2.66 Quad vs 2.26 Octo, usage behavior and which best?

[carlosgavina]

I know, there are a lot of reviews etc. for this, but just wanted to add my two cents on the case, and I would like to know your opinions about this

Although the 2.66Ghz Quad (2009) performs better on the majority photo editing software than the 2.26Ghz octo, in the end doesn't it depends on the workflow of each user?

I'm leaned to buy the 2.26Ghz, 16Gb Ram, 3Tb Disks and the Radeon Graphic Card, over the 2.66Ghz Quad beast, although most I do is Photoshop and Aperture.. BUT, i'm a web and graphic designer myself and a wannabe film director (cof cof) and sometimes 3D almost-artist, and i have photography as a hobby, and I always have like a million programs opened at the same time, like: Photoshop doing the Design, at the same time Aperture for the last import from the camera, Coda for the programing, Safari with lots of tabs, Adium, Illustrator for some vect.

And when doing video: Premiere (although i'm watching to learn final cut, and the price makes it a great deal), After Effects and Logic 8 for some audio at the same time.

I think more cores in these cases are better than the clock speed? Almost all the benchmarks and tests that are made, the machine its tested only with the program they are testing opened and maybe safari, finder or iTunes at most.

But for intensive multi-software worlflow isn't better more cores?

What do you think?

thanks in advance
Carlos Gavina

 

[nightfly13]

I'm inclined to agree with you. More cores > faster clock speed for most jobs. There are some intensive apps that are only written to use one core, and there you'll suffer a bit from clock speed. Snow Leopard is supposed enable most Apple OS-level apps to be 'multi-core aware' and definitely that's the way the market is working.

I'm about to swap processors from a quad 2.66 (2006 Mac Pro) to Octo 2.33 for precisely the same reasoning.

 

[nightfly13]

Oh almost forgot - the memory-storage setup looks good - but really seriously consider a SSD to boost performance most dramatically. Lots of guys are doing RAID 0 SSDs together both because it's faster and also because large-capacity SSDs are really expensive right now.

I recommend 2x Intel x25-m (80 or 160gb). Try for the 2nd Gen version, but I'm using a first gen and LOOOVE it. My disk test score increased 8.5x (with XBench) from my stock 250gb 7200 RPM to a single 1st Gen Intel SSD.

 

[Tesselator]

I know, there are a lot of reviews etc. for this, but just wanted to add my two cents on the case, and I would like to know your opinions about this

Although the 2.66Ghz Quad (2009) performs better on the majority photo editing software than the 2.26Ghz octo, in the end doesn't it depends on the workflow of each user?

Maybe. In a VERY highly automated (scripts and batches) workflow the 2.26 might be faster. Automated is a legitimate workflow.

I'm leaned to buy the 2.26Ghz, 16Gb Ram, 3Tb Disks and the Radeon Graphic Card, over the 2.66Ghz Quad beast, although most I do is Photoshop and Aperture.. BUT, i'm a web and graphic designer myself and a wannabe film director (cof cof) and sometimes 3D almost-artist, and i have photography as a hobby, and I always have like a million programs opened at the same time, like: Photoshop doing the Design, at the same time Aperture for the last import from the camera, Coda for the programing, Safari with lots of tabs, Adium, Illustrator for some vect.

And when doing video: Premiere (although i'm watching to learn final cut, and the price makes it a great deal), After Effects and Logic 8 for some audio at the same time.

I think more cores in these cases are better than the clock speed?

Probably not unless all of those programs are working at the same time. If they're all just loaded into memory about the only thing that's affected is memory. In this case an octad might be better than a quad because there are more RAM slots.

4 slots in the Quad and 8 slots in the octad.

Almost all the benchmarks and tests that are made, the machine its tested only with the program they are testing opened and maybe safari, finder or iTunes at most.

Actually even though there are only one or two applications loaded there are 100 or 200 processes loaded and running in the background just like an idle app would be.

But for intensive multi-software worlflow isn't better more cores?

