Power Mac G5 Quad Benchmarks

[andiwm2003]

No, I won't. Because I'll buy a dual socket quad core version and still be faster and more expensive.

The people that are going to cry are those that buy the sub $1000 multi-core system and it's about 3% faster than the previous generation system because of most applications out there won't know what to do with a multi-threaded processor.

Most of the benchmarks out there that are beautiful on the dual or better systems out there are multi-threaded.

But MS Word, Excel, Powerpoint, Quicken, whatever most regular non-power users use won't run any faster on a dual or quad core system because they don't know what to do with more execution units.

Gamers will be ticked because their machines will be slow...

Just my $0.02

i'm one of the non-power users that want to buy a dual core (of course a mac, not a windows box ).

i don't expect single apps to run faster, i expect the system to stay responsive while something runs in the background (burning a cd, running a video podcast). that is where for the consumer the strength of a dual core system is. running multiple apps without slowing down the system responsiveness.

 

[longofest]

So, this forum has been posting all about the XBench and Cinebench benchmarks, and been drooling over the CPU's (whether it be AMD Opterons or Athlons or PowerPC 970 MP's).

Now, I'm all about that. It's fun and insightful. But I want to draw people's attention briefly to some tools that ATI has recently introduced (right now only for Windoze). http://digg.com/software/ATI_Delivers_GPU-Accelerated_Video_Transcoding_ This App will transcode MPEG-2 to MPEG-4 (5-min clip) in 24 seconds using ATI's latest GPU's to accelerate the process.

Apple has begun to get on this bandwagon with Core Image, but I think we are going to see more and more emphasis on shunting this kind of work from the processor to the GPU. I wouldn't be surprised if it is a feature of QT8 and 10.5.

 

[foniks2020]

GPUs == next big thing??

So, this forum has been posting all about the XBench and Cinebench benchmarks, and been drooling over the CPU's (whether it be AMD Opterons or Athlons or PowerPC 970 MP's).

Now, I'm all about that. It's fun and insightful. But I want to draw people's attention briefly to some tools that ATI has recently introduced (right now only for Windoze). http://digg.com/software/ATI_Delivers_GPU-Accelerated_Video_Transcoding_ This App will transcode MPEG-2 to MPEG-4 (5-min clip) in 24 seconds using ATI's latest GPU's to accelerate the process.

Apple has begun to get on this bandwagon with Core Image, but I think we are going to see more and more emphasis on shunting this kind of work from the processor to the GPU. I wouldn't be surprised if it is a feature of QT8 and 10.5.

I've spoken with multiple knowledgeable friends on this over the last 4 years... we've all agreed that GPUs have the ability to do some incredible things when it comes to number crunching.... and for OS X this should be even more pronounced in the near term as you mentioned... they already have a processing workflow built in to the core of the os, so creating extensions that use those libraries in new innovative ways or 'extend' them beyond their current capabilities to really interesting applications, should not take long.

 

[longofest]

Digg this story!

Can you digg it?

http://digg.com/apple/First_PowerMac_G5_Quad_Benchmarks_Posted

 

[Trekkie]

Yesterday?

Quad-core Intel systems are available now.... (http://www1.us.dell.com/content/pro...0?c=us&cs=555&l=en&s=biz&~tab=specstab#tabtop)

It's just a matter of when Jobs wants to ship them.

Uh, those are dual core.

And you might do a search on Paxville vs. AMD Dual core and see how bad the performance is and how they consume some massive amounts of power by comparison.

 

[Chad A Wright]

When we're all Intel fanboys, the people with expensive quads will feel weird having a machine that went obsolete so quickly. Still they're the early adopters who'll probably always have to have the latest and greatest. NOt me, not anymore, not when I add up what I've spent over the years playing that game...

I'm bewidered by those people enthusimg over a platform thats almost obsolete...the King is dead...long love the King.

For some of us, it's not about what chip is inside, or how much RAM we have. For me it's about the work I do on it. People are impressed when they walk into my office and see the 30" ACD. I hope that they are always more impressed by the work on that display.

Making my living on these things means that I don't have time for computer problems. The Quad is perfect for me. The software out now will do truly great things, and the computer itself will last for many years to come. This will get me through the first few years of the Intel transition, and beyond. The Quad has plenty of head room and is a great value, especially considering that it is only $300 more than what the 2.7 was.

