🚀 go-pugleaf

Of course, the subject should have read "Building a Photoshop PC", but
I'm sure you already guessed that!

In the writings of Al Treacher, the <rqqoa0179fjs7rscf3ddmacufuccf33q32@
4ax.com> scrolls contained these prophetic words:

To give you even more headaches read this article below. Now is *not* a 
good time to build a new machine too many new motherboard designs and 
technology are coming out. Intel have their new socket design due out at 
the end of the year, SATA 2 and PCI-Express is due out as well...

The article is a comparison of Intel/AMD processors and their abilities.

http://techreport.com/reviews/2004q2/opteron-x50/index.x?pg=1

Personally? I am waiting till beginning of next year before I build 
another system to a workstation specification. That way I avoid the 
'sting' of new technology, and the the prices should have stabilised.

Oh and I am running a Athlon 64, very fast system compared to Intel's 
chips... 

Oh and before people accuse that site of being AMD biased, just check 
other reviews, Opteron seems to beat Xeons hands down in most 
disciplines, and those that it doesn't are Intel optimised and we have 
yet to see how the 64Bit optimisations affect both Intel and AMD.
-- 
Scottish Heritage:
http://www.CelticShadows.co.uk

In my experience the four biggest contributors to PS performance
are, in order:

1. CPU speed
2. Memory speed
3. Disk subsystem speed
4. CPU cache size

Don't get too hung up with all the multiprocessor/hyperthreading
marketing glitz.  PS is multiprocessor aware but will use a second
CPU only for a very limited number of operations (e.g. certain
filters).  One is usually better off with a single cpu system with a
faster FSB.  There are exceptions to this, e.g. if the system will
not be a dedicated PS workstation.

Rick

"Al Treacher" <big.al@stones.com.NOSPAM> wrote in message news:rqqoa0179fjs7rscf3ddmacufuccf33q32@4ax.com...
> Hi folks,
>
> The PC I currently run Photoshop in is now a very long way past its
> "best by" date, so the time has come to build a new one.  I've sought
> advice from a couple of techie friends about the basic system
> architecture and I've got essentially three direction that I could
> take, it seems.  Neither of my "advisors" use Photoshop regularly, so
> I'd like to get opinions from other people that do...
>
>
> Option 1:
> Dual-processor motherboard with a pair of 2.8GHz Intel Xeon CPUs.
>
> Considerations:
> Two processors, each with HyperThread tech, so in theory I'd get a
> performance equivalent to maybe three and a bit individual processors.
> **IF** Photoshop supports more than two processors - I know it
> supports two, but would it support (essentially) four?
>
> Also, it looks like the Front-Side-Bus speed of Xeons is low compared
> to the other Intel processors available, at 533MHz or possible even
> only 400MHz.  I'm guessing this is because of the necessity to
> synchronize more threads, but I could be wrong.
>
> Would this result in Photoshop running slower than a single processor
> system running a 800MHz FSB?
>
> I'm having problems finding the details of the cache size for Xeons.
> It looks like there is an average (512KB) level 2 cache, with a 2MB
> level 3 cache on some chips, but finding which ones seems impossible!
> Any ideas?
>
>
> Option 2:
> Single Pentium 4C (Prescott) 3.2GHz CPU.
>
> HT tech again, so essentially a dual processor on a single CPU.
> 800MHz Front-Side-Bus.
> 1MB of level 2 cache.
>
> Seems like a good all-round option, but I'd like to find out if the
> dual Xeons would outperform this when running Photoshop.
>
>
> Option 3:
> Single Pentium 4C (Northwood) 2.8GHz CPU overclocked to about 3.2GHz.
> (Using PC4000 DDR500 RAM.)
>
> Raising the FSB speed to 900+MHz would increase the speed the core was
> running at, and also give faster RAM transfer.  How much of a
> performance increase (if any) would be noticeable?
>
> Of course, there's the slight hit-and-miss of getting a chip that will
> reliably overclock and keeping it cool, but I'd deal with that...!
>
>
>
> I'd thought this was going to be a relatively easy project and I'd
> have had the thing up and running now...  How naive of me...
>
> Your thoughts/observations/experiences and opinions are welcome!
>
> Cheers,
> Al
>

