Processor upgrade

[Alfred Molon]

I'm wondering how much faster an Intel i5 or i7 processor is than the
Intel core 2 duo p8600 processor with my Lenovo X200 notebook uses for
what concerns image processing. The image processing software can make
use of multiple cores.
--

Alfred Molon
------------------------------
Olympus E-series DSLRs and micro 4/3 forum at
http://tech.groups.yahoo.com/group/MyOlympus/
http://myolympus.org/ photo sharing site

[Eric Stevens]

On Sat, 11 May 2013 23:53:46 +0200, Alfred Molon
<alfred...@yahoo.com> wrote:

>I'm wondering how much faster an Intel i5 or i7 processor is than the
>Intel core 2 duo p8600 processor with my Lenovo X200 notebook uses for
>what concerns image processing. The image processing software can make
>use of multiple cores.

I can tell you from personal experience in a different setup from your
own that the i7 is blazingly faster than the Intel core 2 duo p8600.
However, it's not just a question of a faster processor but the
ability of the hardware to support the more powerful processor.

Your should first run the test described in
http://windows.microsoft.com/en-nz/windows-vista/what-is-the-windows-experience-index
to discover where the bottleneck is in your machine.
--

Regards,

Eric Stevens

[Alan Browne]

On 2013.05.11 17:53 , Alfred Molon wrote:
> I'm wondering how much faster an Intel i5 or i7 processor is than the
> Intel core 2 duo p8600 processor with my Lenovo X200 notebook uses for
> what concerns image processing. The image processing software can make
> use of multiple cores.

The i5 and (especially) the i7 are much faster.

But there are other issues that slow down a laptop depending on its
architecture. For example, some laptops share system memory with the
graphics card. So the system is slower (pages more often) and the
graphics response is slow because it's using ordinary memory.

More recent processors use faster memory - and more recent graphics
cards also have higher speed graphics oriented memory.

IOW - it's the system - not just the processor.

That's not even the tip of the iceberg.

--
"A Canadian is someone who knows how to have sex in a canoe."
-Pierre Berton

[PeterN]

Agreed. It can also get crowded on the bus.

--
PeterN

[Alfred Molon]

In article <dlhto8tufd1a80ngf...@4ax.com>, Eric Stevens
says...

>
> I can tell you from personal experience in a different setup from your
> own that the i7 is blazingly faster than the Intel core 2 duo p8600.
> However, it's not just a question of a faster processor but the
> ability of the hardware to support the more powerful processor.
>
> Your should first run the test described in
> http://windows.microsoft.com/en-nz/windows-vista/what-is-the-windows-experience-index
> to discover where the bottleneck is in your machine.

In that test the processor gets a score of 5.3, not far from the highest
possible value of 5.9. The lowest score is 3.8 is of Grafik (Spiele)
(i.e. graphics/games). But what does all this mean? Would a PC with the
i7 get a score of 25?

[Alfred Molon]

In article <BoSdnS1gp4CBTBPM...@giganews.com>, Alan Browne
says...

> The i5 and (especially) the i7 are much faster.

How much faster - is there a comparison table?

> But there are other issues that slow down a laptop depending on its
> architecture. For example, some laptops share system memory with the
> graphics card. So the system is slower (pages more often) and the
> graphics response is slow because it's using ordinary memory.
>
> More recent processors use faster memory - and more recent graphics
> cards also have higher speed graphics oriented memory.
>
> IOW - it's the system - not just the processor.
>
> That's not even the tip of the iceberg.

Well, in my current machine during certain tasks the CPU load indicator
in the task manager goes to 100% for both cores and stays at 100% until
the task is completed. So I would assume that the processor is indeed
the bottleneck.

[Floyd L. Davidson]

Alfred Molon <alfred...@yahoo.com> wrote:
>
>Well, in my current machine during certain tasks the CPU load indicator
>in the task manager goes to 100% for both cores and stays at 100% until
>the task is completed. So I would assume that the processor is indeed
>the bottleneck.

With image processing that is fairly common, and indeed
is probably the most annoying and longest waiting time
when there is a noticeable bottleneck.

The solution is not to look for the fastest CPU as such,
but rather look for the chip with the most cores. Even
if each core is half as fast as those on some other
chip, the times when the speed of any one or two cores
is the problem are rare. And when a process can be
spread out over multiple cores... there ain't nothing
like a pair of 6 core chips!

I currently use two different Dell 7500 system for image
processing. One has a pair of 6-core CPU's and the other
has a pair of 4-core CPU's.

