New mac pro purchase advice ASAP

[jb502]

I just sold my 2008 Mac Pro. It was 8 core 2.8ghz. I need to replace it. What I replace it with MUST be faster at EVERYTHING. 12 core is not an option. Too expensive. I am considering:

2010

3.33 6 core or 2.4 8 core

or

2009

2.66 octo

I can afford any of these, which should I get.

Mostly I need the speed for exporting from Imovie and Adobe Premiere. Wichever one will do these exports of 1080p fastest.

 

[Hellhammer]

3.33GHz 6-core offers a balanced mix of clock speed and cores. IIRC iMovie is still single-threaded so the clock speed is important. 2.66GHz 8-core might be a bit faster when all of its cores can be used but it is slower in tasks where its cores cannot be used.

 

[Loa]

Hello,

Going to a 2009 or 2010 certainly has performance benefits, but nothing that will be impressive compared to a 2008 2.8 8-core.

Among the options you listed, only the 3.33 hexcore will be significantly faster than your 2008 machine. Even then... it's been a while since we've seen impressive performance gains by updating a 2-3 year old machine...

Loa

 

[jb502]

Or would a 2009 8 core be the safest bet to be faster at Everything since it is :

1.higher clockspee
2. 8 cores
3. nehalem > Harpertown

?

that's stretching my budget though at about 3600 refurbished

 

[derbothaus]

Even with full core saturation the 2.66 is the same speed as the 3.33. Roughly. On single threaded the 3.33 will dominate.

Geekbench (Rough) Mactracker numbers:
2x2.66: 14458
3.33: 14183

Both machines are faster than any 2008 as the 3.2GHz gets around 8565.

 

[gameface]

How much ram do you have in your machine now? That machine should be plenty fast at what you are doing and transcoding has a lot to do with RAM.

 

[jb502]

How much ram do you have in your machine now? That machine should be plenty fast at what you are doing and transcoding has a lot to do with RAM.

I had 6 gigs. but i sold it 4 days ago. SO i have to buy a new one. no other option. New one will get 12 gigs, no matter which one i buy.

 

[goMac]

..why'd you sell the 2008? Those machines are honestly hard to beat. You're going to be hard pressed to get a better machine in your budget.

I don't know if either of those would beat the 2008 at movie exporting... It would be pretty damn close. The 2009 2.66 Octo could possibly beat it.

 

[jb502]

..why'd you sell the 2008? Those machines are honestly hard to beat. You're going to be hard pressed to get a better machine in your budget.

I don't know if either of those would beat the 2008 at movie exporting... It would be pretty damn close. The 2009 2.66 Octo could possibly beat it.

I was under the impression that nehalem was quite a bit faster, even at a lower clock speed.

 

[goMac]

I was under the impression that nehalem was quite a bit faster, even at a lower clock speed.

It depends. For memory intensive operations, it can be, and I'm not sure how that would play out for encoding. Encoding does use a fair amount of memory.... But I'd be more worried about disk speeds.

There are still quite a few scenarios where the Nehalem memory architecture does not really come into play and an 08 can beat or stay competitive with an 09 or an '10...

Were you using SSD or RAID on the 2008? That would have given you a significant speed boost.

 

[jb502]

It depends. For memory intensive operations, it can be, and I'm not sure how that would play out for encoding. Encoding does use a fair amount of memory.... But I'd be more worried about disk speeds.

There are still quite a few scenarios where the Nehalem memory architecture does not really come into play and an 08 can beat or stay competitive with an 09 or an '10...

Were you using SSD or RAID on the 2008? That would have given you a significant speed boost.

I did not have an ssd, but i will be buying one for my new mac pro

 

[Hellhammer]

It depends. For memory intensive operations, it can be, and I'm not sure how that would play out for encoding. Encoding does use a fair amount of memory.... But I'd be more worried about disk speeds.

I guess it's time to stop guessing and bring some concrete numbers to the table.

http://www.barefeats.com/nehal04.html
http://www.barefeats.com/wst10c.html
http://www.anandtech.com/show/3969/apple-mac-pro-mid-2010-review/6

Nehalems are quite a bit faster than Harpertowns. Disk speeds are pretty irrelevant as your CPU cannot encode as fast as your disk can write. Otherwise you could encode a full-size Blu-Ray in couple of minutes.

