Nvidia GT 650M on rMBP actually better than Nvidia GTX 660M!

[ivoruest]

How severe is the throttling, did you check the clocks?

I'm puzzled that a little CPU throttling can cause as much as a 20% performance difference in portal. Is portal that cpu-limited at that resolution? Seems strange.

Anyways thanks for sharing

Throttling is not severe but it does show or at least I can see it. When playing GTAIV or doing benchmarks it gets laggy after a long period but still I should say it doesn't feel like a 20%, I get the feeling it just turns Turbo Boost off to cool down a bit.

Temperatures hoover around 90° on this machine when on full load. Which makes me wonder how come there is people pushing their GPU's on full load and Overclocking inclusive and reporting temperatures lower to 80°??

 

[Centsy]

driving more pixels on the screen, even if running at a lower res, might be slowing down the rMBP.

That doesn't have anything to do with it. The resolution is the amount of pixels that are generated by the GPU. Formatting that for the screen is trivial.

So same resolution = same performance on both machines. Differences in the benchmarks could have to do with a variety of factors, such as the slightly different version of Lion that the rMBPs have. GPU performance is probably the same on both machines.

 

[magbarn]

Retina Macbook Pro GPU Clocks: 900 MHz Core/ 1254 MHz Memory
Non Retina Macbook Pro GPU Clocks: 810 MHz Core/ 1000 MHz Memory.

End of Story.

Base Retina Macbook Pro Will be way faster in Games than higher end non Retina MBP. For the same price...
Retina runs cooler, quieter, and has better screen. What more da ya want?

You're conveniently forgetting the rMBP has to push much more pixels and the cMBP most likely can overclock to the same level.

 

[rayman18]

could it be possible to flash the bios of the gt 650m from the rmbp, into the cMBP?

 

[Exana]

could it be possible to flash the bios of the gt 650m from the rmbp, into the cMBP?

No. It's nearly impossible for pc notebook BIOS, thus with Apple EFI and not flashing tools...

 

[Centsy]

You're conveniently forgetting the rMBP has to push much more pixels and the cMBP most likely can overclock to the same level.

See my post above. A resolution of 800x600 = 480,000 pixels pushed from the GPU. Those are then transformed for your screen, but at that point it doesn't matter what the native resolution of the screen is.

 

[ugp]

Sorry, going off topic here:

You wouldn't, by any chance, have heard the 2011 MBPs fans at full blast?

They can be rather high pitched and distracting, even over the sound of a game or music. The main thing stopping me from bothering to upgrade is the difference in fan noise between the 2011s and 2012 non-retina MBP.

I did run a dev>null test in 8 different Terminal windows on an Ultimate unibody 2012 MBP in the Apple Store and the temperatures didn't get as high when I ran the same test on my early 2011... However, because the store was full of people and playing music I couldn't tell if the fans were audible or not.

I have not. But the sound was not bad at all and I was just sitting watching TV while encoding a video. Was not distracting at all. I really didn't even take notice to the sound at all.

 

[leman]

That's funny since MacWorld found it opposite during their Portal 2 test.

These MacWorld benchmarks are a joke. They don't even say which resolution they were testing. Anandtech reports 55 fps for the rMBP at 1680x1050, while MacWorld numbers are off the charts. Were they running 1280 by 800 or what?


Using a seven year old engine with the settings that hardly push the GPU are not what I would call a competent review. I want to see come benchmarks at high resolutions on shader-intensive games, in Windows, compared to other Laptops with similar hardware. There can be tons of reasons why rMBP ends up slower there. But the GPU performance only becomes apparent when you make sure that the GPU is the bottleneck - which in this benchmark, it clearly was not.

 

[Icy1007]

Sorry, going off topic here:

You wouldn't, by any chance, have heard the 2011 MBPs fans at full blast?

They can be rather high pitched and distracting, even over the sound of a game or music. The main thing stopping me from bothering to upgrade is the difference in fan noise between the 2011s and 2012 non-retina MBP.

I did run a dev>null test in 8 different Terminal windows on an Ultimate unibody 2012 MBP in the Apple Store and the temperatures didn't get as high when I ran the same test on my early 2011... However, because the store was full of people and playing music I couldn't tell if the fans were audible or not.

The fans on my 2011 MBP are definitely audible when running at 6200 RPM, but I would hardly call them high pitched or distracting. Any music or game I have playing easily overpowers the fan noise.

 

[doh123]

These MacWorld benchmarks are a joke. They don't even say which resolution they were testing. Anandtech reports 55 fps for the rMBP at 1680x1050, while MacWorld numbers are off the charts. Were they running 1280 by 800 or what?


