From the Horse's Mouth: 2.26 faster than 3.2 (not confirmed)

[thedarkhorse]

I too am mostly interested in harpertown octo vs new Quad, especially in video conversions and rendering. Not just h264, but editing formats like prores and hdv/xdcam.
With only a 1.1x-1.2x performance increase from top end 2008 octo core to top end 2009 octo core(for these codecs) I have a hard time believing that a nehelam quad core will offer an actual increase over the harpertown octos, it seems like it would be a decrease in these circumstances(which would be what I'd be getting it for-editing).

 

[shokunin]

It's obvious that these benchmarks have more holes than Swiss cheese, but we want to get an idea of how the new single quad compares to the dual harpertown model. Not sure why I thought of underclocking (tipsy post).

Instead of a canned benchmark, try converting that file I posted from the current wmv to MPEG4 with HB and see how long of takes. Quick and dirty, but close enough.

m1stake,

I tried downloading your file but I hangs and I end up with an error the URL could not be retrieved message.

 

[Chaos123x]

2.26Ghz Faster than 3.2ghz

Malfunction Malfunction Brain Over Load

Does not compute

1+2=4 1+2=4 1+2=4

System shutdown.

 

[More]

Willing to offer to help crack any chaos theories out there with my 2008 3.2Ghz Mac Pro.

My system: 2008 Mac Pro 8 Core 3.2Ghz, 8GB RAM (4GBx2), 8800 GT, Dual Intel X25-M Drives (RAID-0) for Boot, Dual WD RE3 RAID-0 Scratch/Media.

Let me know what I can do to help

 

[nanofrog]

I meant you could supply the i7 numbers. I'm sure someone around here has a 3.2Ghz Mac Pro.

Looking around the forum, I found another member has already done it, and ran Geekbench.

Well I don't have the 3.2 but my Octo MP is base 2.8ghz and I have underclocked my Hackintosh i7 920 to 2.26 (by lowering the multipler).

Geekbench Score stock Mac Pro 2.8 Octo-core= 9143
Geekbench Score underclock 920 @ 2.26ghz quad-core = 6841

Now if we get an octo 3.2 benchmark..

Geekbench Score 920 @ 3.2ghz quad-core = 9679

And he compared it to his '08 2.8 Octo MP. Not a direct comparison (Quad vs. Octo), but hey, it's a start.

Assuming an Octo version of the 2.26 could produce 1.5x the Quad result (= 10,262), that wouldn't be too bad. But it's hard to estimate the multiplier correctly, so I'd prefer some additional testing.

 

[m1stake]

Ug, I hate GeekBench and xBench. They don't tell you anything

 

[Jouhne]

Ug, I hate GeekBench and xBench. They don't tell you anything

Sure it does show you something look :

Quad Xeon 2,66 (2006) - Mine for the moment
Proc. Int. : 4727
Proc. Float. : 8466
Mem. Perf. : 2142
Mem. Band. : 1934

Quad Xeon 2,80 (2008) - why I didn't upgrade...
Proc. Int. : 5053
Proc. Float. : 8802
Mem. Perf. : 2883
Mem. Band. : 2406

Octo Xeon 2,80 (2008) - well my apps doesn't multicore well, damn Adobe.
Proc. Int. : 8421
Proc. Float. : 15980
Mem. Perf. : 2319
Mem. Band. : 2037

Quad Core i7 940 2,93 GHz(2009) - Mmh
Proc. Int. : 8722
Proc. Float. : 9751
Mem. Perf. : 4468
Mem. Band. : 4607

So there you see why Geekbench can show you things.
QPI is a monster ! And that's the biggest change. Here I hope that the Xeon 3500 series got the QPI to 6,4 GT/s and not locked to 4,8 GT/s (like the core i7). And there you go, a Quad 2,93 will destroy an Octo 2,80... for a lot of things. I can't imagine the speed of an Octo Nehalem with 2 QPI if the mem perf double... scary.
But let's be realistic and create a GB for the Octo Nehalem 2,93 GHz, maybe we can have something like this :
Proc. Int. : 14500
Proc. Float. : 17500
Mem. Perf. : 6000 (QPI 6,4 GT/s)
Mem. Band. : 6000 (QPI 6,4 GT/s)

PS : remember that you can have these kind of performance if your apps are multicore aware.

