Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Apple's New Mac Pro Begins Showing Up in Benchmarks

It is quite remarkable to read all the pros and cons. I have been waiting years for a new Mac Pro and almost went Hackintosh.
I don't buy these benchmarks without a GPU test. I would also like to see the tests with a machine loaded with more than one core
I have to see how they work on photo and video editing for my business. Will there be a marked improvement. I am hoping so.
My wife needs to use this on epidemiology research on insect-borne diseases. That's a lot of scientific number crunching and heavy duty statistical analysis computation.
I will wait until we can test this ourselves before I shout "hooray" or throw up my hands in horror!!!! One cannot judge from these preliminary tests.
 
I haven't read all 25 pages of this thread, so perhaps someone else already pointed this out, but the nMP 12-core is a single processor model, and as such should be compared to the current 6-core model.

As such the geekbench performance has pretty much doubled.

Just sayin'....
 
zephonic said:
I haven't read all 25 pages of this thread, so perhaps someone else already pointed this out, but the nMP 12-core is a single processor model, and as such should be compared to the current 6-core model.

As such the geekbench performance has pretty much doubled.

Just sayin'....
Click to expand...

no you compare 12 core to 12 core 8 to 8, 6 to 6, and 4 to 4.
 
zephonic said:
I haven't read all 25 pages of this thread, so perhaps someone else already pointed this out, but the nMP 12-core is a single processor model, and as such should be compared to the current 6-core model.

As such the geekbench performance has pretty much doubled.

Just sayin'....
Click to expand...

Sure, but you have to consider that that processor will be over $2,000 and gives half the memory capacity of a dual processor system. Also with a dual processor system you could have had the option of two 3.5GHz 6-cores for the same sort of price.
 
Umbongo said:
Sure, but you have to consider that that processor will be over $2,000 and gives half the memory capacity of a dual processor system.
Click to expand...

OK, but what was the price of a W3680 when it first arrived in 2010? Wasn't it around $1500 or so?

Umbongo said:
Also with a dual processor system you could have had the option of two 3.5GHz 6-cores for the same sort of price.
Click to expand...

Fair enough. Again, I don't really have a point :D Just thought we should be comparing single-cpu to single-cpu, regardless of whether a dual-cpu model is available or not.
 
I was looking through the new mac pro specs and at first pretty disheartened by the single CPU offering. I was also annoyed at the lack of CUDA support in the GPU - the only thing I need my GPU to do is have CUDA.

But then I realised why they've chosen AMD FirePros - ECC RAM.

If you've got ECC ram you can farm out processes to the GPUs that normally a graphics card wouldn't be trusted to do. I'm guessing that with 6 teraflops of GPU power at our disposal this machine is going to be very, very fast.

It all now depends on how good Mavericks is - how well it can utilise the GPUs. Maybe that's the 'Maverick' part of Mavericks, uses GPUs like never before.

Either that or the new mac pro is a giant hairdryer.
 
GermanyChris said:
no you compare 12 core to 12 core 8 to 8, 6 to 6, and 4 to 4.
Click to expand...

I would contend that in order to ascertain value for money, you compare machines of the same price, and that the number of CPU cores a machine has is totally irrelevant. Alternatively, depending on your expenses, you could compare performance per unit of electrical energy consumption (which is still in effect a way of figuring out how much a machine will cost to run).
 
aggri1 said:
I would contend that in order to ascertain value for money, you compare machines of the same price, and that the number of CPU cores a machine has is totally irrelevant. Alternatively, depending on your expenses, you could compare performance per unit of electrical energy consumption (which is still in effect a way of figuring out how much a machine will cost to run).
Click to expand...

For many, "performance per system" is critical, since scaling performance by adding boxes is tricky and doesn't work well for many applications. Apple's single socket architecture will compare poorly against dual socket systems in this area.

There's no single "right way" to compare systems - it depends on the user and applications.
 
AidenShaw said:
Apple's single socket architecture will compare poorly against dual socket systems in this area.
Click to expand...

