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

[luigi.lauro]

Yes, it's exactly that, you read me right.

Anandtech's Review: http://www.anandtech.com/show/6023/t...display-review

All of the rMBPs feature NVIDIA’s GeForce GT 650M equipped with 1GB of GDDR5. Apple went aggressive on the Kepler implementation and ships a full 384 core GK107 in the rMBP. The GPU clock is set at a very aggressive 900MHz with a 1254MHz memory clock.

Nvidia Specifications for Nvidia GT 650M:
http://www.geforce.com/hardware/note...specifications (notice the UP TO 900 MHZ)

Nvidia Specifications for Nvidia GTX 660M:
http://www.geforce.com/hardware/note...specifications (notice the 835 MHz)

Here they discuss and confirm this very topic:
http://forum.notebookreview.com/gami...-confused.html

Before:
GT 650M
GDDR3 version: Up to 850MHz
GDDR5 version: Up to 735MHz

Apple have actually GDDR5 version but with 900MHz cores. Looking at the homepage now, Nvidia have completely changed the specifications.

Apple`s GT 650M will beat the Asus GTX 660M (GTX 660M: GDDR5, Up to 835MHz)

Very interesting

Considering the resolution that drives the new mackbook, they would have commited suicide to include a GDDR3, or veru underclocked version of the GT650m. As it seems, they indeed slapped in a "GTX660m".

So Apple clocked the GDDR5 version of 650M so high to actually be better than a GTX 660M.

And also Anandtech confirmed that the new thermal design actually give for the first time in a mac notebook little to no throttling at all, even after going full-burn CPU+GPU for 20+ minutes:

The final experiment focuses on a hardly stressful game by today’s standards: Half Life 2 Episode Two. Once again, for approximately 20 minutes I ran our standard macbench timedemo test in HL2. I ran both systems at 1680 x 1050, without AA, but with all other quality settings maxed out. This is a bit more stressful than 20 minutes of actual gameplay since the timedemo runs renders all frames as quickly as possible rather than playing back the demo in real time. In the end it worked out to be nearly 40 consecutive runs of our benchmark.

Either the CPU or GPU (or both) have to be throttled back over time in order to stay within their thermal and power budgets. As a result, in the 2011 15-inch MacBook Pro, performance drops by over 20% over the course of 20 minutes of this test.

The MacBook Pro with Retina Display, on the other hand, remains relatively stable across all runs. While its performance definitely dips, the impact is around 5% off of peak.

So for the first time we actually have a 'GTX' top-of-the-class mobile GPUs in our macs.

Kudos to Apple, that's all I can say.

[theBostonian]

Originally Posted by luigi.lauro

So Apple clocked the GDDR5 version of 650M so high to actually be better than a GTX 660M.

So for the first time we actually have a 'GTX' top-of-the-class mobile GPUs in our macs.

Kudos to Apple, that's all I can say.

Am I right in thinking that the classic 2012 MBP has the "normal" version of the 650m, ie that only the rMBP has the superior GPU?

[thomaskc]

Originally Posted by theBostonian

Am I right in thinking that the classic 2012 MBP has the "normal" version of the 650m?

I am pretty sure they are exactly the same. And if not it would just be clock speeds, in which case you can just overclock the classic to run the same speeds. But I would be surprised if there is a difference at all.

[Sahee]

...so the 650M graphics card on the rMBR is basically a overclocked version.

[luigi.lauro]

Originally Posted by theBostonian

Am I right in thinking that the classic 2012 MBP has the "normal" version of the 650m, ie that only the rMBP has the superior GPU?

I don't honestly know, but for sure the thermal design of the classis 2012 MBP is the inferior thermal design of the 'old' unibody, which cause throttling of 20% of performance after 20 minutes on the old 2011 MBP.

So even if they would be the same card, with same clocks (which I don't know), after a few minutes of gaming, you would already have a 15% performance difference between the two, if the throttling is the same for the new 2012 MBP (which probably is).

If instead the clock of core and memory is actually less than 900/1254 as on the rMBP, then the gap is even wider.

And that's not even taking into account 512MB vs 1GB (even if I don't think make that much of a difference at the classic uMBP resolutions).

----------

Originally Posted by Sahee

...so the 650M graphics card on the rMBR is basically a overclocked version.

Yeah, and given the fact that the GTX 660M has inferior clocks for core and memory and no other advantage at all (they are EXACTLY the same card, only clocks differ), actually it's an OVERCLOCKED GTX 660M! :-)

So apple GT 650M actually is better than other notebooks GTX 660M, this is the reality, which is quite something, given apple's history about mobile GPUs.

This is for sure a first.

----------

Originally Posted by thomaskc

I am pretty sure they are exactly the same. And if not it would just be clock speeds, in which case you can just overclock the classic to run the same speeds. But I would be surprised if there is a difference at all.

Overclocking the GT 650M on the classic uMBP 2012 would probably give little to no advantage.

Given the thermal design is *EXACTLY* the same as 2011 MBP, you would probably have very bad throttling already at stock clock rates, and things wouldn't improve much overclocking that, you would still be severely limited by the thermal design throttling you back.

Or maybe I'm wrong and given the Ivy Bridge + 28nm Kepler advantage, the throttling on uMBP 2012 is not that bad as on the 2011 MBP, despite having the same thermal design.

We don't honestly know.