There are two places where the difference is very
significant for my workflow. I have a shell script that
rotates all preview JPEG images for proper orientation,
and another that is used to batch process (per
interactively generated configuration files) RAW images
to TIFF's. Both of these scrips work on 1 image per CPU
core. The 12 core machine finishes about 10 times as
fast as a script doing each image sequentially. A 2
core machine where each core was twice as fast as these
cores wouldn't be even half as fast at those particular
tasks.

--
Floyd L. Davidson http://www.apaflo.com/
Ukpeagvik (Barrow, Alaska) fl...@apaflo.com

[Eric Stevens]

On Sun, 12 May 2013 09:47:25 +0200, Alfred Molon
<alfred...@yahoo.com> wrote:

>In article <dlhto8tufd1a80ngf...@4ax.com>, Eric Stevens
>says...
>>
>> I can tell you from personal experience in a different setup from your
>> own that the i7 is blazingly faster than the Intel core 2 duo p8600.
>> However, it's not just a question of a faster processor but the
>> ability of the hardware to support the more powerful processor.
>>
>> Your should first run the test described in
>> http://windows.microsoft.com/en-nz/windows-vista/what-is-the-windows-experience-index
>> to discover where the bottleneck is in your machine.
>
>In that test the processor gets a score of 5.3, not far from the highest
>possible value of 5.9. The lowest score is 3.8 is of Grafik (Spiele)
>(i.e. graphics/games). But what does all this mean? Would a PC with the
>i7 get a score of 25?

There appears to be something screwy here. I've got an i7 and I get a
score for that of 7.6. I also get 7.6 for memory. For Graphics I get
7.0. Gaming graphics is 7.0. Why I said it appears to be screwy is
that the site I cited says "The base scores currently range from 1 to
5.9.". However mine has a maximum of 7.9. They have obviously updated
it.

The final score of mine is for the Primary Hard Disk of 5.9 and that
is the performance given to my machine. In other words, if I want to
get the full benefit of the i7 I should speed up my graphics and my
primary hard disk. Both of those would require new hardware.
--

Regards,

Eric Stevens

[Alfred Molon]

In article <87r4hcz...@pix.apaflo.com>, Floyd L. Davidson says...

> I currently use two different Dell 7500 system for image
> processing. One has a pair of 6-core CPU's and the other
> has a pair of 4-core CPU's.
>
> There are two places where the difference is very
> significant for my workflow. I have a shell script that
> rotates all preview JPEG images for proper orientation,
> and another that is used to batch process (per
> interactively generated configuration files) RAW images
> to TIFF's. Both of these scrips work on 1 image per CPU
> core. The 12 core machine finishes about 10 times as
> fast as a script doing each image sequentially. A 2
> core machine where each core was twice as fast as these
> cores wouldn't be even half as fast at those particular
> tasks.

But there aren't any notebook PCs with two CPUs, are there?

Anyway, how much faster would an i7 compared to the Core 2 Duo P8600 be?

[Alfred Molon]

In article <8emuo81ppnsb6sapv...@4ax.com>, Eric Stevens
says...

> There appears to be something screwy here. I've got an i7 and I get a
> score for that of 7.6. I also get 7.6 for memory. For Graphics I get
> 7.0. Gaming graphics is 7.0. Why I said it appears to be screwy is
> that the site I cited says "The base scores currently range from 1 to
> 5.9.". However mine has a maximum of 7.9. They have obviously updated
> it.

So if my processor gets 5.3 and your 7.6, how much faster is yours than
mine (for instance for batch processing 100 RAW images)?

[nospam]

In article <87r4hcz...@pix.apaflo.com>, Floyd L. Davidson

<fl...@apaflo.com> wrote:

> >Well, in my current machine during certain tasks the CPU load indicator
> >in the task manager goes to 100% for both cores and stays at 100% until
> >the task is completed. So I would assume that the processor is indeed
> >the bottleneck.
>
> With image processing that is fairly common, and indeed
> is probably the most annoying and longest waiting time
> when there is a noticeable bottleneck.

image processing is usually i/o bound, not cpu bound.

that's why photoshop users want the fastest disks they can get, as much
memory as they can get, etc.

video processing, on the other hand, is often cpu bound.

> The solution is not to look for the fastest CPU as such,
> but rather look for the chip with the most cores. Even
> if each core is half as fast as those on some other
> chip, the times when the speed of any one or two cores
> is the problem are rare. And when a process can be
> spread out over multiple cores... there ain't nothing
> like a pair of 6 core chips!

depends on the software. some software benefits from more cores and
other software does not.

[nospam]

In article <MPG.2bf9abd15...@news.supernews.com>, Alfred

Molon <alfred...@yahoo.com> wrote:

> But there aren't any notebook PCs with two CPUs, are there?

these days, most laptops have 2 cores and some have 4.

your core 2 duo has 2 cores.