Using a seven year old engine with the settings that hardly push the GPU are not what I would call a competent review. I want to see come benchmarks at high resolutions on shader-intensive games, in Windows, compared to other Laptops with similar hardware. There can be tons of reasons why rMBP ends up slower there. But the GPU performance only becomes apparent when you make sure that the GPU is the bottleneck - which in this benchmark, it clearly was not.

MacWorld has always been as up-to-date in gaming as Apple themselves... 5+ years behind is pretty normal

 

[luigi.lauro]

MacWorld has always been as up-to-date in gaming as Apple themselves... 5+ years behind is pretty normal

But since Apple for the first time catched on and now rMBP is actually better and faster than the Alienware (m14x) themselves thanks to this special 'GTX 660M', we can hope also MacWorld will catch up ;-)

 

[doh123]

But since Apple for the first time catched on and now rMBP is actually better and faster than the Alienware (m14x) themselves thanks to this special 'GTX 660M', we can hope also MacWorld will catch up ;-)

I don't think they had any gaming in mind when doing this though... probably dumb luck.

 

[leman]

I don't think they had any gaming in mind when doing this though... probably dumb luck.

Then don't quote their 'benchmarks' as argument

BTW, on notebookcheck.net, the 650M in the MBP is confidently outperforming all other 650M's.

 

[asdad123]

This card is pretty insane. Im benchmarking the computer at first like I do with any other laptop I get just to see how high they can go, and Im getting pretty good OCs right now, and the temps arent even increasing.

GPU temps max at about 77C each run, and CPU temps are around 100C just like at stock.

Im currently overclocked to 1000/1354 up from 900/1254, and have no artifacts yet. Best 3Dmark11 score so far is 2646. Ill update with my final results when I finish.

According to NotebookCheck, a GTX 660m with a 3720qm gets a score of 2727. Im very impressed with the gt650m in this computer.

edit: Final Clock was 1035/1729 with a 3dmark11 result of 2860

 

[thomaskc]

I am still pretty sure the reason why the retina gpu is clicked higher is in an attempt to deliver the same user experience on the new more demanding retina setup within osx as people have been used to from previous series of MacBook pros. I highly highly doubt there is any difference what so ever on the the two series other than a preset video bios clock setting. Meaning that the non retina will most likely perform just as well if overclocks the same.

Driving 2880x1800 on a laptop is no joke, and since osx utilises gpu accelleration a great deal ( and by the sound of it mountain lion even more) it makes total sense that the gpu on the retina is clocked slightly higher to try and compensate and to have an easy place to increase performance in the ui.

 

[luigi.lauro]

As you reported on the other thread: https://forums.macrumors.com/threads/1394602/

Core at 1035 (+135) and Memory at 1729 (+475) with just 1 degree more in temperature is a pretty crazy over clocking!

And moreover the card can probably go further, since you were stopped by the software limits of MSI Afterburner (+135 max clock).

Pretty unbelievable, I wouldn't be shocked if with this over clock we even go past GTX 670M performance!

Keep pushing it!

 

[thomaskc]

As you reported on the other thread: https://forums.macrumors.com/threads/1394602/

Core at 1035 (+135) and Memory at 1729 (+475) with just 1 degree more in temperature is a pretty crazy over clocking!

And moreover the card can probably go further, since you were stopped by the software limits of MSI Afterburner (+135 max clock).

Pretty unbelievable, I wouldn't be shocked if with this over clock we even go past GTX 670M performance!

Keep pushing it!

I'm sorry if I missed something but why are you using MSI? EVGA does not have limit in the same way to my experience, but maybe that doesn't see the gpu correctly at this stage?

 

[leman]

I am still pretty sure the reason why the retina gpu is clicked higher is in an attempt to deliver the same user experience on the new more demanding retina setup within osx as people have been used to from previous series of MacBook pros. I highly highly doubt there is any difference what so ever on the the two series other than a preset video bios clock setting. Meaning that the non retina will most likely perform just as well if overclocks the same.

