How is the 1st generation rMBP compared with the 2014 one?

[hajime]

Hello, I got the rMBP 15" 2012 when it came out. It has 2.6GHz i7, 16GB 1600 MHz DDR3, Intel HD Graphics 4000 512MB with Nvidia GeForce GT 650M 512MB, 751 GB SSD. It was great but for the past few weeks, it has image retention problem.

How is it compared with the latest top-of-the-line rMBP in terms of performance on scientific calculations, simulations? For 3D CAD such as 3D Max, SolidWorks, does the latest top of the line system runs noticeably faster than the one I got?

 

[yjchua95]

Hello, I got the rMBP 15" 2012 when it came out. It has 2.6GHz i7, 16GB 1600 MHz DDR3, Intel HD Graphics 4000 512MB with Nvidia GeForce GT 650M 512MB, 751 GB SSD. It was great but for the past few weeks, it has image retention problem.

How is it compared with the latest top-of-the-line rMBP in terms of performance on scientific calculations, simulations? For 3D CAD such as 3D Max, SolidWorks, does the latest top of the line system runs noticeably faster than the one I got?

It won't run noticeably faster. The 2.6GHz i7-3720QM performs similarly to the 2.8GHz i7-4980HQ (mid-2014 15" rMBP).

http://browser.primatelabs.com/geekbench3/compare/687101?baseline=469593

However, in OpenCL tasks, the Iris Pro will perform better than the Intel HD 4000 and the GT650M (and even the GT750M).

 

[leman]

If your software can take advantage of the Haswell's new AVX2 instructions (or you can recompile it with AVX2 support) , then you might see substantial improvement for some numeric computations. Also, the new Intel iGPU is a truly a computation monster, packing more power than 650M or 750M and having better performance with computations that require more random memory access. But this is all circumstantial — you will only benefit from these things if your work is sensible to these improvements, and usually only if you write the software yourself.

As to the CAD, I can't comment from experience, but I don't think that you will see much improvement here. Both won't be able to deal with complex projects, and for simple/medium complexity, it does not really matter anyway.

 

[w00t951]

Depending on the workload, you'll see an increase from 7-17% in CPU constrained workloads, possibly more if your workload leverages Haswell specific instruction sets such as AVX2.

GPU performance is significantly higher, especially for the integrated GPU.

However, the battery life is what I enjoy most. My friends find maybe ~5 hours on battery for light work in browser/word processing applications. My 2013 i7-4850HQ model gets around 8 hours on the same usage scenarios, with brightness at 75% and resolution scaled to 1920x1200 (maximum).

If you've already got a rMBP and don't use the battery that much, upgrading is worthless. I'd wait for Broadwell at the least - it's supposed to be significantly more power efficient than Haswell, and it'll ship with a much more powerful Intel integrated GPU and NVIDIA Maxwell GPU.

Keep in mind, however - Skylake is the generation after Broadwell. It's the first performance oriented processor generation since Sandy Bridge, and we'll be seeing double digit increases in performance over Broadwell. Unless you've got some issues with battery life, I'd even wait for Skylake.

 

[hajime]

What about performance when using MATLAB/Simulink?

My current rMBP 15" is from work. I am trying to decide whether or not it is worth to ask my employer to let me have it when I quit. Note that the machine started to have image retention problem. A local Apple store said that if it is not software issue, it would cost me 1/4 the price of the latest rMBP to have the screen replaced. He said that Apple won't replace it for free.

----------

It won't run noticeably faster. The 2.6GHz i7-3720QM performs similarly to the 2.8GHz i7-4980HQ (mid-2014 15" rMBP).

http://browser.primatelabs.com/geekbench3/compare/687101?baseline=469593

However, in OpenCL tasks, the Iris Pro will perform better than the Intel HD 4000 and the GT650M (and even the GT750M).

You mean the internal Iris PRO GPU is faster than the discrete GT650M?

 

[w00t951]

What about performance when using MATLAB/Simulink?

----------

You mean the internal Iris PRO GPU is faster than the discrete GT650M?

MATLAB appears to support AVX2, but I'm not sure how much of a performance increase that'll buy you. You'll have to do your own research on that.

As for the GPU question, check your specific usage scenario. In general, compute oriented cards will be worse at rendering images than graphics oriented cards, and vice versa.

For example, the NVIDIA GTX Titan Black and the GTX 780 Ti have the same number of cores, but 780 Ti has some compute cores disabled and runs at a higher core frequency that gives it the edge in image performance. The Titan, however, is far faster at compute tasks.

