Larrabee Based On a Bundle of Old Pentium Chips 286
arcticstoat writes "Intel's Pat Gelsinger recently revealed that Larrabee's 32 IA cores will in fact be based on Intel's ancient P54C architecture, which was last seen in the original Pentium chips, such as the Pentium 75, in the early 1990s. The chip will feature 32 of these cores, which will each feature a 512-bit wide SIMD (single input, multiple data) vector processing unit."
What the hell is Larrabee? (Score:5, Insightful)
A little context might help. This isn't the Inquirer for god's sake.
good. (Score:5, Insightful)
on the basis that..
the design is well known, understood and has had rigorous testing in the field
they will no doubt fix any understood errors firstlimits the RnD to the multicore section
as long as the chip performs well for the silicon overhead then they should feel free to cram as many in as they want.
seems perfectly sensible to me.
Re:What the hell is Larrabee? (Score:5, Insightful)
Re:What the hell is Larrabee? (Score:5, Insightful)
Not only is the power retarded, but ATI already can do 100% native ray tracing [techpowerup.com] which crushed intel bigtime.
I welcome intel trying to push for marketshare but it's going to be many generations before intel can play catchup on graphics cards...specifically when we get around to 32+GB of ram and you can afford a couple gigs for graphics (at which point we'll need 4+ gigs for graphics probably), the performance of an integrated solution will still be lacking. Graphics bandwidth and needs increases far exponentially beyond that of processing needs for anything graphics intensive by definition (currently).
Re:Pentium 75? (Score:4, Insightful)
I know I'm wasting my time responding to such a blatant troll, but they're nothing hypocritical about saying that the original Pentium 1 was a pretty bad chip, and the Core 2 Duo is a pretty great one.
Failing to reliably perform basic floating point ops is pretty embarrassing. But Intel's come a long way since then.
Comment removed (Score:5, Insightful)
Marketing Math (Score:3, Insightful)
I don't see how they get to 2 TFLops.
512-bit = 64 bit * 8 way SIMD or 32 bit * 16 way SIMD. Let's go with the bigger of these two and say we are performing 16 single Floating point operations per clock-cycle per core. 16 operations per clock-core * 32 cores * 2 Billion clocks per second = 1024 Single Precision GFlops. It looks more like 512 Double Precision GFlops for 300 Watts which means a DP Teraflop on Larabee will cost you 513 Dollars a Year [google.com] at 10 cents/kWH. If we're considering single precision, we can cut this in half to 257 dollars per years per single precision teraflop.
Compare to Clearspeed which offers 66 DP GFLops at 25 Watts costing 332 dollars [google.com] for a sustained DP teraflop for a year.
even the NVidia Tesla has better performance at single precision: you can buy 4 SP TFlops consuming only 700W or 5.7 GFLops/Watt, for an annual power budget of 153 dollars [google.com].
Yes, "based on" seems to be the key phrase (Score:4, Insightful)
Obviously they're not just going to slap a bunch of Pentium cores on there and call it good. But the high-level design can probably start off with the P54, and just rip out stuff that doesn't need to be supported, possibly including:
Scalar floating-point, 16-bit protected mode, real mode, operand size overrides, segment registers, the whole v86 mode, the i/o address space, BCD arithmetic, virtual memory, interrupts, #LOCK, etc, etc.
Once you've done that, you'll have a much simpler model to synthesize down to an implementation. And with a slightly-modified compiler spec, you can crank out code for it with existing compilers, like ICC and GCC.
Re:Marketing Math (Score:2, Insightful)
I don't see how they get to 2 TFLops. 512-bit = 64 bit * 8 way SIMD or 32 bit * 16 way SIMD. Let's go with the bigger of these two and say we are performing 16 single Floating point operations per clock-cycle per core. 16 operations per clock-core * 32 cores * 2 Billion clocks per second = 1024 Single Precision GFlops.
Most likely there is a muladd unit, which would double the peak FLOPS.
2TFLOPS of processing (Score:2, Insightful)
Oh, so 2 years from now (two lifetimes in the GPU business) Intel will be releasing a chip comparable to this month's ATI HD 4870 X2.
Re:What the hell is Larrabee? (Score:3, Insightful)
Slashdot covers a wide variety of very technical topics, they can't be expected to elaborate on them all.
Granted, but is the following too much to ask?
"Intel's Pat Gelsinger recently revealed that the 32 IA cores in Larrabee, Intel's planned multi-core GPU, will in fact be based on Intel's ancient P54C architecture"
It's the difference between being an editor, and being a trained chimp that has learned to click a POST button in exchange for bananas. It has, of course, been long established that slashdot's "editors" run more along the "chimp" side.
Re:Pentium 75? (Score:5, Insightful)
they're nothing hypocritical about saying that the original Pentium 1 was a pretty bad chip, and the Core 2 Duo is a pretty great one.
Have you compared the total length of Pentium errata with the length of the Core 2 Duo errata?
Re:I doubt it (Score:3, Insightful)
Intel's basically doing here what Sun did with the Niagara series, but without concurrent threading. I suppose it wouldn't be too tough to add it in, though. The cores in the Niagara are really simple 6 or 7 stage pipelines. They don't do any forwarding, and stall at pretty much every hazard they hit. Instead of adding all the complicated circuitry needed for do advanced pipeline stuff (like forwarding and OoO etc), they just defer execution to a new thread. All the threading is in the cores themselves, so there's no need for OS intervention.
I should add that the Niagara's are pretty awful when it comes to single threaded performance, though. I use a T2 daily and starting up firefox can take about 30 or 40 seconds. But they're great if you can manage to parallelize your programs (which is easier said than done).
Re:Pentium 75? (Score:1, Insightful)
Re:Pentium 75? (Score:4, Insightful)
Making math errors at blazing speeds...
To err is human.
To really screw up, you need the aid of a computer.
Re:Pentium 75? (Score:3, Insightful)
/Every/ time you divided? Bull-SHEET. It only affected a certain few pairs of operands.
http://en.wikipedia.org/wiki/Pentium_FDIV_bug [wikipedia.org]
Executive summary: You're a moron who repeats things he doesn't understand.
Re:I'm no expert but (Score:3, Insightful)
Re:Pentium 75? (Score:3, Insightful)
Re:Pentium 75? (Score:3, Insightful)
I guess back then it would have been difficult to notice. If you were using that 75Mhz beast for nothing but floating point processes it would produce about one error every 117 seconds. I'm sure there are/were a lot of applications (not standard users) that would have really been affected by the flaw. Add in multiple core servers and you could have some pretty hefty issues.
Pretty crazy that these days an error like that would rear its ugly head ever 3-4 seconds (Folding@Home or SETI?).