That will certainly be the case in benchmarks and applications that completely saturate all available cores. In reality, for most people and most applications, the core count race has diminishing returns. Compared to 12 cores, 16 cores will not translate to 33% faster in real world terms, nor would 24 cores be doubly fast.

This is where Apple's GPU bet comes in. In applications where parallelization is easily achieved, and where programmers have opened the gateway, the dual GPUs will likely boost performance per watt and per dollar far beyond what is achievable by adding Xeon cores. Count on it.
 
that's a hefty number, i wonder if there is a device with better benchmarks score than that. hoping for an october-november release.
 
clamnectar said:
This is where Apple's GPU bet comes in. In applications where parallelization is easily achieved, and where programmers have opened the gateway, the dual GPUs will likely boost performance per watt and per dollar far beyond what is achievable by adding Xeon cores. Count on it.
Click to expand...

... the dual socket systems have twice as many PCIe 3.0 lanes, so put Quad SLI or Quad Crossfire in them. ;)
 
clamnectar said:
That will certainly be the case in benchmarks and applications that completely saturate all available cores. In reality, for most people and most applications, the core count race has diminishing returns. Compared to 12 cores, 16 cores will not translate to 33% faster in real world terms, nor would 24 cores be doubly fast.

This is where Apple's GPU bet comes in. In applications where parallelization is easily achieved, and where programmers have opened the gateway, the dual GPUs will likely boost performance per watt and per dollar far beyond what is achievable by adding Xeon cores. Count on it.
Click to expand...
I don't know, Intel has been making it a time honored tradition of stealing AMDs lunch. I'm amazed AMD still exists.
Intel does it by consistently bringing out more powerful systems every year.

Point being they have an absurdly well funded R&D. I'd put my money on Intel to win every time whether it's dollar per core, performance per watt...
They may not beat every metric every time, but they prevail in most, and in all eventually.

As far as diminishing returns, what makes parallel GPU cores any better than parallel CPU cores?
 
calaverasgrande said:
I don't know, Intel has been making it a time honored tradition of stealing AMDs lunch. I'm amazed AMD still exists.
Intel does it by consistently bringing out more powerful systems every year.

Point being they have an absurdly well funded R&D. I'd put my money on Intel to win every time whether it's dollar per core, performance per watt...
They may not beat every metric every time, but they prevail in most, and in all eventually.

As far as diminishing returns, what makes parallel GPU cores any better than parallel CPU cores?
Click to expand...

there's more GPU cores than CPU.
 
AidenShaw said:
Hmmmm. Let me think about this logic a bit.

  • More general purpose CPU cores bad because apps don't parallelize well
  • More special purpose GPU cores good because apps do parallelize well
Click to expand...

you misunderstand I'm not defending any of Apples decisions regarding the MP I've sold both of mine since there is no truly viable upgrade path for me. He asked what makes them "better" and that is the answer. We can debate merits all day long but in the end Apple made a call. Now you either jump on or jump off. I choose off.
 
AidenShaw said:
Hmmmm. Let me think about this logic a bit.

  • More general purpose CPU cores bad because apps don't parallelize well
  • More special purpose GPU cores good because apps do parallelize well
Click to expand...

1. CPU cores don't scale well because the kind of code that CPUs excel at is less easily parallelized

2. For tasks that parallelize easily, having a few more CPU cores is not as much of an edge as having thousands of GPU cores.
 
clamnectar said:
1. CPU cores don't scale well because the kind of code that CPUs excel at is less easily parallelized.
Click to expand...

A counter example is embarrassingly parallel code. A common example is a web server - each http request is an independent task that can be handled in parallel.


clamnectar said:
2. For tasks that parallelize easily, having a few more CPU cores is not as much of an edge as having thousands of GPU cores.
Click to expand...

But you're defining "parallelize easily" as "parallelizable on GPGPU architectures".

GPUs are designed specifically for graphics and thus are very restrictive in operations and programming. Due to their design, GPUs are only effective for problems that can be solved using stream processing and the hardware can only be used in certain ways.

http://en.wikipedia.org/wiki/Gpgpu
Click to expand...

