Slashdot Mirror
HP Finally Reveals The Alpha Marvel
brejc8 writes "HP have revealed the new range of AlphaServer systems. The new EV7 processors show very reasonable performance figures. Revealed by the inquirer the 1GHz versions have very similar SPEC scores as the 1GHz Itanium 2 (INT_2000 of 875 and FP_2000 of 1,500). This is very intersting after HP were rumoured to ensure that "...no Alpha benchmark will be released until the Itanium platform(s) is/are faster"."
they should, all current versions of the alpha procs run linux great. (as verified by the alphaserver 5000 sitting under my desk running RH 7.2)
"Two things are infinite: the universe and human stupidity; and I'm not sure about the the universe." --Albert Einstein
... at least on OpenMP type applications. Cribbed shamelessly from realworldtech.com:
...
SPECOMP2001 results, base/peak:
4 cpu:
EV7/1150: 6027/6824
I2/1000: 3762/4091
8 cpu:
EV7/1150: 10349/11929
POWER4+/1450: 9458/ 9694
PA8700+/875: 4375/ 4541
16 cpu:
EV7/1150: 17724/20637
PA8700+/875: 7763/ 8788
R14k/600: 7265/ 7726
Note that this is not a pure CPU test (like SpecINT/FP), but rather a test of SMP performance. Looks like the tin-foil hat "Wait 'til EV8!" brigade might have been on to something
'jfb
To spur "enterprise Linux," Big Bang, the distributed two-phase commit.
Real nicly too....everyone who's demo'd one has drolled at it.
They need 'grout' to fill in the space between people currently on Alpha (or someone needing better performance NOW), before Itanium comes to the point in which they are the in thing. Alpha has MANY VERY loyal customers that would drop ship if they didn't have something to fill the space until Itanium comes of age. AMD doesn't have anything on EV8, it is EV6 they have and licensed technology for. Intel does however have rights to use everything from EV7 and EV8.
Actually you shouldnt assume that. I am British and thats how I spell rumoured rumoured.
Mouse powered Chips, Open source Processors and Lego
I may work for HP, but that does not imply that my opinions are theirs.
Info on SpecOMP, just in case anyone's interested. Also, here's a snippet from the FAQ:
Q3: What components does SPEC OMP measure?
A3: Since the benchmarks are designed to reflect applications requiring compute-intensive parallel processing, they measure performance of the computer's processors, memory architecture, operating system, and compiler. It is important to remember the contribution of the latter three components.
'jfb
To spur "enterprise Linux," Big Bang, the distributed two-phase commit.
It appears they have a program in place to cost effectivly move the Alpha customers to the new Itanium systems when they come out.
They are calling it their Customer Assurance Initiative
HP will introduce Superdome systems running Madison processors (Itanium2 shrink) this year (most likely scaling to 64way). Someone from HP also mentionned that they plan to put 2 Madison processors on a board, so we may see 128 way systems (with reduced bandwidth/proc) before Intel comes out with two-cores processors. Until then HP only sells up to 4 way systems. HP used to resell NEC's Azusa system (up to 16 Itanium (not Itanium2)), however it made no sense to buy one.
an implication that they are planning on keeping the Alpha platform long-term
You are reading too much in that sentence. Alpha systems are being phased out. For HP, the future of the high-end is Itanium only (phasing out Alpha and PA-RISC in the next 5 years). IIRC there should be one more iteration of Alpha processors (EV79?), then nothing new, just support.
they were probably well into working on the itanic when the option to buyout alpha came along.
Its a sad state of affairs when the superior architecture gets cut up and sold to different companies to produce two slightly inferior chips.
yes, it is. and disregarding alpha for a moment, you would think after 20 so years of the pile of crap known as x86, that intel would be intelligent enough to make clean, sane cpu. instead they, of course, design the itanic. i've read about its isa and all i can say is "feature bloat". i also read a little of the hp book about porting linux. the itanic is the most overly complicated, misdesigned cpu i think has ever been made. at least when the 8086 came out, it was a good design (relatively speaking).
it's funny how intel says "epic is simple, no ooo complexity" but doesn't mention the all rediculous crap like rotating register files, etc, etc. afaict, ia64 is MORE complex than any risc chip. NOT simpler. and throwing ooo out the door is stupid. a) compilers can't predict cache misses b) gcc sucks and so, to get anywhere near decent performance, you have to use a different compiler (dec's cc, and i think just about everyone else's, outperforms gcc). i predict that intel will be forced to eventually add ooo back. at best, intel has traded ooo complexity for the complexity of all the features needed for compiler driven scheduling, AND forced compilers to be very good just to get decent performance.
That's probably going to be the single biggest factor in deciding which 64-bit server CPU dominates the marketplace.
Linux on UltraSparc works great, and has excellent support under Debian (although I guess that's no surprise).
HP has claimed that their SuperDome was designed to heterogenously support either PA-RISC or IA-64 cpu modules, in the same frame.
I have some co-workers who were in Palo Alto and say they saw a running system alomost two years ago at HP HQ.
HP's roadmap for ageas has been that Itanium based SuperDomes are on the way real soon. The supposed release for a 64-way Itanium 2 last I heard was Q102...so they have a couple of months to deliver or change the date again.
A great deal of Alpha architect left when Compaq bought Digital. They went mostly to AMD, making the Athlon faster (the main-design was already done), and their influence is also seen readily in the Sledgehammer design.
It's a bit hard to find just by greping the source. Since it's been designed mainly for i386, you will see some code done in assembly. In other cases, they have used some other workarounds specifically to avoid using floating point. Sure, it'll take some retooling, but MPlayer should work much faster on non-i386 platforms when modified to take advantage of floating point.
.ogg player) is still just a CVS snapshot... The current players and encoders use float.
In addition, I was just checking out Vorbis, and ``Tremor" (the int-only
Not that it matters, the video is what takes so much time to encode. It's that this simple fact blows your credibility to hell.
Slashdot gets worse every day... Pipedot: News for nerds, without the corporate slant
The sad news is no EV8. Itanium is far from being debugged and doesn't seem to be a particularly clean architecture compared with Alpha and Intel aren't particularly innovative.
I said typically.
The fact that your particular place of work happens to have it figured out is no contradiction to the general case.
And you're talking local switching. In banking operations, you have remote hot standby in case your datacenter burns down or something else really bad happens (both COs you're connected to die at once, for instance).
With remote hot standby, switching and switching back is often (note the often) much more painful.
In case you still don't get it, note that switching implies that one of your datacenters is DOWN and you are now on a completely separate system with separate disk drives, communications links, etc. Switching back means that you have to bring everything back up, sync it, and fail back again.
Sure, it works. Is it fun? No. Is testing it and retesting it under every failure condition under a new OS port and processer architecture fun?
Um no.