> Anyway, how much faster would an i7 compared to the Core 2 Duo P8600 be?

a lot.

[Martin Brown]

On 12/05/2013 08:47, Alfred Molon wrote:
> In article <dlhto8tufd1a80ngf...@4ax.com>, Eric Stevens
> says...
>>
>> I can tell you from personal experience in a different setup from your
>> own that the i7 is blazingly faster than the Intel core 2 duo p8600.
>> However, it's not just a question of a faster processor but the
>> ability of the hardware to support the more powerful processor.

Comparing a mobile CPU in a portable against a desktop is a little
unfair. My Samsung has an i7-3630QM and is quite nippy but nowhere near
as fast as a full desktop high end i7 CPU which can afford to be less
frugal with power and dump more heat to run faster.

>>
>> Your should first run the test described in
>> http://windows.microsoft.com/en-nz/windows-vista/what-is-the-windows-experience-index
>> to discover where the bottleneck is in your machine.
>
> In that test the processor gets a score of 5.3, not far from the highest
> possible value of 5.9. The lowest score is 3.8 is of Grafik (Spiele)
> (i.e. graphics/games). But what does all this mean? Would a PC with the
> i7 get a score of 25?

You need to be aware that there are different versions of the Windows
confusion index in different versions of their OS releases. The numbers
it gives are almost meaningless on their own.

The various benchmarks that Windoze does are mostly irrelevant to image
processing. A more realistic hard CPU throughput benchmark is online at:

http://www.cpubenchmark.net/high_end_cpus.html

In a desktop I would probably go for an i5-3570K at the moment although
the i7-3770K (which I actually do have) is faster but hyperthreading
isn't as big a boost as you might hope.

Note these benchmarks *only* hold good if the application you intend to
run is fully multithread multiCPU aware. Plenty of code is good with up
to 4 CPUs but benefits much less from extra ones after that!

Either would be exceptionally fast in desktop machines - especially when
combined with 8G ram and a Samsung 840 SSD. For some work on large
datasets a fast SSD will give old hardware a new lease of life! YMMV

--
Regards,
Martin Brown

[Alfred Molon]

In article <YaPjt.8753$KI6....@newsfe09.iad>, Martin Brown says...

> The various benchmarks that Windoze does are mostly irrelevant to image
> processing. A more realistic hard CPU throughput benchmark is online at:
>
> http://www.cpubenchmark.net/high_end_cpus.html

And here are the notebook processor benchmarks:
http://www.cpubenchmark.net/laptop.html

The P8600 shows up with a score of 1505, while a typical Core i7 for a
notebook is around 4500, i.e. three times faster.

[Floyd L. Davidson]

nospam <nos...@nospam.invalid> wrote:
>In article <87r4hcz...@pix.apaflo.com>, Floyd L. Davidson
><fl...@apaflo.com> wrote:
>
>> >Well, in my current machine during certain tasks the CPU load indicator
>> >in the task manager goes to 100% for both cores and stays at 100% until
>> >the task is completed. So I would assume that the processor is indeed
>> >the bottleneck.
>>
>> With image processing that is fairly common, and indeed
>> is probably the most annoying and longest waiting time
>> when there is a noticeable bottleneck.
>
>image processing is usually i/o bound, not cpu bound.
>
>that's why photoshop users want the fastest disks they can get, as much
>memory as they can get, etc.

If Photoshop users see it that way, they are wrong. If
you spend any time at all editing an image, the editing
time vastly out distances the time it takes to read and
later write the image to disk.

>video processing, on the other hand, is often cpu bound.

And so is image editing in the way that I described above:
the annoying waits that are noticed and frustrating.

>> The solution is not to look for the fastest CPU as such,
>> but rather look for the chip with the most cores. Even
>> if each core is half as fast as those on some other
>> chip, the times when the speed of any one or two cores
>> is the problem are rare. And when a process can be
>> spread out over multiple cores... there ain't nothing
>> like a pair of 6 core chips!
>
>depends on the software. some software benefits from more cores and
>other software does not.

Which was already noted, so you are nagging and not
helping.