If all the programs are working then yes. Most people don't multi-task too well tho - so even though we have 4 to 8 applications loaded at once we usually only use one at a time. In such a case more cores or even hyper-threading is a waste. Where more cores really shines for single creative users such as yourself is in 3D renderings. For setting up the scene, modeling, and so forth you want clock speed but when you set about rendering out a multi-frame animation the number of cores you have will scale almost linearly to the time required to complete the job.

If it were me I would skip the 2009 models all together unless I was fabulously wealthy, and get myself a 2008 2.8, 3.0, or 3.2 eight core for less than the price you're talking for any of the 2009 models. You get the 8-cores, the 8 RAM slots, cheaper RAM, more SATA connections, an IDE interface, the faster clocks, and all for a lower price!

 

[DoctorX]

Similar situation, hard to decide on model

...most I do is Photoshop and Aperture.. BUT, i'm a web and graphic designer myself and a wannabe film director (cof cof) and sometimes 3D almost-artist, and i have photography as a hobby, and I always have like a million programs opened at the same time, like: Photoshop doing the Design, at the same time Aperture for the last import from the camera, Coda for the programing, Safari with lots of tabs, Adium, Illustrator for some vect.

Right now I'm in a very similar situation, having had a hard time justifying getting 8-core for current usage, but feels that with the projected 5 year lifetime of the machine I might miss out a lot of potential speedbumps down the line.

If it were me I would skip the 2009 models all together unless I was fabulously wealthy, and get myself a 2008 2.8, 3.0, or 3.2 eight core for less than the price you're talking for any of the 2009 models. You get the 8-cores, the 8 RAM slots, cheaper RAM, more SATA connections, an IDE interface, the faster clocks, and all for a lower price!

Faster clocks doesn't seem to count for much when it's a generation behind, IDE matters only when you have legacy devices you need to support somehow (but with SATA drives so cheap, why bother?) but I do agree that value wise 08 model is much better... for now.

Right now I can safely say I do NOT need octo-core. But I can also safely say that there's no way I can justify spending for another 3k machine in at least 2 years.

08s are off my list for being the last of their breed (architecture), while 09 quad seems kinda awkward for being newer yet having less cores than the 08 octos. Entry 09 octos seem to be best positioned to leverage future speedbumps while minimize paying for ridiculous CPU markup.

So I guess the main question is will the potential of the octo pay for itself within 2-5 year timeframe? I'm leaning toward yes, but someone could prove me wrong (and save me a coupla hundreds).

 

[VirtualRain]

Right now I'm in a very similar situation, having had a hard time justifying getting 8-core for current usage, but feels that with the projected 5 year lifetime of the machine I might miss out a lot of potential speedbumps down the line.



Faster clocks doesn't seem to count for much when it's a generation behind, IDE matters only when you have legacy devices you need to support somehow (but with SATA drives so cheap, why bother?) but I do agree that value wise 08 model is much better... for now.

Right now I can safely say I do NOT need octo-core. But I can also safely say that there's no way I can justify spending for another 3k machine in at least 2 years.

08s are off my list for being the last of their breed (architecture), while 09 quad seems kinda awkward for being newer yet having less cores than the 08 octos. Entry 09 octos seem to be best positioned to leverage future speedbumps while minimize paying for ridiculous CPU markup.

So I guess the main question is will the potential of the octo pay for itself within 2-5 year timeframe? I'm leaning toward yes, but someone could prove me wrong (and save me a coupla hundreds).

Here's what has become my boilerplate response to this debate...

I always say that the Octo's can really only be justified if you make money with the machine (and can realize a positive ROI on the added investment) or someone else is buying it for you (company, grant, etc.). Otherwise, the diminishing returns are just not worth it.

Some will argue that as software becomes more mutli-threaded, you will be better able to exploit the added cores in an Octo - making it more future proof. The thing is that by the time that happens, middle of the road workstations of the day will have 6-8 cores per CPU so why waste money on that now when you can't really leverage it?

Even with an improved multi-threaded framework available in Snow Leopard, it will take time for applications to be coded to take advantage of it and I strongly feel that developers will get more of a boost from leveraging the GPU (OpenCL) instead and will therefore head in this direction. When you look at the tasks that currently benefit from multi-threading such as encoding, rendering, or image processing, you can immediately see that they will benefit tremendously from the hundreds of cores a GPU has rather than trying to leverage the 4 or 8 cores available in the CPU(s).

