back to article Quanta opens servers to 100-core Tilera

Upstart multicore, Linux-compatible chip maker Tilera don't need no stinking tier one server makers. That means no IBM, Dell, Hewlett-Packard, or Sun Microsystems. At least not yet. The company has just lined up $25m in its C round of funding, which includes $10m from Quanta Computer, the Taiwanese PC maker that is the volume …


This topic is closed for new posts.


Wouldnt they benefit immensely from this chip?

Silver badge


While I'm less excited about this chip personally because of the lack of floating-point, no doubt it's specifically because this unneeded feature was omitted that they could fit so many cores on a chip, and have such high performance with low power requirements.

For the best possible performance, a chip does need to be specialized to the workload it will be facing.

Thumb Up

I'd like to test drive one...

I expect it's a whole array of simple short-pipelined in-order cores. I wonder if they spared silicon for even SIMD. Did I hear someone say MIPS-like in a prior article?

BTW, lack of an FPU is a bother these days but surely not an absolute deterrent for some problems.

There is, of course, fixed point. Sounds like it's a 32-bitter. If it's a 64 bit even better. If you know the limits of your problem (which can be tricky, I guess) and your integer performance is 'good enough', it may be possible.

However, 'good enough' in this chip may be just good enough to load, store.. push data around.

Hmmm... Interesting nevertheless. Let's support this, if it works as it says on the tin, or just about.

As ever, anything to deny intel the monopoly. Might I suggest to our esteemed El Reg editors the inclusion of an Anti-Intel icon?


Re: Floating-point

I wonder if the architecture even supports FP. Naturally, trapping to an FP library won't be viable for any FP-intense code, but for ordinary code it should suffice. However, if the arch doesn't even have the data types and opcodes to support it, it's going to make it that much more expensive to layer on top.


Floating point.

There are ways and ways; but FP has always been a compromise. It was ages before I convinced one bunch that single precision (out to say,something to 10-38th) was not going to do it) for the numbers they were calculating at (hey!, a preppy! ends! it!).


As for as precision, Integer at 32 beats double at 8; at this point who cares.

Strap-on processors (What? don't they call them that?).

The 8087 line could come back to life; I remember wanting an upgrade for my old Amstrad so bad.

(Which does kind of date me and when you date a computer you are riding with the naughty bits.)

Or just use one of the new GPU boards for your co-processor.

(OK, is it just insane or have I seen that 'God is my co-processor' sticker on a co-workers computer in the last ten years? Damn, how did I miss that?)

Software emulation.

Less said the better here; it probably can be done but I have lost money on this in the past.



Absolutely grand for LAMP boxes, tho you'd wonder whether there'd be somewhere to do the mod_ssl and ssh FP in hardware .. reminds me of 486DX ;)

This topic is closed for new posts.


Biting the hand that feeds IT © 1998–2017