[nospam]

In article <87ehdcy...@pix.apaflo.com>, Floyd L. Davidson

<fl...@apaflo.com> wrote:

> >> >Well, in my current machine during certain tasks the CPU load indicator
> >> >in the task manager goes to 100% for both cores and stays at 100% until
> >> >the task is completed. So I would assume that the processor is indeed
> >> >the bottleneck.
> >>
> >> With image processing that is fairly common, and indeed
> >> is probably the most annoying and longest waiting time
> >> when there is a noticeable bottleneck.
> >
> >image processing is usually i/o bound, not cpu bound.
> >
> >that's why photoshop users want the fastest disks they can get, as much
> >memory as they can get, etc.
>
> If Photoshop users see it that way, they are wrong. If
> you spend any time at all editing an image, the editing
> time vastly out distances the time it takes to read and
> later write the image to disk.

photoshop users see a dramatic increase in performance when they have
more memory and faster hard drives. the cpu is *not* the bottleneck.

adobe has repeatedly said that to optimize photoshop to be as fast as
possible, give it fast drives and memory. they have a much better idea
than you do about its bottlenecks.

cpus have greatly increased in speed and the number of cores, but i/o
hasn't kept pace.

most image processing does simple calculations (e.g., adjust colour
balance) on a *lot* of pixels and moving them in and out of the cpu is
the bottleneck.

on the other hand, if you spend your days doing cpu intensive filters
such as radial blur, then it might be different but few people do that
more than occasionally, if they do at all.

> >video processing, on the other hand, is often cpu bound.
>
> And so is image editing in the way that I described above:
> the annoying waits that are noticed and frustrating.

nope. video editing involves more complex calculations, unless you're
doing something simple such as adjusting colour balance on all frames.

[Me]

On 13/05/2013 12:39 a.m., Alfred Molon wrote:
> In article <87r4hcz...@pix.apaflo.com>, Floyd L. Davidson says...
>> I currently use two different Dell 7500 system for image
>> processing. One has a pair of 6-core CPU's and the other
>> has a pair of 4-core CPU's.
>>
>> There are two places where the difference is very
>> significant for my workflow. I have a shell script that
>> rotates all preview JPEG images for proper orientation,
>> and another that is used to batch process (per
>> interactively generated configuration files) RAW images
>> to TIFF's. Both of these scrips work on 1 image per CPU
>> core. The 12 core machine finishes about 10 times as
>> fast as a script doing each image sequentially. A 2
>> core machine where each core was twice as fast as these
>> cores wouldn't be even half as fast at those particular
>> tasks.
>
> But there aren't any notebook PCs with two CPUs, are there?
>
> Anyway, how much faster would an i7 compared to the Core 2 Duo P8600 be?
>

As an example, I've got two notebooks here, one with Core 2 Duo, one
with Core i7. Nominal core speed is about the same for both.
Converting and processing D300 (12mp) raw files with the Core 2 duo is a
little slow, also with some annoying lag when making processor intensive
adjustments once the image file is loaded - but it is usable. With D800
raw files (36mp) it was so slow and laggy that I don't consider it
practically usable - it would drive me nuts.
The core i7 machine handles D800 raw file conversion /much/ faster than
the core 2 duo handles D300 raw files.
When ripping / transcoding DVDs for backup, the Core i7 is about 8x
faster than the Core 2 duo, so it's at least 4x faster than a simple
calculation of core speed x number of cores would suggest (the core i7
is many generations ahead of the core 2 duo).

[Eric Stevens]

On Sun, 12 May 2013 14:41:14 +0200, Alfred Molon
<alfred...@yahoo.com> wrote:

>In article <8emuo81ppnsb6sapv...@4ax.com>, Eric Stevens
>says...
>> There appears to be something screwy here. I've got an i7 and I get a
>> score for that of 7.6. I also get 7.6 for memory. For Graphics I get
>> 7.0. Gaming graphics is 7.0. Why I said it appears to be screwy is
>> that the site I cited says "The base scores currently range from 1 to
>> 5.9.". However mine has a maximum of 7.9. They have obviously updated
>> it.
>
>So if my processor gets 5.3 and your 7.6, how much faster is yours than
>mine (for instance for batch processing 100 RAW images)?

I don't know the answer. Clearly my processor is faster than yours but
the overall performance would be held back by my slower disk. I've got
8GB of RAM but that may not be enough for 100 RAW images. The overall
performance may be held back by the speed of disk transfers.

The other thing is that I don't know whether the Windows performance
scale is linear, logarithmic or what, so what going from 5.3 to 7.6
means is beyond me.
--

Regards,

Eric Stevens

[Alan Browne]

On 2013.05.12 03:51 , Alfred Molon wrote:
> In article <BoSdnS1gp4CBTBPM...@giganews.com>, Alan Browne
> says...
>
>> The i5 and (especially) the i7 are much faster.
>
> How much faster - is there a comparison table?

http://www.cpubenchmark.net/

>
>> But there are other issues that slow down a laptop depending on its
>> architecture. For example, some laptops share system memory with the
>> graphics card. So the system is slower (pages more often) and the
>> graphics response is slow because it's using ordinary memory.
>>
>> More recent processors use faster memory - and more recent graphics
>> cards also have higher speed graphics oriented memory.
>>
>> IOW - it's the system - not just the processor.
>>
>> That's not even the tip of the iceberg.
>
> Well, in my current machine during certain tasks the CPU load indicator
> in the task manager goes to 100% for both cores and stays at 100% until
> the task is completed. So I would assume that the processor is indeed
> the bottleneck.

