On 14-7-2010 17:17, Darren New wrote:
> Invisible wrote:
>> And my point is "where the **** do you get a system that can handle 
>> more than 16GB of RAM from?!"
> 
> http://tinyurl.com/25z4y9r
> 
or e.g. http://www.sgi.com/pdfs/4185.pdf

Post a reply to this message

Le 14/07/2010 17:51, Jim Henderson nous fit lire :
> On Wed, 14 Jul 2010 09:11:04 +0100, Invisible wrote:
> 
>>> Sure, but the system may well have 128 GB of memory installed in it
>>> (for example), and only 23 GB allocated to that specific machine -
>>> that's my point.
>>
>> And my point is "where the **** do you get a system that can handle more
>> than 16GB of RAM from?!"
> 
> GIYF. ;-)
> 
> Hint:  It's not your typical off-the-shelf desktop PC.  We're talking 
> very specialized hardware here.

Well, a 24GB system is expensive but main-street now.
(I built one in june): based on i7 implied triple channel memory, and
therefore most single i7-supporting motherboard are limited to 24GB.
Finding a set of 6 Ram-stick to fill the slots is rather easy too, as
memory-maker/seller make kit ready to install (and with warranty).

Motherboard: about 200€
Memory: about 800€
CPU: anywhere from 200 to 1000€
Box: from 50€ to 200€ (according to the cooling and other
easy-life-making: hand-screw, clip-on...)
and so on for the small details (mouse, keyboard, graphics card(s),
display...)

There you go, a system with more than 16GB.

Post a reply to this message

> GIYF. ;-)
> 
> Hint:  It's not your typical off-the-shelf desktop PC.  We're talking 
> very specialized hardware here.

What like this:

http://www.ebuyer.com/product/173974

Post a reply to this message

Darren New wrote:

> It was funny, tho. I read someone talking about all the "nosql" 
> databases, and how people were complaining SQL doesn't "scale".  He said 
> "SQL scales fine. What the people mean is that SQL doesn't scale if you 
> want to use a commodity piece of hardware like Amazon rents. This is 
> what we have in my office for testing:" and gives the stats on a 96-CPU 
> 256G RAM SQL processor. He says "That's the low-end base machine for 
> testing, not the production system."

Presumably that's not x86 though.

Post a reply to this message

Invisible wrote:
> Presumably that's not x86 though.

I didn't look into it that deeply. Why would you think it's not?

-- 
Darren New, San Diego CA, USA (PST)
    C# - a language whose greatest drawback
    is that its best implementation comes
    from a company that doesn't hate Microsoft.

Post a reply to this message

>> Presumably that's not x86 though.
> 
> I didn't look into it that deeply. Why would you think it's not?

I got the impression that x86 CPUs are designed to communicate with only 
a fixed number of peers.

For example, a socket 939 Athlon can only be run in a single-chip 
configuration. But a socket 940 Opteron has an extra pin which enables 
it to run in single-chip or dual-chip configuration. (But not 
triple-chip or more.) In later Opteron generations, part of the model 
number tells you what the maximum number of chips you can wire together 
is. I gather Intel Xeon chips use a similar system.

Given that the reviews talk about 1-chip, 2-chip, 4-chip and even 
8-chip, I very much doubt anybody has bothered to design a 96-chip part. 
Who would they sell it to, after all? (And wouldn't it need 96 extra pins??)

Post a reply to this message

Darren New <dne### [at] sanrrcom> wrote:
>
http://aws.typepad.com/aws/2010/07/the-new-amazon-ec2-instance-type-the-cluster-compute-instance.html

> Rent a quad-core 2HT 2.93GHz 23G RAM 1.5TB disk compute engine for $1.60/hour.

  I wonder if POV-Ray 3.7 could be compiled for this monster:
http://en.wikipedia.org/wiki/Jaguar_%28computer%29

  It has more cores than the maximum number of threads POV-Ray would use for
an average-sized image (iow. you basically get a core for each square to which
the image is subdivided by POV-Ray for rendering).

  A giant poster-sized image could perhaps require rendering more than one
square per core...

-- 
                                                          - Warp

Post a reply to this message

Warp wrote:
>   A giant poster-sized image could perhaps require rendering more than one
> square per core...

I think the first job should be rendering a 300 DPI 1:1 scale image of the 
computer it ran on. :-)

-- 
Darren New, San Diego CA, USA (PST)
    C# - a language whose greatest drawback
    is that its best implementation comes
    from a company that doesn't hate Microsoft.

Post a reply to this message

On 7/14/2010 12:09 PM, Darren New wrote:
> Jim Henderson wrote:
>> Well, yeah - but I'm not going to find it going to the local Best Buy,
>> that's what I mean. :-)
>
> It is indeed not a "Personal Computer" sort of thing. :-)
>

http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=6426353&Sku=E145-2082

I'm thinking one that supports 4-way SLI is probably a gaming system. ;)

Supports up to 48GB of ram, dual hex-core xeons ... yeah..

A bit pricey, though.

-- 
~Mike

Post a reply to this message

On 7/15/2010 10:32 AM, Invisible wrote:
>>> Presumably that's not x86 though.
>>
>> I didn't look into it that deeply. Why would you think it's not?
>
> I got the impression that x86 CPUs are designed to communicate with only
> a fixed number of peers.

http://www.cray.com/Products/XT/Specifications.aspx

Opterons ... x86 ;)



http://www.ewiz.com/detail.php?p=MB-H8DGUF&c=fr&pid=ea8e96795a53689a8306610b250b775faaf92905edde9d71f42a3e02a2c3931c

So, its a dual chip mobo ... unless you make it yourself like Cray ...
24 cores is probably enough.

> Given that the reviews talk about 1-chip, 2-chip, 4-chip and even
> 8-chip, I very much doubt anybody has bothered to design a 96-chip part.
> Who would they sell it to, after all? (And wouldn't it need 96 extra
> pins??)

Cray has, but then, it's a purpose-built machine.
-- 
~Mike

Post a reply to this message