The Mini MacPro will be good for some classes of applications.

A dual socket system will be better at other classes, and a dual socket with dual, triple or quad GPUs will be better at the CPU tasks and at the GPGPU tasks.

There's no "special sauce" to the MMP - it uses the same CPU chips and GPUs as other workstations. It just has much less flexibility (like max memory).
 
AidenShaw said:
Hmmmm. Let me think about this logic a bit.

  • More general purpose CPU cores bad because apps don't parallelize well
  • More special purpose GPU cores good because apps do parallelize well
Click to expand...
My point exactly.

I'd bet on the general purpose CPU cores as they have faster more direct access to, well everything except the GPGPU ram cache. Now some GPUs do have blazing faster DDR ram than what you ordinarily find on a desktop. But the difference in throughput is not nearly as big as it was just a couple years ago.
 
calaverasgrande said:
My point exactly.

I'd bet on the general purpose CPU cores as they have faster more direct access to, well everything except the GPGPU ram cache. Now some GPUs do have blazing faster DDR ram than what you ordinarily find on a desktop. But the difference in throughput is not nearly as big as it was just a couple years ago.
Click to expand...

My point is that the people who think that the MMP (Mini MacPro) is revolutionary are missing a few things...

  • a dual socket system has up to twice as many CPU cores, twice as many PCIe lanes, up to four times as many DIMM slots, up to twice as many GPUs

If CPU cores are good, the dual socket system can have twice as many. If GPGPU cores are good, the dual socket system can have twice as many.

Want to propose some serious benchmark that an HP Z820 with 24 cores, 512 GiB RAM and Quad SLI GPUs will lose to an MMP? (And please don't bring up some absurd metric like "web pages viewed per hour per dollar" - focus on applications where people will gladly spend from several thousand to several tens of thousands of dollars on the workstation.)

There are some important apps that can use GPGPU stream acceleration - but with a dual socket system you can have twice as many GPGPU cores to run them. No advantage for the MMP here.

The MMP is pretty, but so was the Cube (except for the cracks).

And the MMP won't be pretty with a half-dozen T-Bolt add-ons cabled to it.

The MMP will sell well to the posers who want to think that they are too important to use an Imac.
 
AidenShaw said:
My point is that the people who think that the MMP (Mini MacPro) is revolutionary are missing a few things...

  • a dual socket system has up to twice as many CPU cores, twice as many PCIe lanes, up to four times as many DIMM slots, up to twice as many GPUs

If CPU cores are good, the dual socket system can have twice as many. If GPGPU cores are good, the dual socket system can have twice as many.

Want to propose some serious benchmark that an HP Z820 with 24 cores, 512 GiB RAM and Quad SLI GPUs will lose to an MMP? (And please don't bring up some absurd metric like "web pages viewed per hour per dollar" - focus on applications where people will gladly spend from several thousand to several tens of thousands of dollars on the workstation.)

There are some important apps that can use GPGPU stream acceleration - but with a dual socket system you can have twice as many GPGPU cores to run them. No advantage for the MMP here.

The MMP is pretty, but so was the Cube (except for the cracks).

And the MMP won't be pretty with a half-dozen T-Bolt add-ons cabled to it.

The MMP will sell well to the posers who want to think that they are too important to use an Imac.
Click to expand...
Honestly, I should shut up now and walk away, but I cannot. The only real problem I see with what you are saying here is that I will have to use windows 8. Now if YOU tell me I can use the HP Z820 with 24 cores and have a lot more ram, and put Mavericks on it you might have me sold.
 
pastimage said:
Honestly, I should shut up now and walk away, but I cannot. The only real problem I see with what you are saying here is that I will have to use windows 8. Now if YOU tell me I can use the HP Z820 with 24 cores and have a lot more ram, and put Mavericks on it you might have me sold.
Click to expand...

That's Apple's call, and Apple's loss. HP won't care if you run Apple OSX on your Z820.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back