There are still quite a few scenarios where the Nehalem memory architecture does not really come into play and an 08 can beat or stay competitive with an 09 or an '10...

How is that? Nehalems support 1066MHz DDR3 and triple-channeling while Harpertowns support only 800MHz DDR2 and dual-channeling.

 

[goMac]

I guess it's time to stop guessing and bring some concrete numbers to the table.

http://www.barefeats.com/nehal04.html
http://www.barefeats.com/wst10c.html
http://www.anandtech.com/show/3969/apple-mac-pro-mid-2010-review/6

None of which are particularly impressive benchmarks, and also, as noted, you at least need to have the same number of cores to see a speed gain.

Nehalems are quite a bit faster than Harpertowns. Disk speeds are pretty irrelevant as your CPU cannot encode as fast as your disk can write. Otherwise you could encode a full-size Blu-Ray in couple of minutes.

Errrr no, and write speeds aren't really even the problem. It's read speeds. That's why most editors at least have RAID. You don't edit in H.264 or necessarily compressed video which means the files you are dealing with are much much bigger than a Bluray (heh).

How is that? Nehalems support 1066MHz DDR3 and triple-channeling while Harpertowns support only 800MHz DDR2 and dual-channeling.

If you are running an operation that is CPU intensive and not memory intensive, the Nehalems have little speed advantage, and sometimes are even slower. You can usually see this in GPU heavy benchmarks, where most of the CPU processing being done is not RAM heavy. I've seen the 2008's beat some of the 2010s.

Don't get me wrong, the Nehalems have a lot of cool tech (I actually got to talk with some Intel engineers extensively about it), but it doesn't necessarily translate to an across the board speed boost, and the machines are much more expensive.

Regardless, if you get a Nehalem and put it on a slow hard disk, you're going to get a slowdown. Getting a faster memory bus is meaningless if you can't pull data off disk fast enough. I'd still suggest the 09 to the OP (as it's 8 core and at a good speed), but you need to at least get some RAID on there.

 

[Hellhammer]

I've seen the 2008's beat some of the 2010s.

I would like to have a source for that.

Errrr no, and write speeds aren't really even the problem. It's read speeds. That's why most editors at least have RAID. You don't edit in H.264 or necessarily compressed video which means the files you are dealing with are much much bigger than a Bluray (heh).

OP was talking about encoding where disk speeds have no relevance. In editing, the disk speeds matter though, you are right about that. A couple of 1-2TB HDs in software RAID 0 should be decent.

 

[goMac]

I would like to have a source for that.

http://www.barefeats.com/wst10g7.html

The Harpertown is extremely close (~1%) and even wins sometimes.

OP was talking about encoding where disk speeds have no relevance.

...so where exactly is the video he's encoding coming from then? Because if it's not a disk I'd like to know where.

In editing, the disk speeds matter though, you are right about that. A couple of 1-2TB HDs in software RAID 0 should be decent.

I'd argue that disk speeds matter for everything now, but that's a different debate.

It's the biggest issue with the new memory architecture though, it's great that in theory you've got all this bandwidth to memory, but you're still limited by the rate at which you can pull data off devices into memory. If you're running an operation that touches the hard disk, you start losing any advantage Nehalem brings.

 

[Hellhammer]

The Harpertown is extremely close (~1%) and even wins sometimes.

I guess that is more dependent on the GPU. You can clearly see that the 2006 Mac Pro is lagging behind due to its PCIe 1.1. Harpertowns are still great CPUs, I have to agree with that. 2008 Mac Pro was the best bang for the buck. If I was OP, I wouldn't have sold mine.

 

[jb502]

I would like to have a source for that.

http://www.barefeats.com/wst10g7.html

The Harpertown is extremely close (~1%) and even wins sometimes.



...so where exactly is the video he's encoding coming from then? Because if it's not a disk I'd like to know where.



I'd argue that disk speeds matter for everything now, but that's a different debate.

It's the biggest issue with the new memory architecture though, it's great that in theory you've got all this bandwidth to memory, but you're still limited by the rate at which you can pull data off devices into memory. If you're running an operation that touches the hard disk, you start losing any advantage Nehalem brings.