Computer load indicators are funny things. Your CPU can be sitting and
waiting for memory and indicate full load. It could even be i/o bound
but the code you're running is written to loop on poll rather than
return control to the OS. That makes it look like it's "working" when
it's just actively waiting. You just can't really tell.

In threaded cores it can be very misleading - an i7 quad core can show
well over 800%. But the fact is the threaded cores can only do about
30% more than a solo core. So it shows 800% that is really 520% of real
processing, or so. OTOH, they also over-clock withing thermal limits
which gives it a bit more "real" work. All that to say that 10% of CPU
load is probably an accurate indicator of nothing much happening but a
100% indication is probably misleading and does not mean, necessarily,
that it is the bottleneck.

Computers today are incredibly complex.

And laptops are usually end-to-end compromises of low-value and/or
low-power consumption parts. It all adds up to compromises that rob the
processor. High end portables have fewer compromises and better
performance. Of course you pay for it.

[Alan Browne]

On 2013.05.12 08:39 , Alfred Molon wrote:
> In article <87r4hcz...@pix.apaflo.com>, Floyd L. Davidson says...
>> I currently use two different Dell 7500 system for image
>> processing. One has a pair of 6-core CPU's and the other
>> has a pair of 4-core CPU's.
>>
>> There are two places where the difference is very
>> significant for my workflow. I have a shell script that
>> rotates all preview JPEG images for proper orientation,
>> and another that is used to batch process (per
>> interactively generated configuration files) RAW images
>> to TIFF's. Both of these scrips work on 1 image per CPU
>> core. The 12 core machine finishes about 10 times as
>> fast as a script doing each image sequentially. A 2
>> core machine where each core was twice as fast as these
>> cores wouldn't be even half as fast at those particular
>> tasks.
>
> But there aren't any notebook PCs with two CPUs, are there?

I don't know of any.

Let's be clear. There are dual processor machines (such as the high end
Mac Pro).

But processors divide to cores (full capability CPU's) and cores divide
(in some cases like the i7) to hyperthreaded models.

Example: the Mac Pro with dual 6 core Xeon processors:

2 processors X 6 cores/processor X 2 threads/core = 24 threads running
concurrently.

However, only 1 thread per core is actually "ticking" at a time. When 1
thread is waiting for instruction or data fetch, the registers set for
the opposite thread is switched in while the prior register set goes
dormant.

It's worth it though - it results in approximately 30% greater
throughput per core.

>
> Anyway, how much faster would an i7 compared to the Core 2 Duo P8600 be?

The web is your friend. I looked at benchmarks for my 2.8 GHz Dual core
machine v. the 3.4 GHz i7 Quad Core. IIRC, it was about 3.8X faster on
benchmarks. (I left you a link in another post)

Also of importance was the memory speed difference. My current machine
is 667 MHz. The i7 based iMac is 1600 and multi channel for an effective
transfer rate of 12 GB/s. That helps keep all those cores busy.

Then "how much memory". More is usually better as you avoid page
swapping. My iMac is limited to 6 GB. My next one will ship with 8,
and I'll immediately add 16 for 24 GB total. Eventually up it to 32 GB.
That will reduce paging considerably. Possibly altogether.

[Floyd L. Davidson]

nospam <nos...@nospam.invalid> wrote:
>In article <87ehdcy...@pix.apaflo.com>, Floyd L. Davidson
><fl...@apaflo.com> wrote:
>
>> >> >Well, in my current machine during certain tasks the CPU load indicator
>> >> >in the task manager goes to 100% for both cores and stays at 100% until
>> >> >the task is completed. So I would assume that the processor is indeed
>> >> >the bottleneck.
>> >>
>> >> With image processing that is fairly common, and indeed
>> >> is probably the most annoying and longest waiting time
>> >> when there is a noticeable bottleneck.
>> >
>> >image processing is usually i/o bound, not cpu bound.
>> >
>> >that's why photoshop users want the fastest disks they can get, as much
>> >memory as they can get, etc.
>>
>> If Photoshop users see it that way, they are wrong. If
>> you spend any time at all editing an image, the editing
>> time vastly out distances the time it takes to read and
>> later write the image to disk.
>
>photoshop users see a dramatic increase in performance when they have
>more memory and faster hard drives. the cpu is *not* the bottleneck.
>
>adobe has repeatedly said that to optimize photoshop to be as fast as
>possible, give it fast drives and memory. they have a much better idea
>than you do about its bottlenecks.