For some of us, it's the perfect time and the perfect computer.

 

[Trekkie]

i don't expect single apps to run faster, i expect the system to stay responsive while something runs in the background (burning a cd, running a video podcast). that is where for the consumer the strength of a dual core system is. running multiple apps without slowing down the system responsiveness.

expect to be disappointed (at least on an Intel platform speaking for todays architectures as dual core is just now appearing)

Multi core architectures aren't something you can just toss in an expect the software to take advntage of. There is going ot need to be some serious overhauling done of both the operating systems as well as the applications to easily benefit from what you're describing.

Intel knows this. Intel is offering tons of tools (sorry can't find a web link) for helping coprorate endusers, and what not, to re-write their applications to do what you're wanting. But as it stands today right now dual core isn't going to double the performance of most things.

What would happen now is the OS would try it's best to chuck requests to the other cores but you'd have some stalling happening because thread one on core 1 needs to finish before thread two on core two can execute, and what not.

Thought I really think the concern on the power mac's is all moot. I'd bet money we won't see powermacs with intel until late '06 at the earliest. the first generation or so of Intel's processors (dual socket) are going to be 145W of power per socket. That's a crapload of power consumption. When they go 65 nm it'll get a lot better and that's when I bet Apple will move to them. For reference looking at IBM's website (I can't find the link) was around 35W per socket. that's one heck of a difference and a lot more liquid cooling or noisy fans...

That being said my finger has been hovering over a 'BUY' button on the dual core dual socket PM 2.5GHz for a few days now...I really want to use Aperture and it won't run on my Rev A iMac G5 (video card not supported)

 

[rantsky]

Dual Core Works

expect to be disappointed (at least on an Intel platform speaking for todays architectures as dual core is just now appearing)

Multi core architectures aren't something you can just toss in an expect the software to take advntage of. There is going ot need to be some serious overhauling done of both the operating systems as well as the applications to easily benefit from what you're describing.

Intel knows this. Intel is offering tons of tools (sorry can't find a web link) for helping coprorate endusers, and what not, to re-write their applications to do what you're wanting. But as it stands today right now dual core isn't going to double the performance of most things.

What would happen now is the OS would try it's best to chuck requests to the other cores but you'd have some stalling happening because thread one on core 1 needs to finish before thread two on core two can execute, and what not.

Thought I really think the concern on the power mac's is all moot. I'd bet money we won't see powermacs with intel until late '06 at the earliest. the first generation or so of Intel's processors (dual socket) are going to be 145W of power per socket. That's a crapload of power consumption. When they go 65 nm it'll get a lot better and that's when I bet Apple will move to them. For reference looking at IBM's website (I can't find the link) was around 35W per socket. that's one heck of a difference and a lot more liquid cooling or noisy fans...

That being said my finger has been hovering over a 'BUY' button on the dual core dual socket PM 2.5GHz for a few days now...I really want to use Aperture and it won't run on my Rev A iMac G5 (video card not supported)

My experience is really different! A friend of mine has an AMD X2 3800+ and the system works great with dual core. You can run any single- or multi-threaded program and select on which core (or cores if is multi-thr) it will run on. You get one core busy crunching numbers, while the system is still fully functional on the other core. Makes you wish you had many more cores actually! Nowadays everyone is web surfing + downloading + burning + playing music + chatting + skyping + compressing + working(?) + ... at the same time. The gain from multicores can be incredible even if every app is single-threaded! Quad-core really seems quite natural at this point.

 

[scifiman]

Uh, those are dual core.

And you might do a search on Paxville vs. AMD Dual core and see how bad the performance is and how they consume some massive amounts of power by comparison.

You're right that those are dual core but the system can be configured to have 2 dual core procs so AidenShaw is right in that the system has 4 computing cores. As for the Intel vs. AMD debate, you're absolutely right that AMD trumps Intel in power consumption and performance per watt. Any current gen. Intel processor, save the Pentium M, is obsolete right now compared to an AMD proc imho. Having said that, the future Intel roadmap does loop pretty impressive. They still have to fix the limitations with their FSB (front-side bus) and the move to a smaller die should really help with the power consumption.

Edited because I misread the post above.

 

[AidenShaw]

Uh, those are dual core.