You posted a benchmark for video cards. The results are in fps. This is a pointless benchmark for this topic

 

[goMac]

You posted a benchmark for video cards. The results are in fps. This is a pointless benchmark for this topic

Because video processing has nothing to do with frames, I guess? And encoding certainly wouldn't be measured in FPS, now would it?

I guess that is more dependent on the GPU. You can clearly see that the 2006 Mac Pro is lagging behind due to its PCIe 1.1. Harpertowns are still great CPUs, I have to agree with that. 2008 Mac Pro was the best bang for the buck. If I was OP, I wouldn't have sold mine.

The reasoning why is because the benchmarks were CPU heavy but not memory heavy. But yeah, the 2006 Mac Pro lags because of it's PCI 1.1 bus.

 

[Transporteur]

The 2010 2.4GHz pretty much has the same performance as the 2008 2.8GHz.

There aren't many direct comparisons available, as the most Nehalem - Harpertown benches were made with the release of the 2009 model.
The 2009 2.26GHz model was a little slower in single threaded tasks, faster in multi-threaded tasks, though.
Given the slightly higher clock speed of the 2010 octad, it should beat the 2008 model, not by much, so no worth being called an "upgrade".

The 2009 2.66GHz model certainly is quite faster in all tasks, but again, not worth upgrading IMHO.

The only real upgrade would be the 2010 hex model, which runs circles around ANY of the mentioned machines in single threaded tasks and has roughly the same multithreaded performance as the 2.4GHz 2010 octad model.

 

[jb502]

Im not gaming. This is a video card comparison.

Quote from benchmark.

As you can also see from the graphs, certain models of Mac Pro make better use of the both the 5870 and 5770. There appears to be a correlation between CPU clock speed and relative performance. That is born out by the fact that the 3.33GHz 6-core Mac Pro leads the pack in 6 out of 7 tests. And since most GPU intensive apps use one or two cores, having an 8-core or 12-core Mac Pro will seldom provide an advantage for graphic intensive apps.

 

[goMac]

Im not gaming. This is a video card comparison.

It's actually quite comparable. You're rendering video (i.e. you're rendering a certain number of frames per second), you're likely rendering effects from the GPU.... The big difference between video and games is that your disks enter into play.

As you can also see from the graphs, certain models of Mac Pro make better use of the both the 5870 and 5770. There appears to be a correlation between CPU clock speed and relative performance. That is born out by the fact that the 3.33GHz 6-core Mac Pro leads the pack in 6 out of 7 tests. And since most GPU intensive apps use one or two cores, having an 8-core or 12-core Mac Pro will seldom provide an advantage for graphic intensive apps.

Video rendering/encoding is actually heavily multicore (except for, as one poster noted, iMovie).

You might want to go back and learn a bit about the tools before you start making hardware decisions.

 

[Evil Spoonman]

It's actually quite comparable. You're rendering video (i.e. you're rendering a certain number of frames per second), you're likely rendering effects from the GPU.... The big difference between video and games is that your disks enter into play.

Wait, what? A GPU benchmark is completely not applicable to encoding in any way. Unless you are going to use an encoder that leverages GPGPU (CUDA/OpenCL). Almost all programs are still using the CPU to encode, and thus the GPU has no benefit whatsoever.

Video rendering/encoding is actually heavily multicore (except for, as one poster noted, iMovie).

Ehhh… kinda. Depends what you are encoding and what application you are using to do it. H.264 for example has a limited amount of parallelism in the encoding process, no matter what tool you're using. Some encoding tools are heavily threaded with some codecs. This is something you'd need to look at for your own personal application and ask: is this task more dependent on clock or threads?

Nehalem is much faster than Harpertown. There is not an argument to be made here. In memory bandwidth, in performance per clock, and you have to consider turbo here. That 2.66GHz will be running nearer 2.9GHz under lightly threaded workloads, and around 2.8GHz under even heavily threaded workloads. That's one of the places the good Xeons show an advantage over the normal i7. The binning allows the chips to take advantage of more aggressive turbo modes even under heavily threaded workloads. The X5550 for example runs at 2.8-2.9GHz all the time, and then 3GHz when lightly threaded.