"Al Treacher" <big.al@stones.com.NOSPAM> wrote in message
news:rqqoa0179fjs7rscf3ddmacufuccf33q32@4ax.com...
> Hi folks,
>
> The PC I currently run Photoshop in is now a very long way past its
> "best by" date, so the time has come to build a new one.  I've sought
> advice from a couple of techie friends about the basic system
> architecture and I've got essentially three direction that I could
> take, it seems.  Neither of my "advisors" use Photoshop regularly, so
> I'd like to get opinions from other people that do...
>
>snip<


Sounds like overkill in the three cases.
If your PC is a tool you use for your trade AND if you are very busy you
"could" need something that fast.
Like someone said here not long ago, a few seconds gained on rendition of
filters can add up and make you more productive at the end of the day.But
are you really that busy that seconds count?
Working with photoshop is made much easier with a multiple display system
and a good graphic tablet.
I would get two or three very good monitors and a calibrating tool instead
of gamers motherboards and chips.


Stephan

On Thu, 20 May 2004 09:44:58 +0100, Al Treacher
<big.al@stones.com.NOSPAM> wrote:

>Hi folks,
>
>The PC I currently run Photoshop in is now a very long way past its
>"best by" date, so the time has come to build a new one.  I've sought
>advice from a couple of techie friends about the basic system
>architecture and I've got essentially three direction that I could
>take, it seems.  Neither of my "advisors" use Photoshop regularly, so
>I'd like to get opinions from other people that do...
>
>
>Option 1:
>Dual-processor motherboard with a pair of 2.8GHz Intel Xeon CPUs.
>

>Option 2:
>Single Pentium 4C (Prescott) 3.2GHz CPU.
>

>Option 3:
>Single Pentium 4C (Northwood) 2.8GHz CPU overclocked to about 3.2GHz.
>(Using PC4000 DDR500 RAM.)
>
Or you could do Option 4 and have either a single or dual Athlon
system in which case you'd have a faster PC, but there you go...

 --

Hecate
Hecate@newsguy.com 
veni, vidi, reliqui

On Thu, 20 May 2004 09:44:58 +0100, Al Treacher
<big.al@stones.com.NOSPAM> wrote:

>Hi folks,
>
>The PC I currently run Photoshop in is now a very long way past its
>"best by" date, so the time has come to build a new one.  I've sought
>advice from a couple of techie friends about the basic system
>architecture and I've got essentially three direction that I could
>take, it seems.  Neither of my "advisors" use Photoshop regularly, so
>I'd like to get opinions from other people that do...
>
>
>Option 1:
>Dual-processor motherboard with a pair of 2.8GHz Intel Xeon CPUs.
>
>Considerations:
>Two processors, each with HyperThread tech, so in theory I'd get a
>performance equivalent to maybe three and a bit individual processors.
>**IF** Photoshop supports more than two processors - I know it
>supports two, but would it support (essentially) four?

If your applications are multithreaded and support more than one
processor you will be fine, but even with applications that do not
support multiple processors a dual processor system is fine if you
find one application keeping one processor busy while the other
handles OS and anything else (Oracle 7 database on a two processor
system worked like that).

Fo all other considerations there is a simple preference list:

1) Available real memory (and accessing speed)
2) Processor speed (no longer relevant once you run out of memory)
3) Virtual memory speed (put scratch and pagefile on RAID-0)

1) Graphic apps live on available memory. If you have a single huge
file or a smaller one with lots of layers doesn't matter, you will end
up using large chunks of memory. If you have not enough memory you
will fall from RAM-speed (several 2-digit-nanoseconds access time) to
disk-speed (1-digit-milliseconds at best; 100.000 times slower).
Access speed and caching is less relevant, more memory is always
better than going to the disk drive, even if more memory may be slower
than less memory on some systems.

2) Most current processors are able to process more data than their
memory interface can provide. FSB 800 is promising but still: getting
the hardware that gives you the last five percent of performance will
never pay.