 

[SarcasticJoe]

For example, the NVIDIA GTX Titan Black and the GTX 780 Ti have the same number of cores, but 780 Ti has some compute cores disabled and runs at a higher core frequency that gives it the edge in image performance. The Titan, however, is far faster at compute tasks.

The compute performance edge is also partly caused by drivers intentionally hampering compute performance on consumer cards. This is done so that people who use their cards for GPGPU tasks will have to buy the more expensive versions of essentially the same cards. Remember reading years ago about how people were able to get significant performance gains in compute tasks from just putting new firmware on consumer cards, but I think the card makers decided to put a stop to this by making otherwise useless modifications into the compute cards so that their firmware wouldn't work on consumer cards.

 

[udflyer]

Well, I work with both.

Personally - Just bough $1999 base 2014 rMBP

Work - using original rMBP base 15" model.


The only thing I notice.....and I do notice , is the SSD drive speed.

The newer Macs are twice as fast and it now sucks for me to work on my work Mac.

If that is worth the money is your call.

 

[CaffeinatedNoms]

What about performance when using MATLAB/Simulink?

My current rMBP 15" is from work. I am trying to decide whether or not it is worth to ask my employer to let me have it when I quit. Note that the machine started to have image retention problem. A local Apple store said that if it is not software issue, it would cost me 1/4 the price of the latest rMBP to have the screen replaced. He said that Apple won't replace it for free.

----------



You mean the internal Iris PRO GPU is faster than the discrete GT650M?

Actually you'd be surprised - Quite a few OpenCL and quite a lot of OpenGL actions will perform better on the Iris Pro than on a better-than-equivalent nVidia card. This is, however, down to optimisation: nVidia chose to focus on CUDA technologies, which MacOS does not use, rather then the OpenCL/GL platform that Intel and AMD support - the platform that Apple specifically uses 😛

I should point out though, that the 650 upwards will still beat Iris Pro for gaming performance.

 

[w00t951]

The compute performance edge is also partly caused by drivers intentionally hampering compute performance on consumer cards. This is done so that people who use their cards for GPGPU tasks will have to buy the more expensive versions of essentially the same cards. Remember reading years ago about how people were able to get significant performance gains in compute tasks from just putting new firmware on consumer cards, but I think the card makers decided to put a stop to this by making otherwise useless modifications into the compute cards so that their firmware wouldn't work on consumer cards.

Yeah, but crimping SMX's on Kepler cards is the main source of the compute performance loss, isn't it?

Anyway, if you're required to do serious GPGPU stuff for work, they should provide you with a workstation-class desktop/laptop. They're freaking expensive compared to their consumer counterparts. And to my knowledge, Apple only makes one true workstation, the Mac Pro.

The MacBook Pro is a consumer device that carries a consumer class dedicated GPU inside.

 

[hajime]

The only thing I notice.....and I do notice , is the SSD drive speed.

The newer Macs are twice as fast and it now sucks for me to work on my work Mac.

If that is worth the money is your call.

Thanks for the experience. How come it is twice as fast? This seems to be useful for those in video editing.

 

[Dilster3k]

For the price to performance ratio, it's not worth it at all. You won't notice much of a performance bump in day-to-day usage. Just hold around for another generation or two.

 

[venom600]

Thanks for the experience. How come it is twice as fast? This seems to be useful for those in video editing.

The original rMBP used a SATAIII connection, so it's limited to 6GBPS. The new ones use direct PCI-E connections, so they have much higher peak speeds.

 

[yjchua95]

My current rMBP 15" is from work. I am trying to decide whether or not it is worth to ask my employer to let me have
You mean the internal Iris PRO GPU is faster than the discrete GT650M?

In OpenCL tasks, yes. I even disable my GT 750M in my late-2013 15" when doing OpenCL jobs.

 

[hajime]

The original rMBP used a SATAIII connection, so it's limited to 6GBPS. The new ones use direct PCI-E connections, so they have much higher peak speeds.

In theory they do. In practice, they do as well?

----------

In OpenCL tasks, yes. I even disable my GT 750M in my late-2013 15" when doing OpenCL jobs.

So, for OpenCL (do you mean OpenGL?), it is faster and more energy saving to use the Iris PRO GPU. Am I right? I read somewhere that the Iris PRO GPU is faster than the discrete Nvidia GPU. That is what got me confused...

