Re: Crashing ARM's party?
Obviously having some armless fun.
Imagination Technologies has decided today – right in the middle of its arch rival ARM's annual technical conference in California – is the best day to announce some new chip designs. As a friendly reminder that ARM doesn't have the world of embedded processors completely sewn up, Imagination will unveil the 64-bit MIPS P6600 …
This post has been deleted by its author
it has a 16-stage pipeline with multi-issue out-of-order execution, can scale from single core to hexacore
I am not a chip guy but that sounds to me a bit like the P4 pipeline which required a lot of silicon and performance often suffered with o-o-o (it's damn hard to get this right). How much "general purpose" do you need for this where the option of dedicated (let's say encryption or something similar) units wouldn't be a better choice? Okay, this puts the onus on the compiler and the programmer to produce the right code for the right unit but so far this seems to be beating hardware heuristics.
Just like MHz is a useless measure of CPU power between different architectures, so is pipeline depth.
x86 is a horrendous instruction architecture that needs amazing complexity to execute fast. No wonder we're seeing 20 + stage pipelines.
ARM is relatively simple and thus can execute really well on short pipelines.
O-O-O means Out-Of-Order.
Thanks, yes I know all about O-O-O then, I just didn't recognise the acronym!
From what I can gather that's one of the things that ARM's big.LITTLE can differentiates between: a device ticks over using a low power consumption non-O-O-O core, but switches to a high-performance core when needed.
... I can't see how if a company as big as Silicon Graphics was in its day couldn't make this architecture popular enough for it to sell itself , a small company can do it now. I wish them all the best but I can't see them being the next ARM to be honest.
MIPS couldn't come out with processor improvements fast enough to keep far enough ahead of x86 to justify the premium cost of SGIs.
We had a bunch of SGI machines (workstations and servers) back in the late 80's/early 90s and switched over to Linux in the mid 90's.
We also made the switch from SGI and Sun to x86 and Linux in the late '90s. I don't think the problem was so much architectural improvements, but the cost of building state of the art foundries. Intel had the money from the Wintel duopoly to build the best fabs.
Then SGI committed suicide by dumping MIPS and going with the Itanic.
Despite my grandmother's cautionary response to any gossip: "Were you there?", I suspect I am not the only person who remembers things a bit differently. Never an SGI employee, but a customer both of SGI computers and MIPS processors (for embedded systems). What it looked like from the outside was that some suit in a corner office bet the company on WinNT/(Itanium?), quite possibly without running that idea past the folks in the trenches. I don't recall the exact timing, but SGI also sold the "no use to us" line of Sparc-based Cray supers (that they got bundled with other Cray IP) to Sun, thus kickstarting some of their most serious competition. Reminds me of when Intel fobbed off that "useless" high-performance StrongArm they accidentally acquired from the ashes of DEC, to Marvel.
There's a wide variety of stupid at the top of many (formerly) successful companies. Rarely do they die of a single cause.