And you might do a search on Paxville vs. AMD Dual core and see how bad the performance is and how they consume some massive amounts of power by comparison.

I said "Quad-core Intel systems", not quad-core chips.

Since the number of cores is the most important factor (not the number of chips), calling the dual-dual PMG5 a "quad" is reasonable. Calling dual-dual Xeons "quads" is also reasonable.

As for performance, the postings seen so far lack a certain rigour in their analyses - insufficient numbers of benchmarks done on early hardware that don't test the whole system. It's too early to dismiss Paxville as a failure....

 

[bdjones]

Does this make any sense to anybody?

yes! Check out this page. It has a the results for CineBench run on many machines Mac and PC. It makes a lot of sense in relation like that, particularly to users of Cinema4D

http://www.3dfluff.com/mash/cbtop.php

It would be great if you could post your results up there.

 

[AidenShaw]

this doesn't add up

What would happen now is the OS would try it's best to chuck requests to the other cores but you'd have some stalling happening because thread one on core 1 needs to finish before thread two on core two can execute, and what not.

You seem to be confusing "Hyper Threading" (or SMT) with dual cores.

Dual cores are equivalent to dual processors - especially on Intel (since the cores currently communicate on the FSB - just like a two-chip system).

There is no need imposed by the CPU for one thread to finish before another can start. Application synchronization protocols may require this - but again, it's the same as any other multi-processor.

The original poster was describing running multiple applications at once. These will have no synchronization issues whether on dual-core or dual-chip systems - the threads in the apps are independent.

 

[Synchro]

Car analogies

Please people stop the Car analogies it's stupid and really dosen't make any comparision to mac and PC. Guys don't buy BMW for speed/performance , they buy them so chicks will hop in and screw thier brains out. Those cars are all about showing off , not abosolute value.

I agree that car analogies do get very silly, however, there's an interesting stat that might come as something of a shock to US readers (given typical viewpoints like Jiggie2g's and the frequent comparison of Apple to BMW in order to characterise them as a minority premium product): in the UK, in the midrange (Mondeo/3-series), BMW now outsells Ford. In fact BMW outsells everyone in this market.

Bodes well for Apple I'd say. Or maybe us Brits just have better taste in cars.

 

[Naimfan]

I've got to say I find the posted benchmarks for the Quad to be impressive, even allowing for the difficulty in making direct comparisons, although I'd also agree with those who want to see it in action on real world work. I was actually inspired, if that is the right word, to run Xbench and Cinebench on my iMac. The highest score I got on Xbench was 82.66--a little more than half of what what the Quad achieves.

And a question on Cinebench--the higher the number, the better? The only number I remember with specificity is the "OpenGL speedup" or something like that--it was 5.88. Is that good, bad, or indifferent?

Best,

Bob

 

[AidenShaw]

compare prices

in the UK, in the midrange (Mondeo/3-series), BMW now outsells Ford. In fact BMW outsells everyone in this market.

Bodes well for Apple I'd say. Or maybe us Brits just have better taste in cars.

What's the average selling price of the Mondeo vs. the 3-series in the UK?

In the US, BMW loads each car will just about every conceivable option package - which makes the entry price for the 3-series much, much higher than the entry price for the Ford. ($30K vs $21K)

 

[840quadra]

As for performance, the postings seen so far lack a certain rigour in their analyses - insufficient numbers of benchmarks done on early hardware that don't test the whole system. It's too early to dismiss Paxville as a failure....

That is the same opinion I have over the benchmarks on this new Quad core G5 Powermac. People are getting too hyper sensitive, and posting other benchmarks comparing scores with the G5 calling one faster or slower.

I have run into situations were a customer will have a SQL query run faster on one of their older servers then when the same query was run on a new system with faster processors. They then use this as a benchmark, and complain that the new $5000 HP server we sold, configured and manage for them was a waste of money. They ignore the fact that this new server is faster at running their application, terminal server, and at processing their data.

The problem with benchmarks is the same as you listed, and they don't tell the whole story about the system. The fact of the matter is, perceived or benchmarked speeds all depend on how the device is used, and if processes being used are optimized for that system.