 

[leman]

So, for OpenCL (do you mean OpenGL?), it is faster and more energy saving to use the Iris PRO GPU. Am I right? I read somewhere that the Iris PRO GPU is faster than the discrete Nvidia GPU. That is what got me confused...

OpenCL is a general-purpose parallel computation library (not to be confused with OpenGL). The Iris Pro IS faster than 650M or 750M, if you look on its peak theoretical computation ability. The reality is much more complicated. As a rule of thumb, Iris Pro will fare better for complex computation tasks with random memory accesses (i.e., there is a lot of work to be done for each work item, and memory access pattern is less regular), and the Nvidia GPU is better suited for more trivial computations which access the memory in more or less sequential fashion (3D graphics or most practical cases of video/image processing). In games, 650M is up to 50% faster.

At any rate, if its the first time you hear about OpenCL, then the 1st gen rMBP will be fast enough for you. You can update to Skylake or its successor later, by which time the software should also catch up with the new CPU and GPU capabilities. That should give you a healthy processing boost. With 1st gen retina vs. current ten one, the real-world difference in existing software is 10-15% at most (and that is quite optimistic), so not really worth the money.

 

[yjchua95]

So, for OpenCL (do you mean OpenGL?), it is faster and more energy saving to use the Iris PRO GPU. Am I right? I read somewhere that the Iris PRO GPU is faster than the discrete Nvidia GPU. That is what got me confused...

OpenCL and OpenGL aren't the same.

OpenCL is a GPGPU API that enables programmers to use the GPU as a co-processor for massively parallelizable tasks such as physics and image/signal processing.

OpenGL is only 'important' if games you want to play don't support Direct3D. OpenCL is not used by many games since most people would rather put their GPU's compute power into graphics than splitting it with stuff like physics.

Photoshop CS6 and CC is an example of an OpenCL software.

FlightGear is an example of an OpenGL game.

 

[brdeveloper]

Actually you'd be surprised - Quite a few OpenCL and quite a lot of OpenGL actions will perform better on the Iris Pro than on a better-than-equivalent nVidia card. This is, however, down to optimisation: nVidia chose to focus on CUDA technologies, which MacOS does not use, rather then the OpenCL/GL platform that Intel and AMD support - the platform that Apple specifically uses 😛

I should point out though, that the 650 upwards will still beat Iris Pro for gaming performance.

OSX doesn't use, although CUDA is supported by third-party software. Actually I'm more interested on CUDA since its framework provides better tools than xCode in relation to OpenCL. Also, nVidia provides better drivers than the ones bundled with OSX. I installed a "hacked" driver (actually a user from the MBP forum edited the list of supported GPUs inside the downloaded package) which performed better than the standard Mavericks driver on my MacMini with a 320M.

That is... as a developer I feel like I have more options with an nVidia card. Gamers will feel the same.

 

[UBS28]

OpenCL is a general-purpose parallel computation library (not to be confused with OpenGL). The Iris Pro IS faster than 650M or 750M, if you look on its peak theoretical computation ability. The reality is much more complicated. As a rule of thumb, Iris Pro will fare better for complex computation tasks with random memory accesses (i.e., there is a lot of work to be done for each work item, and memory access pattern is less regular), and the Nvidia GPU is better suited for more trivial computations which access the memory in more or less sequential fashion (3D graphics or most practical cases of video/image processing). In games, 650M is up to 50% faster.

At any rate, if its the first time you hear about OpenCL, then the 1st gen rMBP will be fast enough for you. You can update to Skylake or its successor later, by which time the software should also catch up with the new CPU and GPU capabilities. That should give you a healthy processing boost. With 1st gen retina vs. current ten one, the real-world difference in existing software is 10-15% at most (and that is quite optimistic), so not really worth the money.

It is not always faster than the 750m on OpenCl according to some benchmarks. But wouldn't it make more sense to use CUDA over OpenCL when using the 750m since NVIDIA drivers are optimised for CUDA? NVIDIA doesn't pay much attention to OpenCl.

 

[yjchua95]

It is not always faster than the 750m on OpenCl according to some benchmarks. But wouldn't it make more sense to use CUDA over OpenCL when using the 750m since NVIDIA drivers are optimised for CUDA? NVIDIA doesn't pay much attention to OpenCl.

From a performance and developer's standpoint, yes, it does make more sense to use CUDA.

But the problem is, some developers, like Adobe, more or less turned a blind eye on CUDA and moved to OpenCL since the days of CS6.

And not all software is optimised for CUDA.

And you can't choose whether you want to use CUDA or OpenCL. It's up to the developer of the software 😛