I also don't think trading clock speed for cores is a sound decision right now... nor will it ever be. Clock speed is still king on CPU's... it impacts every task you do. Really, beyond 2 cores, added cores are of value in very limited circumstances (encoding, etc. as I mentioned above). So if you do these tasks, and time is money, absolutely buy as many cores as you can.

While I know I'm probably in the minority here, I forsee the ultimate computer hardware build out in the near future to be a high-clock CPU, dual GPU's with 1-2GB VRAM (or as much as possible), and SSD's in RAID.

 

[nanofrog]

If it were me I would skip the 2009 models all together unless I was fabulously wealthy, and get myself a 2008 2.8, 3.0, or 3.2 eight core for less than the price you're talking for any of the 2009 models. You get the 8-cores, the 8 RAM slots, cheaper RAM, more SATA connections, an IDE interface, the faster clocks, and all for a lower price!

Not to mention they're better for internal hardware RAID, if needed for the intended usage. Given the added costs associated with 3rd party controllers and the needed adapter kit at an extra $165USD, it matters to some.

And as many, if not the majority of applications are still single threaded, clock speeds are still desireable if at all possible. If the applications are limited to those that are multithreaded, then a slightly lower clocked pair of processors may make more sense.

I always say that the Octo's can really only be justified if you make money with the machine (and can realize a positive ROI on the added investment) or someone else is buying it for you (company, grant, etc.). Otherwise, the diminishing returns are just not worth it.

I agree. Given the cost associated with these machines, if it can't be utilized now, it's not worth it. It will just end up a waste of funds.

Some will argue that as software becomes more mutli-threaded, you will be better able to exploit the added cores in an Octo - making it more future proof. The thing is that by the time that happens, middle of the road workstations of the day will have 6-8 cores per CPU so why waste money on that now when you can't really leverage it?

To me, it seems applications that would best be served by going multithreaded, already have. A few straglers may come along, but not many IMO.

Even with an improved multi-threaded framework available in Snow Leopard, it will take time for applications to be coded to take advantage of it and I strongly feel that developers will get more of a boost from leveraging the GPU (OpenCL) instead and will therefore head in this direction. When you look at the tasks that currently benefit from multi-threading such as encoding, rendering, or image processing, you can immediately see that they will benefit tremendously from the hundreds of cores a GPU has rather than trying to leverage the 4 or 8 cores available in the CPU(s).

It would make sense for FPU bound tasks, as that's what GPU's really are.

I also don't think trading clock speed for cores is a sound decision right now... nor will it ever be. Clock speed is still king on CPU's... it impacts every task you do. Really, beyond 2 cores, added cores are of value in very limited circumstances (encoding, etc. as I mentioned above). So if you do these tasks, and time is money, absolutely buy as many cores as you can.

I agree that clock speed will remain important, but core count, I'd up to 4 for some users. At any rate, 2 - 4 cores would suffice for most applications.

Anything greater, would certianly be for a specific usage, that can be implemented from the day the system is plugged in. Not waiting around for the software to play catch up.

While I know I'm probably in the minority here, I forsee the ultimate computer hardware build out in the near future to be a high-clock CPU, dual GPU's with 1-2GB VRAM (or as much as possible), and SSD's in RAID.

Maybe. Definitely nice setup.

Unfortunately, things don't always turn out the way we hope, or make any sense.

 

[Tesselator]

Yep! There's almost zero advantage to the "new generation" Dr. X is referring to here. The actual difference between the 5500 series and the 5400 series is 8% at most and typically much less. I guess 8% increase is not worth over 200% the price?

IDE is perfect for ODD devices and it's even difficult to find ODDs in SATA form. Why waste an SATA connection on an ODD when you don't have to and when there's no advantage in doing so?

 

[nanofrog]

Yep! There's almost zero advantage to the "new generation" Dr. X is referring to here. The actual difference between the 5500 series and the 5400 series is 8% at most and typically much less. I guess 8% increase is not worth over 200% the price?