 

[Azurael]

I can't come close to anything else, even with a little overclocking on the side... Still, nice to know my poor 'lil Sempron 3000+ lappy doesn't overheat or crash when OC'd to 1980MHz Especially impressive considering the fact that it's smaller than a 12" iBook. I might have to get myself a new CPU... maybe a Turion MT-40 to replace the crappy 'lil Sempron. I want a new Mac!

CINEBENCH 2003 v1
****************************************************

Tester : Azurael

Processor : MSI MS-1013 Sempron 3000+
MHz : 1980
Number of CPUs : 1
Operating System : Windows XP SP2

Graphics Card : Radeon XPRESS 200M
Resolution : 1280x800
Color Depth : 32-bit

****************************************************

Rendering (Single CPU): 277 CB-CPU

****************************************************

 

[finchna]

i'm one of the non-power users that want to buy a dual core (of course a mac, not a windows box ).

i don't expect single apps to run faster, i expect the system to stay responsive while something runs in the background (burning a cd, running a video podcast). that is where for the consumer the strength of a dual core system is. running multiple apps without slowing down the system responsiveness.

You can also get that with one of the older dual G5 macs--might save a bit if you can find one on sale.

 

[manu chao]

My experience is really different! A friend of mine has an AMD X2 3800+ and the system works great with dual core. You can run any single- or multi-threaded program and select on which core (or cores if is multi-thr) it will run on. You get one core busy crunching numbers, while the system is still fully functional on the other core. Makes you wish you had many more cores actually! Nowadays everyone is web surfing + downloading + burning + playing music + chatting + skyping + compressing + working(?) + ... at the same time. The gain from multicores can be incredible even if every app is single-threaded! Quad-core really seems quite natural at this point.

And with OS X you don't have to select the core the OS will do for you automatically. Considering that, according to a poll on macosxhints (representative for geeks only, I guess), half of all users have seven or more GUI applications open at the same time, which translates into a multiple of this in terms of the number of threads. There are plenty of tasks already which can be processed parallely.

 

[dmglover]

Sure...

CINEBENCH 2003 v1
****************************************************

Tester : Pie

Processor : Quad G5
MHz : 2.5 GHZ
Number of CPUs : 4
Operating System : 10.4.2

Graphics Card : GeForce 6600
Resolution : <fill this out>
Color Depth : <fill this out>

****************************************************

Rendering (Single CPU): 359 CB-CPU
Rendering (Multiple CPU): 1016 CB-CPU

Multiprocessor Speedup: 2.83

Shading (CINEMA 4D) : 353 CB-GFX
Shading (OpenGL Software Lighting) : 1051 CB-GFX
Shading (OpenGL Hardware Lighting) : 1871 CB-GFX

OpenGL Speedup: 5.29

****************************************************


Does this make any sense to anybody?

I want a QUAD PowerMac NOW!! I am interested if anyone knows why the difference in Shading numbers.

Here is my Cinebench 2003 v1 results for my AMD System.

CINEBENCH 2003 v1
****************************************************

Tester : MG

Processor : ASUSA8VDAMD4200X2
MHz : 2200 Overclocked to 2425
Number of CPUs : 2
Operating System : Microsoft Windows XP Pro SP2

Graphics Card : NVidia GeForce 6800 Ultraa AGP
Resolution : 1280 x 1024
Color Depth : Highest 32bit

****************************************************

Rendering (Single CPU): 342 CB-CPU
Rendering (Multiple CPU): 636 CB-CPU

Multiprocessor Speedup: 1.86

Shading (CINEMA 4D) : 333 CB-GFX
Shading (OpenGL Software Lighting) : 1827 CB-GFX
Shading (OpenGL Hardware Lighting) : 3277 CB-GFX

OpenGL Speedup: 9.83

****************************************************

 

[AidenShaw]

only fools and experts set affinity

And with OS X you don't have to select the core the OS will do for you automatically.

Windows will also do it automatically, and probably better than you can do it by hand unless you have a pretty unique and static workload.

Setting affinity manually often leads to one core sitting idle while multiple threads are fighting for the other core.

 

[Trekkie]

You seem to be confusing "Hyper Threading" (or SMT) with dual cores.

No, I'm not, I must not have gotten my point across

Dual cores are equivalent to dual processors - especially on Intel (since the cores currently communicate on the FSB - just like a two-chip system).

no, they're not. On a dual socket single core system you have two processors each with a dedicated socket and access to the memory controller.