They understand, yes. YOU don't. Skip the fast hard
drives, and add memory. If you have enough memory there
will never be a wait, during editing, for a disk write,
due to caching.

>cpus have greatly increased in speed and the number of cores, but i/o
>hasn't kept pace.
>
>most image processing does simple calculations (e.g., adjust colour
>balance) on a *lot* of pixels and moving them in and out of the cpu is
>the bottleneck.

The CPU is the bottleneck. If the same simple
calculation is done on 10 million pixels in sequence it
takes ten times as long as it would to do that using 10
cores to process 10 pixels at a time.

>on the other hand, if you spend your days doing cpu intensive filters
>such as radial blur, then it might be different but few people do that
>more than occasionally, if they do at all.
>
>> >video processing, on the other hand, is often cpu bound.
>>
>> And so is image editing in the way that I described above:
>> the annoying waits that are noticed and frustrating.
>
>nope. video editing involves more complex calculations, unless you're
>doing something simple such as adjusting colour balance on all frames.

Yep, but image editing of even simple things like
adjusting color balance on all pixels represents the
same kind of a load.

[nospam]

In article <8761ynz...@pix.apaflo.com>, Floyd L. Davidson

<fl...@apaflo.com> wrote:

> >> >> >Well, in my current machine during certain tasks the CPU load indicator
> >> >> >in the task manager goes to 100% for both cores and stays at 100% until
> >> >> >the task is completed. So I would assume that the processor is indeed
> >> >> >the bottleneck.
> >> >>
> >> >> With image processing that is fairly common, and indeed
> >> >> is probably the most annoying and longest waiting time
> >> >> when there is a noticeable bottleneck.
> >> >
> >> >image processing is usually i/o bound, not cpu bound.
> >> >
> >> >that's why photoshop users want the fastest disks they can get, as much
> >> >memory as they can get, etc.
> >>
> >> If Photoshop users see it that way, they are wrong. If
> >> you spend any time at all editing an image, the editing
> >> time vastly out distances the time it takes to read and
> >> later write the image to disk.
> >
> >photoshop users see a dramatic increase in performance when they have
> >more memory and faster hard drives. the cpu is *not* the bottleneck.
> >
> >adobe has repeatedly said that to optimize photoshop to be as fast as
> >possible, give it fast drives and memory. they have a much better idea
> >than you do about its bottlenecks.
>
> They understand, yes. YOU don't.

yes i do, and it's in agreement with adobe.

> Skip the fast hard
> drives, and add memory. If you have enough memory there
> will never be a wait, during editing, for a disk write,
> due to caching.

if you have enough memory, then you've removed the i/o bottleneck.
thanks for confirming it.

> >cpus have greatly increased in speed and the number of cores, but i/o
> >hasn't kept pace.
> >
> >most image processing does simple calculations (e.g., adjust colour
> >balance) on a *lot* of pixels and moving them in and out of the cpu is
> >the bottleneck.
>
> The CPU is the bottleneck.

photoshop's code is very highly optimized and is not the bottleneck,
except with certain filters such as radial blur that require a lot of
computation. the common stuff, like adjusting colour, exposure,
sharpening, etc. are all i/o bound.

long ago, the cpu was the bottleneck, but cpu performance has
dramatically increased, including offloading to the gpu, while memory
bandwidth has not kept pace. now the bottleneck is i/o.

i assume you're using the gimp, which is quite a bit slower than
photoshop doing the same operations, so i can see why you think image
processing is cpu bound.

> If the same simple
> calculation is done on 10 million pixels in sequence it
> takes ten times as long as it would to do that using 10
> cores to process 10 pixels at a time.

no it doesn't. there's always an overhead no matter how many cores you
have, and you will almost never realize a 10x improvement with 10
cores.

you also need 10x the memory bandwidth to keep all 10 cores fed with
data.

[Me]

That /was/ a big issue with the Gimp. It didn't utilise multi-core cpus
very well.
The newer versions do much better. The user may need to go into
Preferences|Environment|number of processors to use, and manually select
number of cores (including virtual cores).

>
>> If the same simple
>> calculation is done on 10 million pixels in sequence it
>> takes ten times as long as it would to do that using 10
>> cores to process 10 pixels at a time.
>
> no it doesn't. there's always an overhead no matter how many cores you
> have, and you will almost never realize a 10x improvement with 10
> cores.
>
> you also need 10x the memory bandwidth to keep all 10 cores fed with
> data.
>

This is all very nice, but PCs are terribly inefficient for image
processing compared to the dedicated processors in cameras, which manage
to perform complex operations (demosaicing raw data, complex
interpolation for automatic lens distortion and CA correction, colour
correction, encoding to jpeg etc) in the blink of an eye, and using very
little power.

