News - Intel at last ready to break into parallel computing with Xeon Phi

Quote:

---

One-Teraflop of FP64 double-precision performance and 8GB+ GDDR5 RAM.

---

Read more.

50 Enhanced Pentium 1 cores... so that's 50.0000000000000000000112394 :)

Quote:

---

Originally Posted by d032sh

50 Enhanced Pentium 1 cores... so that's 50.0000000000000000000112394 :)

---

You win the internet today :-)

i tried parallel comuting once, didnt know whether I was coming or going :(


(pointing out spelling errors in a vaguely humorous way since 2006)

From the image - 'synergistically' - What the hell sort of rubbish is that. Sounds like made up marketing words to me...

Quote:

---

Originally Posted by cheesemp

From the image - 'synergistically' - What the hell sort of rubbish is that. Sounds like made up marketing words to me...

---

Oh, those marketing guys - they probably think that sort of talk is cutting edge :p

Bah, I have 80 cores on my desk.

But... it doesn't run Linux. And it only does 40GIPS (integer instructions not floating point ops).

Syngergistic talk is probably designed to counter AMD fusion talk with onboard GPU being part of the CPU.

Doesn't look like much of an EMI shield out back...

Still, nice to finally get this from Intel.

Also talking about pentiums 1s (I had a P60 as my first PC!) - does the idea of a coprocessor remind anyone of the 386 days when you needed a coprocessor for floating point. It good to see nothing changes...

Quote:

---

Originally Posted by cheesemp

Does the idea of a coprocessor remind anyone of the 386 days when you needed a coprocessor for floating point.

---

The x87 co-processor? It's still there, sorta. You can certainly decide to use it or otherwise with the gcc compiler flag -mfpmath=387 :) It has slightly different floating point accuracy characteristics when compared to the SSE floating-point units, so I found.

Quote:

---

Originally Posted by Fraz

The x87 co-processor? It's still there, sorta. You can certainly decide to use it or otherwise with the gcc compiler flag -mfpmath=387 :) It has slightly different floating point accuracy characteristics when compared to the SSE floating-point units, so I found.

---

Actually the i{3,4,5,6}86 targets still defaults to 387, you have to explicitly set -mfpmath=sse to enable SSE intrinsics on 32bit machines, I haven't noticed any ill effects running it globally on Gentoo. The x86_64 port made the break to SSE by default.

I'd be interested in seeing how this coprocessor card will be implemented in practice, whether it is actually a coprocessor at all for a start. The description appears to be more discrete.

Quote:

---

Originally Posted by aidanjt

Actually the i{3,4,5,6}86 targets still defaults to 387, you have to explicitly set -mfpmath=sse to enable SSE intrinsics on 32bit machines

---

You still use 32-bit machines?! :surprised: ;)

Quote:

---

Originally Posted by Fraz

You still use 32-bit machines?! :surprised: ;)

---

Shocking, I know :lol:

I have an ole atom netbook and a really old 'Pentium Dual Core' laptop, so yeah, no x86_64 on those. :(