@#^*% SU01A/U8A out the door for real! Last I heard was January 2013 for it to start appearing at Digikey. I, um, can't wait!


mikronauts
Member++
Posts: 28
Joined: Sun Aug 28, 2011 5:18 am
Contact:

Post by mikronauts »

I am guessing that "getr rX,XS1_RES_TYPE_THREAD" just returns some ROM or fuse bits.

Six instructions every 8/10ns would be wonderful, however I suspect that such a significant change has not been made.

I do like the hints about a 700Mhz speed grade :) and look forward to USB enabled versions with integrated PHY... and ADC...

I would LOVE to see a "16 core" XMOS TQFP-128 chip at 700Mhz+ (two hard cores, eight threads each) for the simple reason that it is easier to work with (for me at least).

Soapbox mode:

Basically I find it distasteful whenever manufacturers (AMD, Intel, Microchip, Atmel et al) deliberately cripple parts to sell at a lower cost... which costs them more to do than to NOT cripple them.

Perhaps I am old fashioned, but the way I'd address different market segments is to bring out newer more powerful products, and reduce the prices on the older products, package with different number of I/O's, and have several different guaranteed speed grades.
JohnWilson wrote:
mikronauts wrote:where threads per core < 8 some threads just appear to be disabled; and I would not be surprised at all if the limitation is based purely on the processor model number selected in the IDE perhaps enforced by checking a processor-internal fuse/rom setting.
Whatever else it is, it can't be as simple as an IDE hack, since the "getr rX,XS1_RES_TYPE_THREAD" instruction has to be in on it. I'm fantasizing that they've added more execution-unit slots so that you get *six* instructions every 8/10 nsec instead of only four, but even if it's only a crippled version of the regular quad-issue core I'm sure the point is that it's priced accordingly so you only pay for what you use and XMOS can reach more customers. It'll be a little surprising if that flies (it seems to me the economics would pull in opposite directions) but, whatever!

I also hope this doesn't distract them from getting the (*@#^*% SU01A/U8A out the door for real! Last I heard was January 2013 for it to start appearing at Digikey. I, um, can't wait!
Post Reply