IDE is perfect for ODD devices and it's even difficult to find ODDs in SATA form. Why waste an SATA connection on an ODD when you don't have to and when there's no advantage in doing so?

Fewer problems it seems with drivers and booting as well. IDE's not dead yet.

 

[tome viewer]

Here's what has become my boilerplate response to this debate...

I always say that the Octo's can really only be justified if you make money with the machine (and can realize a positive ROI on the added investment) or someone else is buying it for you (company, grant, etc.). Otherwise, the diminishing returns are just not worth it.

Some will argue that as software becomes more mutli-threaded, you will be better able to exploit the added cores in an Octo - making it more future proof. The thing is that by the time that happens, middle of the road workstations of the day will have 6-8 cores per CPU so why waste money on that now when you can't really leverage it?

Even with an improved multi-threaded framework available in Snow Leopard, it will take time for applications to be coded to take advantage of it and I strongly feel that developers will get more of a boost from leveraging the GPU (OpenCL) instead and will therefore head in this direction. When you look at the tasks that currently benefit from multi-threading such as encoding, rendering, or image processing, you can immediately see that they will benefit tremendously from the hundreds of cores a GPU has rather than trying to leverage the 4 or 8 cores available in the CPU(s).

I also don't think trading clock speed for cores is a sound decision right now... nor will it ever be. Clock speed is still king on CPU's... it impacts every task you do. Really, beyond 2 cores, added cores are of value in very limited circumstances (encoding, etc. as I mentioned above). So if you do these tasks, and time is money, absolutely buy as many cores as you can.

While I know I'm probably in the minority here, I forsee the ultimate computer hardware build out in the near future to be a high-clock CPU, dual GPU's with 1-2GB VRAM (or as much as possible), and SSD's in RAID.

This is not only the best explanation I have heard but you also offer a very good future prediction IMO. So it seems everyone is in agreement...sort of. And all else being equal, clocks are better than cores at this point in time unless someone has a specific need for the extra cores beyond 4. (We also can't forget hyperthreading if the need for more than four cores arises.) The only thing that seems to be missing from this clocks versus cores debate, are the changes in the Nehalem's CPU architecture. i.e. QPI versus FSB, Turbo versus a fixed clock, etc. And I think this "little" difference is what makes the 2009 offerings better than the 2008. So if we are temporarily focusing on clocks instead of cores at this point in time the changes in latency and the Nehalem's ability to adjust its clock upward is what allows the 2.66 Quad to keep pace with the 3.2 Octo in single thread, and the 2.93 Quad to out perform the 3.2 Octo in single thread.

 

[Tesselator]

is what allows the 2.66 Quad to keep pace with the 3.2 Octo in single thread, and the 2.93 Quad to out perform the 3.2 Octo in single thread.

You have misunderstood the CineBench results. Bigger is better! Not the other way around.



EDIT: Whoops. And I misread your sentence missing the part where you say "in single thread.". Sorry about that.

Yeah, I think single thread is a tiny bit faster mostly because of the RAM speed and the embedded Memory controller.

 

[nanofrog]

Yeah, I think single thread is a tiny bit faster mostly because of the RAM speed and the embedded Memory controller.

Yep. And Turbo Boost may have kicked in a little too.

The architectural differences are there, but given the current pricing, not worth it compared to the '08's (covered rather well in another thread).

 

[jjahshik32]

Single threads will be SO yesterday when SL arrives and continues to develop.

I'd like to think about it as 8 cores as one vs. 4 cores as one. Well the general idea of more cores to begin with was to be created to take advantage of multithreads afterall since its too hot to clock above 3.2GHz.

Go octo.

 

[nanofrog]

Single threads will be SO yesterday when SL arrives and continues to develop.

I'd like to think about it as 8 cores as one vs. 4 cores as one. Well the general idea of more cores to begin with was to be created to take advantage of multithreads afterall since its too hot to clock above 3.2GHz.

Go octo.

Not everything can be multithreaded though, as the processes rely on the previous data generated.