 

[m1stake]

They're arbitrary numbers in canned situations. While they give you a general idea that X is faster than Y, using "real" files to do "real" things (FPS in games, or render time, etc) gives you an idea of what the computer will do for applications you actually use.

Your numbers are fairly helpful though, would you mind giving us your configurations?

 

[Sun Baked]

Maybe for a database search, if it is something you can load into memory.

But not today, and not until most update their apps to be Snow Leopard optimized and highly multi threaded applications.

 

[shokunin]

Finally downloaded the video file m1stake posted. I'm using Handbrake 0.9.3 using the cobrent.wmv file by m1stake and encoding using the preset Handbrake -> High Profile -> Animation settings and then letting it rip.

Baseline:
2.8ghz Octo Mac Pro (Early 2008) = 7 min 59 seconds

Nehalem 920:

Turbo boost is active in Bios, which gives an extra 5% clock speed increase above the stated processor speed. The Multiplier goes from 20 to 21 when 4 cores are in use and thermal throttling has not kicked in. All times listed below are QUAD core. The Ghz listed is the actual speed before any turbo-boosting is engaged.

920 @ 3.6ghz = 6:50 14.4% Faster
920 @ 3.2ghz = 7:42 3.5% Faster
920 @ 2.92ghz = 8:30 6.5% slower
920 @ 2.66ghz = 9:19 16.7% slower
920 @ 2.26ghz = 11:25 ** No turbo clock applied 43% slower
920 @ 2.40ghz = 10:49 ** Theoretically 2.26 w/ turbo boost by upping the multiplier by one. 35.5% Slower

Just for kicks I did the same conversion on my 1.86ghz /128GB SSD MacBook Air = 1:07:22, yes 1 hour, 7 Minutes and 22 seconds.

 

[m1stake]

Would you mind sharing your configurations with us?

 

[shokunin]

Here's my system info for reference, both systems are running Leopard 10.5.6.

2.8ghz Octo Mac Pro
16GB Ram
Velociraptor Boot Drive
nVidia 8800GT

Hackintosh Core i7 920
Vanilla install Leopard 10.5.6

DFI x58-T3H6 micro ATX motherboard
Core i7 920 Processor 2.66ghz overclocked to 3.6ghz
Noctua NH-U12P se1366 heatsink / fan
6GB (2GB x 3) DDR3 Super Talent RAM
WD Raptor Boot Drive
Corsair HX520 PSU
ATI Radeon 2600 XT (pulled from Mac Pro)
Silverstone TJ-08 Micro ATX tower
no optical drives

 

[Nik]

Could you please run a photoshop benchmark for us with the same clock speeds? Especially the i7 at 2.66 and 2.93 GHz vs. the 2x2.8GHz MacPRo.

This would be much appreciated! Thanks!

 

[myca]

Ouch, so heres more (preliminary) proof that last years octo 2.8 may out perform the new quad 2.66, meaning that a year old system that was cheaper and more expandable could prove to be more powerful (at some tasks at least) than the new model.

Theses results do bode well for the higher clocked dual configurations though, they're gonna fly, if only they'd have stuck 9 ram slots in there for extra chewiness

 

[nanofrog]

Here I hope that the Xeon 3500 series got the QPI to 6,4 GT/s and not locked to 4,8 GT/s (like the core i7). And there you go, a Quad 2,93 will destroy an Octo 2,80... for a lot of things. I can't imagine the speed of an Octo Nehalem with 2 QPI if the mem perf double... scary.

Assuming Intel sticks to what they've done on the Core i7 alone, the top model (W3570), is likely the only part to have an unlocked multiplier.

The addition of the 2nd QPI won't double the memory throughput. It's used to connect to the other CPU. As each has a 2nd QPI, the configuration produces a full duplex implementation to prevent it from being a bottleneck.

 

[G4DP]

I rang up apple, they told me the sun shines out of Steve's arse.



The new base line will be the same as the old base line. People who think that they will give a 40% increase over the last generation are either stupid or easily fooled.