[Bo...@somewhere.com]

On Sun, 12 May 2013 20:35:38 +0200, Alfred Molon <alfred...@yahoo.com> wrote:

>In article <YaPjt.8753$KI6....@newsfe09.iad>, Martin Brown says...
>> The various benchmarks that Windoze does are mostly irrelevant to image
>> processing. A more realistic hard CPU throughput benchmark is online at:
>>
>> http://www.cpubenchmark.net/high_end_cpus.html
>
>And here are the notebook processor benchmarks:
>http://www.cpubenchmark.net/laptop.html
>
>The P8600 shows up with a score of 1505, while a typical Core i7 for a
>notebook is around 4500, i.e. three times faster.

I wouldn't pay much attention to those scores...

According to their charts, of 2 computers I own, one CPU scored at 8324, and the
other at 1983... a difference of over 4 times...

But in the real world, a MIPS test shows 1164 to 870, and a MFLOPS test shows
1210 to 540.

A math problem times out to 1.4 sec to complete on one machine, and 1.7 sec on
the other. (no big deal there!)

I don't see a lot of difference between them while rendering HD video, one is a
bit faster rendering at about 2x real time. (A 10min video renders in 5 min's,
including a size change from 1080 down to 720..)

One machine is an i7-2600 running Win7 in 64bit, the other is a dual core E6850
running win XP 32bit.

I need to install a game on the i7 to see what it can do, the other machine was
a "top of the line game machine" at one time... a few years back! I usually run
games at full HD with all graphics options at max.

There are so many variables!

[Martin Brown]

On 12/05/2013 19:35, Alfred Molon wrote:
> In article <YaPjt.8753$KI6....@newsfe09.iad>, Martin Brown says...
>> The various benchmarks that Windoze does are mostly irrelevant to image
>> processing. A more realistic hard CPU throughput benchmark is online at:
>>
>> http://www.cpubenchmark.net/high_end_cpus.html
>
> And here are the notebook processor benchmarks:
> http://www.cpubenchmark.net/laptop.html
>
> The P8600 shows up with a score of 1505, while a typical Core i7 for a
> notebook is around 4500, i.e. three times faster.

I generally work on the basis that when the CPU is about 3x faster than
what I have it is time to upgrade. It isn't a bad heuristic. This used
to be roughly every other year for a long time but is now about 5 years.

Also unless you are doing very complex demanding work almost anything
with a benchmark score of 3000 or more will do for video editing - you
just have to wait a bit longer while it transcodes it. Image editing is
even less demanding, but you may want at least 8GB ram for large ones.

The spec I had my new machine built to was a gaming machine with the
high end graphics card deleted. The i7 actually has a decent built in 2D
graphics engine provided that you don't want 3D gaming graphics. It runs
at a shade over 120W flat out. Its predecessor was over 300W.

You should check that your software really can use all 8 hyperthreads
before going for an i7 though. Much software only really knows how to
exploit upto 4 cores and would be almost as fast on a cheaper i5.

Non-multicore code will get pathetic benchmark scores on these CPUs.

--
Regards,
Martin Brown

[Wolfgang Weisselberg]

Floyd L. Davidson <fl...@apaflo.com> wrote:

> The solution is not to look for the fastest CPU as such,
> but rather look for the chip with the most cores. Even
> if each core is half as fast as those on some other
> chip, the times when the speed of any one or two cores
> is the problem are rare. And when a process can be
> spread out over multiple cores... there ain't nothing
> like a pair of 6 core chips!

Sure there is: a rig with 24 cores. Or 100 cores. Or 1000
cores. (See Beowulf cluster).

-Wolfgang

[Wolfgang Weisselberg]

Floyd L. Davidson <fl...@apaflo.com> wrote:

> nospam <nos...@nospam.invalid> wrote:

>>cpus have greatly increased in speed and the number of cores, but i/o
>>hasn't kept pace.

>>most image processing does simple calculations (e.g., adjust colour
>>balance) on a *lot* of pixels and moving them in and out of the cpu is
>>the bottleneck.

> The CPU is the bottleneck.

When the IO to the CPU is the bottleneck, then, yes, the CPU

is the bottleneck.

> If the same simple
> calculation is done on 10 million pixels in sequence it
> takes ten times as long as it would to do that using 10
> cores to process 10 pixels at a time.

How about a few 100 or even a couple 1000 cores? Called
graphic card ...

-Wolfgang

[Wolfgang Weisselberg]

Alfred Molon <alfred...@yahoo.com> wrote:

> Anyway, how much faster would an i7 compared to the Core 2 Duo P8600 be?