I highly doubt that the 650m is even driving the display most of the time. If it did, the battery life would be in the gutter. The majority of work is done by the HD 4000. Much of the UI rendering is done on the CPU (text is a prime example, although it is possible to move this to the GPU maybe ML does this?), the GPU is mostly responsible for simpler shapes and image compositing. The HD 4000 has a benchmarked fillrate of over 2000 Mpixels/sec . The full 2880x1800 display is 'only' 5 Mpixels. Thus, the HD 4000 could redraw the full screen at around 200 fps with relative ease. Given that real rendering is much more complex, but that window contents rarely needs to be refreshed completely (usually only small details change, like animations or when user initiates an UI action), the HD 4000 should be competent enough to drive the retina UI at around 30 fps - incidentally, this is what Anandtech shows in their review. The performance will drop with complex UIs which redraw themselves completely multiple times per second - but this again is a problem of bad programming

 

[asdad123]

I'm sorry if I missed something but why are you using MSI? EVGA does not have limit in the same way to my experience, but maybe that doesn't see the gpu correctly at this stage?

I tried both MSI Afterburner and EVGA Precision and both have the core clock limited at +135

 

[Centsy]

I did a benchmark in 3DMark06 and with a +135/+300 overclock plus throttlestop, I got 17189.

For reference:
Average for 650m: 13676.3
Average for 660m: 16025.0
Average for 670m: 19650.3

 

[cirus]

I highly doubt that the 650m is even driving the display most of the time. If it did, the battery life would be in the gutter. The majority of work is done by the HD 4000. Much of the UI rendering is done on the CPU (text is a prime example, although it is possible to move this to the GPU maybe ML does this?), the GPU is mostly responsible for simpler shapes and image compositing. The HD 4000 has a benchmarked fillrate of over 2000 Mpixels/sec . The full 2880x1800 display is 'only' 5 Mpixels. Thus, the HD 4000 could redraw the full screen at around 200 fps with relative ease. Given that real rendering is much more complex, but that window contents rarely needs to be refreshed completely (usually only small details change, like animations or when user initiates an UI action), the HD 4000 should be competent enough to drive the retina UI at around 30 fps - incidentally, this is what Anandtech shows in their review. The performance will drop with complex UIs which redraw themselves completely multiple times per second - but this again is a problem of bad programming

The laptop has a 95 watt hour battery. That is a huge battery. Considering most laptops not running switchable graphics get about 2.5 to 4 hours on a 48-55 watt hour battery and the rmbp gets about 6-7, its battery life is pretty much "in the gutter" in terms of hours/watt. Pretty much any laptop will last 7 hours with a 95 watt hour battery (except 15 pound gaming laptops), I think about 12-15 watts/hour is average.

 

[Aftershocker]

This card is pretty insane. Im benchmarking the computer at first like I do with any other laptop I get just to see how high they can go, and Im getting pretty good OCs right now, and the temps arent even increasing.

GPU temps max at about 77C each run, and CPU temps are around 100C just like at stock.

Im currently overclocked to 1000/1354 up from 900/1254, and have no artifacts yet. Best 3Dmark11 score so far is 2646. Ill update with my final results when I finish.

According to NotebookCheck, a GTX 660m with a 3720qm gets a score of 2727. Im very impressed with the gt650m in this computer.

edit: Final Clock was 1035/1729 with a 3dmark11 result of 2860

Has anybody tried "lubbo's fan control", or any other dedicated fan control program to see if it works with the retina macbook? Could be useful to try and bring down temps.

 

[golf1410]

Bump for fun.

 

[unwinded]

That's funny since MacWorld found it opposite during their Portal 2 test.

The fans full blast on the Non-Retina Model are quiet. I can barely hear them. Maybe if I were is a completely silent room you would notice them.

The only difference between the Retina and Non-Retina is the SSD drive. But once you put a SSD in the Non-Retina they are identical as far as speed or pretty close.

As a few have noticed, some serious scaling is to blame. I think somebody else confirmed Portal 2 is being rendered at 2880x1800 no matter what resolution you choose (performance is the same in lower resolutions so that's likely the case). On the rMBP I mean. Then 2880x1800 is scaled down to whatever your desktop environment is. Which means if it's 1920x1200 (equivalent) that's what it is scaled to. Then it's scaled to whatever resolution you choose in the game. Or something like that.

tl;dr Portal 2 needs a fix or something to run properly and it's wasting resources scaling like 4x. Even with full AA it's pixelated.

I'm afraid of OCing this. I OC'd my Asus 17" which had a decent cooling system and was huge. It would crash the driver and shut down a lot despite no artifacts and good temps. Was common for that model and it's successors with similar hardware. This mac gets pretty hot and it already eclipses my OC'd 460m in the Asus (which was better than a stock 560m when OC'd despite the problems). I just hope stock doesn't fry it over time.

 

[leman]

You're conveniently forgetting the rMBP has to push much more pixels and the cMBP most likely can overclock to the same level.

You don't have to run it at the native resolution, you know?