 

[jb502]

Wait, what? A GPU benchmark is completely not applicable to encoding in any way. Unless you are going to use an encoder that leverages GPGPU (CUDA/OpenCL). Almost all programs are still using the CPU to encode, and thus the GPU has no benefit whatsoever.


Ehhh… kinda. Depends what you are encoding and what application you are using to do it. H.264 for example has a limited amount of parallelism in the encoding process, no matter what tool you're using. Some encoding tools are heavily threaded with some codecs. This is something you'd need to look at for your own personal application and ask: is this task more dependent on clock or threads?

Nehalem is much faster than Harpertown. There is not an argument to be made here. In memory bandwidth, in performance per clock, and you have to consider turbo here. That 2.66GHz will be running nearer 2.9GHz under lightly threaded workloads, and around 2.8GHz under even heavily threaded workloads. That's one of the places the good Xeons show an advantage over the normal i7. The binning allows the chips to take advantage of more aggressive turbo modes even under heavily threaded workloads. The X5550 for example runs at 2.8-2.9GHz all the time, and then 3GHz when lightly threaded.

Thank you for agreeing with me. I was about to lose all trust in this forum IF everyone agreed with the guy posting the video card comparison! LOL!


I am more interested in benchmarks like this one, where the 2.26 octo nehalem beats my 2.8 octo harpertown

http://www.macworld.com/article/139507/2009/03/macpro2009.html


quote:

It’s harder to recommend that people spend $800 more for the new $3,299 2.26GHz eight-core Mac Pro if they don’t run software written for the top-of-the-line Mac Pro’s eight processor cores. Though the 2.26GHz model was faster at most individual tasks than the previous eightcore 2.8GHz Mac Pro, and very fast in a few of our professional applications tests, it was slower than the new quad-core model at the majority of our application tests.

 

[goMac]

Wait, what? A GPU benchmark is completely not applicable to encoding in any way. Unless you are going to use an encoder that leverages GPGPU (CUDA/OpenCL). Almost all programs are still using the CPU to encode, and thus the GPU has no benefit whatsoever.

Depends. Are you encoding, or rendering AND encoding? If you're rendering, than a GPU benchmark is totally %100 applicable.

OP needs to decide what he's doing. Considering he's asking about iMovie and Premiere in the same sentence, he needs to do some tool research. If he's just encoding, heck, he could do that with free software.

If he's actually working with projects, then the GPU becomes very important, ESPECIALLY if he's using Premiere.

Ehhh… kinda. Depends what you are encoding and what application you are using to do it. H.264 for example has a limited amount of parallelism in the encoding process, no matter what tool you're using. Some encoding tools are heavily threaded with some codecs. This is something you'd need to look at for your own personal application and ask: is this task more dependent on clock or threads?

Ehhhhh, H2.64 is still pretty seriously parallelizable, unless you've got like 3 keyframes.

Nehalem is much faster than Harpertown. There is not an argument to be made here. In memory bandwidth, in performance per clock, and you have to consider turbo here. That 2.66GHz will be running nearer 2.9GHz under lightly threaded workloads, and around 2.8GHz under even heavily threaded workloads. That's one of the places the good Xeons show an advantage over the normal i7. The binning allows the chips to take advantage of more aggressive turbo modes even under heavily threaded workloads. The X5550 for example runs at 2.8-2.9GHz all the time, and then 3GHz when lightly threaded.

Performance per clock there is very little difference, almost all the differences are in the bus design.

(Note: For context, I've worked as a developer on video software that works with CS suite and Final Cut. I currently do real time video and audio processing development.)

 

[Evil Spoonman]

I am more interested in benchmarks like this one, where the 2.26 octo nehalem beats my 2.8 octo harpertown

http://www.macworld.com/article/139507/2009/03/macpro2009.html

That must be a fairly heavily threaded test. The 2.8GHz Harpertown will beat the 2.26GHz Nehalem some of the time, until you get to decently threaded applications, or certain memory bandwidth dependent applications. The 2.8GHz is not a slow machine by any means, it's almost just a wash. It just depends what task you're optimizing for. Also the 2.26GHz uses cheaper and faster DDR3 1066MHz memory without heat spreaders, which is a big plus IMO.