 

[myca]

The new base line will be the same as the old base line. People who think that they will give a 40% increase over the last generation are either stupid or easily fooled.

I would beg to differ, the first gen mac pros were all dual dual cores, then (after the intro of the high end octo core bump) the next refresh was all dual quad cores, so that refresh gave a massive boost to the base line model, whereas this refresh may very well see you paying more (by UK price) and possibly getting a lesser machine in some respects, at the base end of course.

I've quoted this a number of times, but I'm gonna quote it again.

From Apples marketing of the early 2008 mac pros

8-core processing was once reserved as a high-end option. Now it's at the heart of the Mac Pro line.

So going by Apples marketing, it would seem that they have ripped the heart out of the new base mac pro

 

[G4DP]

Myca, I was just stating that the new base machine is not going to be a significant improvement over the old machine.

I just ran the suggested benchmark. Took 8 minutes 2 seconds to convert file. This was with Seti running on 4 cores in the background.

Set up:

2 x 2.8GHz
10GB Ram
320GB Drive
Radeon 2600

 

[myca]

Myca, I was just stating that the new base machine is not going to be a significant improvement over the old machine.

I just ran the suggested benchmark. Took 8 minutes 2 seconds to convert file. This was with Seti running on 4 cores in the background.

Set up:

2 x 2.8GHz
10GB Ram
320GB Drive
Radeon 2600

I know, sorry that's me just getting picky

But the thing that's miffed me is that it looks as though the new base model for real world uses could be worse than the previous base model, which when it costs more is a bit of a kick in the teeth.

At least the higher end models do look pretty strong, but they're beyond my needs and budget, that's why I've ordered some parts to build my own machine.

BTW if the base model had remained a dual quad core I'd have splashed out on one, as I had £2000 set aside, with the expectation that the price would increase by a couple of hundred quid.

 

[Abidubi]

920 @ 3.6ghz = 6:50 14.4% Faster
920 @ 3.2ghz = 7:42 3.5% Faster
920 @ 2.92ghz = 8:30 6.5% slower
920 @ 2.66ghz = 9:19 16.7% slower
920 @ 2.26ghz = 11:25 ** No turbo clock applied 43% slower
920 @ 2.40ghz = 10:49 ** Theoretically 2.26 w/ turbo boost by upping the multiplier by one. 35.5% Slower

We are talking about a 4 core i7 right? Meaning a 2.66GHz 4 core (the base mac pro) would be only 16.7% slower than an 8 core 2.8 last gen.

Also meaning that an 8 core 2.26 2009 could "potentially" do it in 6:12 (using your no turbo number), making a 2.26 octo 28.7% faster than a 2.8 octo. Still costs more than the 2.8 octo did though.

How fast would the octo 3.0 do this test and how much did that cost?

 

[phairphan]

I'm sure this was from the horse, I just don't think it was from his mouth.

From the Horse's Mouth: 2.26 faster than 3.2 (not confirmed)

 

[Boneoh]

I'm sure this was from the horse, I just don't think it was from his mouth. [/I]

LOL That was a good one

 

[shokunin]

We are talking about a 4 core i7 right? Meaning a 2.66GHz 4 core (the base mac pro) would be only 16.7% slower than an 8 core 2.8 last gen.

Also meaning that an 8 core 2.26 2009 could "potentially" do it in 6:12 (using your no turbo number), making a 2.26 octo 28.7% faster than a 2.8 octo. Still costs more than the 2.8 octo did though.

How fast would the octo 3.0 do this test and how much did that cost?

Yes, we are comparing 4 cores versus 8. Amazing isn't it that Nehalem can smoke 8 cores. However, this is the where the MOST performance gains there will be, don't expect all benchmarks to show as large of gain.

The "turbo-boost" is an Intel Nehalem specific process and even Apple tout's the use of turbo-boot on the new Mac Pro, stating that the 2.93ghz version can turbo boost up to 3.33ghz (and additional 3x multiplier (3x133 bus speed) when only ONE core is needed. My 920 is limited to a 2x multiplier when 1 core is needed. This is kind of confusing since most the time the processor is running faster than the advertised speed.

