No one got fired for using Intel?

  • Thread starter Thread starter YKhan
  • Start date Start date
If the new CEO is smart and wants to keep his job, he should very
quickly consider building Opteron servers. The recently fired CEO must
be the third one who stayed loyal to the Intel vision for far too long.


Yousuf Khan
 
YKhan said:
If the new CEO is smart and wants to keep his job, he should very
quickly consider building Opteron servers. The recently fired CEO must
be the third one who stayed loyal to the Intel vision for far too long.


Yousuf Khan
What's the value add that SGI can contribute to an Opteron Server? They
had Numalink or whatever for the big numa Itanium and Mips boxes. I
guess they could put on high end graphics and sell them as workstations.

del
 
The whole industry should blame themselves for folding their hands at
the mere announcement of Intel's new "super chip".

You have _no_ idea how close to the truth you are!
 
What's the value add that SGI can contribute to an Opteron Server? They
had Numalink or whatever for the big numa Itanium and Mips boxes. I
guess they could put on high end graphics and sell them as workstations.

....and what is it they do? ...and do they have any other real choice?
 
...and what is it they do? ...and do they have any other real choice?

Yes, they have a choice to try and sell themselves while SGI name
still holds some residual value. The rest of their business probably
has a negative value. Being delisted doesn't help to prop up the
business, either. Too bad - it used to be a big name.
Sic transit...

NNN
 
Yes, they have a choice to try and sell themselves while SGI name
still holds some residual value. The rest of their business probably
has a negative value. Being delisted doesn't help to prop up the
business, either. Too bad - it used to be a big name.
Sic transit...

They have to have *something*, other than a name. Their name is a tad
tarnished these days.
 
Del said:
What's the value add that SGI can contribute to an Opteron Server? They
had Numalink or whatever for the big numa Itanium and Mips boxes. I
guess they could put on high end graphics and sell them as workstations.

Yeah, that's exactly what I was thinking of, they could adapt their
NUMAlink to Opteron. So far, none of the announced Opteron NUMA
solutions have shown up yet. Not Newisys' Horus, nor the Serverworks
chipset that was supposed to have some form of Sun Microsystems IP in
them; in fact, Sun released their Galaxy servers with Nvidia chipsets in
them instead.

Yousuf Khan
 
Keith said:
They have to have *something*, other than a name. Their name is a tad
tarnished these days.
They have a nice building in Chippawa Falls, just down the street from
the Leiney Lodge and Leinenkugel brewery, and the Big Eddie Spring.

I presume there are still people working in that building as well.

del
 
YKhan said:
If the new CEO is smart and wants to keep his job, he should very
quickly consider building Opteron servers. The recently fired CEO must
be the third one who stayed loyal to the Intel vision for far too long.

You do realize that Opterons cannot address enough physical memory for
most of SGI's applications, right? 1TB limit currently. SGI systems
go upto 10TB+...I think the largest I read about was a 14TB system in
japan, but I'm not 100% sure.

Now that could change, we'll have to wait and see what's up with the
F-stepping I guess.

DK
 
David said:
You do realize that Opterons cannot address enough physical memory for
most of SGI's applications, right? 1TB limit currently. SGI systems
go upto 10TB+...I think the largest I read about was a 14TB system in
japan, but I'm not 100% sure.

1TB can be installed per processor (40-bit), but 256TB overall (48-bit)
is addressable virtually. So if you have 256 Opterons, each with 1TB of
memory attached to them, then you'll have reached your limit.

But I don't think even SGI's have 1TB of RAM per processor yet.

Yousuf Khan
 
So I did a little reading at www.cray.com and www.sgi.com ...

SGI Altix clusters can theoretically go up to 128 TB.
Cray XT3 (Opteron) clusters can theoretically go up to 240 TB.
1TB can be installed per processor (40-bit), but 256TB overall (48-bit)
is addressable virtually. So if you have 256 Opterons, each with 1TB of
memory attached to them, then you'll have reached your limit.

But I don't think even SGI's have 1TB of RAM per processor yet.

SGI's max RAM solution is by clustering Altix 4700's.
One or two CPUs and up to 48 GB of DDR-2 per blade.

Cray's XT3, IIRC, has single-CPU blades with up to 8 GB/CPU.

HP sells 4P Opty blades with up to 64 GB/blade but I haven't read
anything about them being used in any super-sized clusters.
 
Back
Top