3) There will be a day when you run out of memory. If you have 2 GB
this day will come later, but it will come. 
If you are prepared, you will drink less coffee than otherwise. 
Even if RAID-0 is a high risk configuration that I'd never recommend
for storing important data, it is perfect for high speed access to
pagefile and scratch files - contents that will be gone when you close
the application or switch off the system.

On the other hand: If you have multiple disks then distributing
pagefile and scratch files over as many of them as possible will
probably give better results. Graphic apps tend to have certain access
patterns for reading contiguous chunks of data from one location while
writing it to some other location - if source and target reside on the
same disk, this will result in slow transfer and heavy disk activity. 
Source and target on different drives reduces positioning activity and
enables higher transfer speed.

And, by the way, there is no need for a high speed graphics card.

Michael

"Xalinai" <xalinai_Two@xalinai.de> wrote in message news:40adbaee.669052187@news.t-online.de...
> 2) Most current processors are able to process more data than their
> memory interface can provide. FSB 800 is promising but still: getting
> the hardware that gives you the last five percent of performance will
> never pay.

Promising?  Five percent?  Where have you been the last year?

FSB 800MHz pumps twice as much data to a cpu (or cpus) as
400MHz, and 50% more than 533MHz.

Rick

Rick wrote:
> "Xalinai" <xalinai_Two@xalinai.de> wrote in message news:40adbaee.669052187@news.t-online.de...
> 
>>2) Most current processors are able to process more data than their
>>memory interface can provide. FSB 800 is promising but still: getting
>>the hardware that gives you the last five percent of performance will
>>never pay.
> 
> 
> Promising?  Five percent?  Where have you been the last year?
> 
> FSB 800MHz pumps twice as much data to a cpu (or cpus) as
> 400MHz, and 50% more than 533MHz.
> 
> Rick

Where did you get those figures from? In the real world it will not be 
quite as cut and dried as that, it may reach that at burst rates but not 
continually.

Stuart

On Fri, 21 May 2004 02:01:03 -0700, "Rick" <me@privacy.net> wrote:

>"Xalinai" <xalinai_Two@xalinai.de> wrote in message news:40adbaee.669052187@news.t-online.de...
>> 2) Most current processors are able to process more data than their
>> memory interface can provide. FSB 800 is promising but still: getting
>> the hardware that gives you the last five percent of performance will
>> never pay.
>
>Promising?  Five percent?  Where have you been the last year?
>
>FSB 800MHz pumps twice as much data to a cpu (or cpus) as
>400MHz, and 50% more than 533MHz.

Sure. 

But it does so for how much of the total code processed? And it does
so in what address mode?

How does a graphics application work? Adressing pixels, 3 bytes per
pixel. PS is processing data bytewise or double-byte-wise (in 16 bit
mode) adressing multiple pixels in a grid that makes memory locations
of pixels in adjacent rows three times the image width apart (Filters,
sharpen, blur). This isn't very much the thing that high speed burst
mode was designed for.

You won't get much more than five to ten percent on application level
from the faster memory bus on the average system. There will be more
if memory access is the major bottleneck in two (otherwise similar)
competing systems for specific applications (like Video streams).

Michael

"Stuart" <stuart@nospam.uk> wrote in message news:c8khs8$lvt$2@sunbeam.coventry.ac.uk...
> Rick wrote:
> > "Xalinai" <xalinai_Two@xalinai.de> wrote in message news:40adbaee.669052187@news.t-online.de...
> >
> >>2) Most current processors are able to process more data than their
> >>memory interface can provide. FSB 800 is promising but still: getting
> >>the hardware that gives you the last five percent of performance will
> >>never pay.
> >
> >
> > Promising?  Five percent?  Where have you been the last year?
> >
> > FSB 800MHz pumps twice as much data to a cpu (or cpus) as
> > 400MHz, and 50% more than 533MHz.
>
> Where did you get those figures from? In the real world it will not be
> quite as cut and dried as that, it may reach that at burst rates but not
> continually.

With the current P4, which can't do much (or in many cases,
anything) with bus speeds higher than 533MHz, that's correct.
But that will be changing in the very near future.

Rick