Most of the applications that would benefit already have been developed. SL's API's will help the fence-sitters that have waited, presumably as they didn't want to develop it from scratch.

 

[Tesselator]

Yep. And Turbo Boost may have kicked in a little too.

The architectural differences are there, but given the current pricing, not worth it compared to the '08's (covered rather well in another thread).

Oh for sure. To get those results turbo-boost was maxed! This kinda thing will show up in benchmark environments far more often than in real-world use. Probably that's why Windoze users have already figured out how to turn on turbo-boost on a full-time basis instead of letting the system decide when to.

And yeah, it's been discussed to death in several other threads with benchmarks galore.

Not everything can be multithreaded though, as the processes rely on the previous data generated.

Most of the applications that would benefit already have been developed. SL's API's will help the fence-sitters that have waited, presumably as they didn't want to develop it from scratch.

Yup! SL won't make any difference at all initially! It only offers a new framework and that's it. Essentially a new set of tools for developers to use should they so decide to use them. The same tools or better have already been designed by some application developers - so in those cases it likely will make no difference to them either.

The fact the SL will more robustly support 64-bit applications will probably make a bigger difference than the Grand Central stuff. It will for sure initially!


.

 

[jjahshik32]

Not everything can be multithreaded though, as the processes rely on the previous data generated.

Most of the applications that would benefit already have been developed. SL's API's will help the fence-sitters that have waited, presumably as they didn't want to develop it from scratch.

Its all good. I dont need a 3.2GHz to run safari or ichat.

 

[Tesselator]

I want 3.2 for safari! I don't need it - but there's a lot of web applets and stuff that can really benefit from faster clocks! Heck even my favorite game (Quake III Arena) is now playable in a browser. It's free and called Quake Live.

 

[nanofrog]

Oh for sure. To get those results turbo-boost was maxed! This kinda thing will show up in benchmark environments far more often than in real-world use. Probably that's why Windoze users have already figured out how to turn on turbo-boost on a full-time basis instead of letting the system decide when to.

Over Clocking... YUM. Me likes speed. And it happens to allow more than Turbo Boost can, and definitely on a full time basis. 4.2Ghz on air (&stable) has been achieved. I've done it, but I didn't like the heat, and backed it down a hair (4.12GHz).

Yup! SL won't make any difference at all initially! It only offers a new framework and that's it. Essentially a new set of tools for developers to use should they so decide to use them. The same tools or better have already been designed by some application developers - so in those cases it likely will make no difference to them either.

The fact the SL will more robustly support 64-bit applications will probably make a bigger difference than the Grand Central stuff. It will for sure initially!


.

Yeah, I see GC as more of a marketing gimick ATM, but with potential down the road.

The 64bit Kernel and the optimizations (or really the removal of the 32bit mess) is far more interesting. Complete 64bit-ness.

Its all good. I dont need a 3.2GHz to run safari or ichat.

An Octad '09MP for web browsing and iChat. You should be ashamed of yourself!

But faster is always good, even if it's just Calc.

 

[DoctorX]

I'm convinced, quad it is.

Seriously in the beginning I just don't want to deal with windows anymore after my 4 year old desktop finally croaked. Wanted expandability to handle my ever growing photoshop / raw files so Mac Pro became my only choice.

Then I get suckered into the thinking "if I'm spending this much already why not make it the best value?" and got into an upsell mentality.

When you get down to it, it just bugged me to no end when there are certain tasks that can be sped up 2x by forking out extra 1/4.

Oh well, power untapped is power wasted, right? Expandability wise intel already have 8-core single Nehalem, so being limited to single socket doesn't seem to be all that bad, and I heard the quad has the heat spreader intact so much easier upgrade post-Applecare?

Btw. There are very few pics of the quad core Nehalem, but seen in this pic here:

the quad seems to have a MUCH bigger/beefier heatsink. Does that mean it can potentially accept much higher clocked cpus down the line? (or over-clocked) If so then that levels the field even more, and could potentially end up being a nice balance between single / multi-thread tasks performance down the line.

Pure speculation at this point of course.