That depends on the program and data used, as the CPUs don't
simply increase the speed (more MHz, blanket speed increase),
but also efficiency: less cycles needed for a given operation
--- this varies a lot by the operation looked at, the quality
of the branch predictor, and thus by the program and data used
to compare. (This can be even more complicated if you compare
Intel and AMD: in some disciplines one is quite faster, in
others the other one is. Here are some various benchmarks
between AMD's octocore and some i5 and an i7:
http://www.phoronix.com/scan.php?page=article&item=amd_fx8350_visherabdver2&num=1
)

Oh, and there's compiling for the extra CPU capabilities
instead of using only basic instruction sets, in specific
circumstances that can give quite a speed boast --- but
unless you're compiling the stuff yourself you'll probably
won't get these optimisation. (And then there's how well
opimizing the compiler is for your specific CPU, now and in
future.)
See here for an example how 2 good compilers aren't equal:
http://www.phoronix.com/scan.php?page=article&item=llvm_clang33_fx8350&num=1

Then there's the fact that your i7 won't be faster while you're
typing text into notepad (it's waiting for you all the time),
but may be vastly faster crunching numbers --- and the same i7
may be much slower on image manipulation than an identical i7
with a different graphics card (which the image manipulation
program uses to speed up some operations).


But if you want a single, unreliable number:
http://www.cpubenchmark.net/pt7_cpu_list.php
Intel Core2 Duo P8600 @ 2.40GHz 1633
Intel Core i7 620UM @ 1.07GHz 1577
[more i7's in raising performance] :
Intel Core i7-3970X @ 3.50GHz 12923

for what it's worth. Single CPU, multi-core, AIUI.

So an i7 dual core notebook (low power) CPU at 1 GHz may in
fact be a tad slower or a high end i7 may be almost 8 times
as fast --- for what it's worth.

Remember you'lll (probably, I'm not sure) need a new mainboard
(probably a different socket) and a cooler and you'll probably
like more RAM (not enough RAM costs lots of speed loss, what
is "enough" depends on what you do) and likely you'll want a
SSD (and you want more RAM for that to reduce writing there)
and you might need a new PSU ... so you could just as well
switch to AMD ---
http://www.cpubenchmark.net/cpu.php?cpu=AMD+FX-8350+Eight-Core&id=1780
give way more bang for the buck than Intel: 50+ PassMarks/USD
--- the Intel Core i7-3970X @ 3.50GHz offers only 12.61 ---
but they'll be only 5,5 times at fast, and being an octacore
instead of a hexcore the individial core is only 1.4 times
faster than your P8600's core (with the i7-3970X each core is
2.6 times as fast).

And single-thread the i7-3970X delivers 2,047 PassMarks, the
AMD 1,522, and the P8600 936 (many CPUs can go into a boost
mode with a single thread: 2*936 > 1633): the i7 is 2,1 times as
fast, the AMD 1,6 times (I doubt these numbers, though, they
don't fit --- the AMD only has a tiny bit of single thread
boost!):

That means the AMD is quite a bit slower and vastly cheaper
than a top of the line i7; both will be a bit faster
with single core use, the i7 notably so, the AMD possibly
only noticeably. Both will leave the P8600 in the dust *if*
you can keep all cores working at the same time; again, the
i7-3970X will probably outperform the AMD by quite a margin.
The AMD is around USD 150, the i7-3970X around USD 1,000.

AfterShot Pro uses all cores, for the record.

Do note that there are a lot of i7's that are vastly cheaper
(and somewhat slower) than the i7-3970X ...

Does that answer your question?

-Wolfgang

[Eric Stevens]

On Wed, 15 May 2013 23:26:12 +0200, Wolfgang Weisselberg
<ozcv...@sneakemail.com> wrote:

>Floyd L. Davidson <fl...@apaflo.com> wrote:
>> nospam <nos...@nospam.invalid> wrote:
>
>>>cpus have greatly increased in speed and the number of cores, but i/o
>>>hasn't kept pace.
>
>>>most image processing does simple calculations (e.g., adjust colour
>>>balance) on a *lot* of pixels and moving them in and out of the cpu is
>>>the bottleneck.
>
>> The CPU is the bottleneck.
>
>When the IO to the CPU is the bottleneck, then, yes, the CPU
>is the bottleneck.

I love your logic.

>
>> If the same simple
>> calculation is done on 10 million pixels in sequence it
>> takes ten times as long as it would to do that using 10
>> cores to process 10 pixels at a time.
>
>How about a few 100 or even a couple 1000 cores? Called
>graphic card ...
>
>-Wolfgang

--

Regards,

Eric Stevens