 

[AndyCorleone]

Mnnn I just did the same with my q6600 oveclocked at 2.71 (Intel stock Fan waiting postman for a better Cooler ) I takes me 00:12 min 25secons

Finally downloaded the video file m1stake posted. I'm using Handbrake 0.9.3 using the cobrent.wmv file by m1stake and encoding using the preset Handbrake -> High Profile -> Animation settings and then letting it rip.

Baseline:
2.8ghz Octo Mac Pro (Early 2008) = 7 min 59 seconds

Nehalem 920:

Turbo boost is active in Bios, which gives an extra 5% clock speed increase above the stated processor speed. The Multiplier goes from 20 to 21 when 4 cores are in use and thermal throttling has not kicked in. All times listed below are QUAD core. The Ghz listed is the actual speed before any turbo-boosting is engaged.

920 @ 3.6ghz = 6:50 14.4% Faster
920 @ 3.2ghz = 7:42 3.5% Faster
920 @ 2.92ghz = 8:30 6.5% slower
920 @ 2.66ghz = 9:19 16.7% slower
920 @ 2.26ghz = 11:25 ** No turbo clock applied 43% slower
920 @ 2.40ghz = 10:49 ** Theoretically 2.26 w/ turbo boost by upping the multiplier by one. 35.5% Slower

Just for kicks I did the same conversion on my 1.86ghz /128GB SSD MacBook Air = 1:07:22, yes 1 hour, 7 Minutes and 22 seconds.

 

[Tesselator]

Sure it does show you something look :

Quad Xeon 2,66 (2006) - Mine for the moment
Proc. Int. : 4727
Proc. Float. : 8466
Mem. Perf. : 2142
Mem. Band. : 1934

Quad Xeon 2,80 (2008) - why I didn't upgrade...
Proc. Int. : 5053
Proc. Float. : 8802
Mem. Perf. : 2883
Mem. Band. : 2406

Octo Xeon 2,80 (2008) - well my apps doesn't multicore well, damn Adobe.
Proc. Int. : 8421
Proc. Float. : 15980
Mem. Perf. : 2319
Mem. Band. : 2037

Quad Core i7 940 2,93 GHz(2009) - Mmh
Proc. Int. : 8722
Proc. Float. : 9751
Mem. Perf. : 4468
Mem. Band. : 4607

So there you see why Geekbench can show you things.

Many apps and games are multi-threaded these days, LightRoom 2.3 & 2.2, parts of Photoshop CS4 and many of it's plugins. Capture One, Lightwave 3D, FinalCut, Houdini, Quake3, Quake4, etc. etc. But I really wanted to comment on GeekBench.

Yes it can show you things. It's not that great for profiling and comparing very similar systems. Even on the same system running GeekBench multiple times can yield results with ±100 point differences in just about every category. It is limited to about that though so when you see a massive increase (like in your Core i7 940 score for memory) or decrease you will know there really are some differences.

Also I bet my 2007, 2.66 Quad (upgraded to a 2.66 Octad) does NOT score worse than the new $3,300 2.26GHz Octad. And I further bet it's real close to the $4,700 2.66Ghz Octad! Since I paid only $2,900 for it all total (after selling off the old procs and installing new x5355 ones) I'm going to agree with you that the difference of the new machines is not worth the money if you already have a Intel based MacPro of any kind. In my case i would be paying $3,000 (after selling my current Mac Pro for $1,700) for about a 8% ~ 12% increase in speed.

Looky here. http://browse.geekbench.ca/geekbench2/view/94982 These are my GeekBench results:

2007 MacPro 2.66 (Quad upgraded to an x5355 2.66 Octad)
Proc. Int. : 9087
Proc. Float. : 13231
Mem. Perf. : 2097
Mem. Band. : 1825

And I'm not even the best score. The best (with the exact same system) is:

2007 MacPro 2.66 (Quad upgraded to an x5355 2.66 Octad)
Proc. Int. : 10374
Proc. Float. : 16177
Mem. Perf. : 2168
Mem. Band. : 1926
http://browse.geekbench.ca/geekbench2/view/113749