With dual core you gave a dog more heads but didn't give it a thicker neck or stomach. This is true for current shipping AMD products or Intel products, they are socket swaps for existing single core to go to dual core. This will change over time as newer FSBs from Intel and the 'rev F' socket adds more HT links next year but for now this is a true statement.

There is no need imposed by the CPU for one thread to finish before another can start. Application synchronization protocols may require this - but again, it's the same as any other multi-processor.

I know, that's not what I said. What I said was Thread 2 relies on something Thread 1 is doing and until thread 1 completes thread 2 will stall. This has nothing to do with SMT where you have registers in use by thread 1 so thread 2 stalls in the pipeline.

The original poster was describing running multiple applications at once. These will have no synchronization issues whether on dual-core or dual-chip systems - the threads in the apps are independent.

Yes, there is. Because the OS is going to take all these applicaitons and try to multiplex them through the system threads it can receive. When you've got a single threaded app spread across two to four cores it's going to stall backing up other apps being spread like peanut butter across the cores. The OS needs to get smart about how it handles the scheduling of threads and you can and will see performance improvements. But not every app is a multi-core aware application and the OS tries its best to make them that way, causing delay and latency.

 

[gugy]

I think these quad machine are made for those PPC people who just have to have bragging rights, now, until the Intel Macs come out, which would have to smoke the quads, else why would Apple go that way.

When we're all Intel fanboys, the people with expensive quads will feel weird having a machine that went obsolete so quickly. Still they're the early adopters who'll probably always have to have the latest and greatest. NOt me, not anymore, not when I add up what I've spent over the years playing that game...

Apple is just milking a small segment of its market, with always too much money to spend.

I doubt if any/much software which truly uses the quad PPC power will ever be written...for such a small, soon to be an always declining percentage of Mac users, as the Intel move takes hold.

I'm bewidered by those people enthusimg over a platform thats almost obsolete...the King is dead...long love the King.

Sorry, I don't agree with you at all.
The PM Intel machines might be here around middle of 2007.
For people who needs new PM now the Quad seems to be the right thing to purchase it.
Even if Intel is amazing and smoke away the quad in 2007 it doesn't make any sense to hold of a quad purchase now. Personally, I think it's safer for professionals like me to have a quad and get a rev. B PM Intel. The revB might be here at the end 2008. So that would be enough time to have a solid machine on the Quad working for you.
I would rather wait and see how Rosetta works under the Intel.
The most important thing is if you need a computer now and that will help you make money go and get the Quad. People who wait for the latest and most amazing computer will only get disappointed, because always something better is on the horizon.
The quads seems to be very solid. I just worked on a 2.5 single core PMs and it was amazing. I can't imagine what my new Quad 2.5 with 7800 GT card will look like. Unfortunately, I'll have to wait until Xmas to find out!

 

[AidenShaw]

prize winner - worst analogy

no, they're not. On a dual socket single core system you have two processors each with a dedicated socket and access to the memory controller.

Intel has a shared FSB - it doesn't matter if two cores share the single FSB through one socket or two.

The G5 has per-socket FSBs to a shared memory controller, so the dual chip might have an FSB advantage - but one that's unlikely to matter since the memory controller is a bottleneck.

AMD has per-socket memory controllers. A dual-core chip might outperform two single-cores because all the memory is accessible to a dual-core without going over the high-latency HT link which the two single cores have to use. This, of course, will be application and OS dependent (Win2k3 and the latest Linux kernels know about the NUMA topology and can do some improved memory allocation and scheduling).

When you've got a single threaded app spread across two to four cores it's going to stall backing up other apps being spread like peanut butter across the cores.

This is not only one of the worst analogies I've ever seen posted, it's not even close to the way OS schedulers work.

A single thread cannot be "spread" across more than one core simultaneously - it can only be scheduled on one CPU at a time. The other CPUs (cores) can be scheduled with other threads from other single threaded applications, without any CPU level contention. (Disks, network and memory contention may be happening - but your description of one thread clogging all the cores is absurd.)

But not every app is a multi-core aware application and the OS tries its best to make them that way, causing delay and latency.

Please explain this, I can't make any sense of it.

 

[Booga]

Four benchmarks

Someone should write an AppleSCript that starts 4 copies of MacBench simultaneously.