 

[nanofrog]

Oh well, power untapped is power wasted, right? Expandability wise Intel already have 8-core single Nehalem, so being limited to single socket doesn't seem to be all that bad, and I heard the quad has the heat spreader intact so much easier upgrade post-Applecare?

That's the way I see it, and even if the IHS is missing, it's not hard to get around, as you just use a thermally conductive material, such as SilPad (polydisiloxane with boron nitride sheet) to fill in the gap, and still remove the heat off the Voltage Regs.

the quad seems to have a MUCH bigger/beefier heatsink. Does that mean it can potentially accept much higher clocked cpus down the line? (or over-clocked) If so then that levels the field even more, and could potentially end up being a nice balance between single / multi-thread tasks performance down the line.

Pure speculation at this point of course.

Apparently, it can take CPU's with 150W TDP, and another member (Tutor), has already installed an i7-965 (130W TDP) with no problems.

As per Over Clocking, you can't change the settings in the firmware, but it may be possible if a 3rd party utility is developed (one was created by ZDNet of Germany for previous machines '06 - '08, but won't work on the '09's due to the architectural differences), or something along the lines of a BSEL mod is done. Not easy, but doable I think. Rather risky, but to each their own...

 

[VirtualRain]

Single threads will be SO yesterday when SL arrives and continues to develop.

I'd like to think about it as 8 cores as one vs. 4 cores as one. Well the general idea of more cores to begin with was to be created to take advantage of multithreads afterall since its too hot to clock above 3.2GHz.

Go octo.

We will see in a few days, but besides an overall snappiness as many have reported, there is little in the way of performance improvements out of the box... folks have been using SL for months and the gold build for a couple of weeks... if there was gold in them there hills, we would have heard about it by now.

I surmise that if Leopard did not already account for this that the SL task scheduler might now be HT aware, and therefore more intelligently assign tasks to physical cores vs. logical cores when it makes sense. That may make a difference on some apps.

The key new frameworks that developers can now take advantage of include:
1. 64-bit architecture. This will offer some benefits in FP intensive apps (if they are recompiled), probably in the order of 5-10% and support for more address space which may finally start to tax the memory architecture on the 08's .
2. Grand central. This will make it easier for developers to enable multi-threading in their apps. However, as someone else pointed out, most apps that lend themselves to multi-threading have already implemented it. It's unlikely to benefit a developer to re-architect their app to use Grand Central vs. their existing threading architecture. I expect this to have minimal impact.
3. OpenCL. This will provide orders of magnitude difference for apps that lend themselves to massive parallelism. Within a couple of release cycles, I would hope that most multi-threaded pro apps would take advantage of this capability, but it will take investment of valuable development cycles that are also earmarked for new feature development, so even this could take some time to materialize depending on a given apps market position and priorities.

 

[Tesselator]

Apparently, it can take CPU's with 150W TDP, and another member (Tutor), has already installed an i7-965 (130W TDP) with no problems.

I too like the 2008 models MUCH better but I agree here that if you're going to buy a 2009 quad then get the cheapy one and buy separately a nice fast processor!

 

[nanofrog]

I too like the 2008 models MUCH better but I agree here that if you're going to buy a 2009 quad then get the cheapy one and buy separately a nice fast processor!

I agree. My like for the '08's is more than just due to similar performance for less money though (certainly good enough on this fact alone), add in the ease for internal hardware RAID, and it screams "Buy Me".

 

[Tesselator]

Yup, and if you add a RAID card to the 2009 then that's it. All your PCIe slots are full and it's not expandable any further. In a 2008 machine there's still room for more. I don't get why Apple did that. It's like they cut the machine in half and then jacked the price to double??? What gives? Too weird!

 

[nanofrog]

Yup, and if you add a RAID card to the 2009 then that's it. All your PCIe slots are full and it's not expandable any further. In a 2008 machine there's still room for more. I don't get why Apple did that. It's like they cut the machine in half and then jacked the price to double??? What gives? Too weird!

Control + Greed

Despite sales figures in the current economy, there's an increase in margin that created the pricing structure IMO. The differences in parts isn't that different, as though some areas increased, some also decreased. R&D wouldn't be inordinate either, as most of it was developed by Intel.