[grahamnp]

Deleted.

[TheMacBookPro]

So it's actually better than my Alienware laptop for gaming? Pretty cool

[luigi.lauro]

Originally Posted by TheMacBookPro

So it's actually better than my Alienware laptop for gaming? Pretty cool

It depends on which Alienware, there are alienware with 18'' displays and *DUAL* GTX 680M that run around the rMBP in circles :P

But they also have very different battery duration/size/weight/price (a maxed out alienware is 12 lbs, has the size of a backpack, and the price is just short of 8k $...) ;-)

Which model is yours, out of curiosity?

rMBP is better than M11x/M14x, but worse than M17x/M18x, performance-wise.

Unluckily alienware does not do a M15x which would be comparable to the rMBP, but I guess they would be really close performance-wise.

[thomaskc]

Overclocking the GT 650M on the classic uMBP 2012 would probably give little to no advantage.

Given the thermal design is *EXACTLY* the same as 2011 MBP, you would probably have very bad throttling already at stock clock rates, and things wouldn't improve much overclocking that, you would still be severely limited by the thermal design throttling you back.

Or maybe I'm wrong and given the Ivy Bridge + 28nm Kepler advantage, the throttling on uMBP 2012 is not that bad as on the 2011 MBP, despite having the same thermal design.

We don't honestly know.

I agree, we don't really know, but I don't actually think it is a problem, basically because of the points you already mentioned. The 2011 and 2012 versions, shell and fans aside are actually not THAT similar, and the new fans in the rmbp compared to the old fans are not really that effective either, they are how ever probably a lot quieter, but I doubt they cool much more when all comes to all.

How ever, I will recieve my non-retina beafed up MBP2012 late this week or early next week, and then I can let you know

[TheMacBookPro]

Originally Posted by luigi.lauro

It depends on which Alienware, there are alienware with 18'' displays and *DUAL* GTX 680M that run around the rMBP in circles :P

But they also have very different battery duration/size/weight ;-)

Which model is yours, out of curiosity?

rMBP is better than M11x/M14x, but worse than M17x/M18x, performance-wise.

Unluckily alienware does not do a M15x which would be comparable to the rMBP, but I guess they would be really close performance-wise.

I had an M11x and now have an M14x.

It's a revision 1 so it has the Core i7-2860QM, GT555M, 8GB RAM, 750GB 7200rpm HDD.
The r2 has the GT 650M like the Pros so it should not be that far off from the rMBP's performance. A little worse no doubt but the lower resolution screen means games should run smoother when comparing native resolutions.

I'll still be doing my gaming on that thing though since the keyboard is a lot more tactile and there's obviously a lot more storage space on that too.

They used to make an M15x but that was sized more like a 16-17" laptop so it still wouldn't be all that comparable with the 15" MacBooks either.

Originally Posted by luigi.lauro

Overclocking the GT 650M on the classic uMBP 2012 would probably give little to no advantage.

Given the thermal design is *EXACTLY* the same as 2011 MBP, you would probably have very bad throttling already at stock clock rates, and things wouldn't improve much overclocking that, you would still be severely limited by the thermal design throttling you back.

Overheating is easily solved by using a cheap external laptop cooler

[boto]

Anyone with Windows 7 that can confirm this through Nvidia drivers or using EVGA application? Just want to confirm if these are default settings on the Windows side as well.

[luigi.lauro]

Originally Posted by boto

Anyone with Windows 7 that can confirm this through Nvidia drivers or using EVGA application? Just want to confirm if these are default settings on the Windows side as well.

Anandtech confirmed this by using exactly this method, checking on windows side, and given anandtech reliability I doubt they did it wrong.

Anyways confirmations are always good, and would be good if clocks on the non retina are the same!

[rays09]

and then what about non retina? anyone know?

[luigi.lauro]

Originally Posted by rays09

and then what about non retina? anyone know

No one knows yet, but we hope that soon someone with a non retina MBP will test it out on Windows 7 and confirm the clocks on the uMBP.

[Moriarty]

Interesting that NVidia dos now list 900 MHz and 80GB/sec max memory bandwidth. So GPU-Z probably tells the truth. FWIW there is another thread where the GPU-Z of the non-retina has been posted and it has lower clocks.

So the rMBP is faster at games etc than the regular one at the same resolutions. And it's much thinner. Nice engineering work, Apple!

[luigi.lauro]

Originally Posted by Moriarty

Interesting that NVidia dos now list 900 MHz and 80GB/sec max memory bandwidth. So GPU-Z probably tells the truth. FWIW there is another thread where the GPU-Z of the non-retina has been posted and it has lower clocks.

So the rMBP is faster at games etc than the regular one at the same resolutions. And it's much thinner. Nice engineering work, Apple!

If what I see is true, the non retina MBP is clocked much lower, at a 810/1000 (core/memory), much lower than the retina MBP at 900/1254.

[arctic]

Interesting. Can't wait for my Retina to arrive. But according to this comparison from Macworld, the cMBP seem to have a slight edge in frame rates.

[doh123]

don't either of these up their clock at times... when needed? I thought I read the 660s would over-clock themselves if they could... but I haven't followed GPUs closely for several years.

[ugp]

I am very satifised with the performance on my 15" MBP. This CPU/GPU Combo is amazing.

[prfrma]

Hmmmmm....

[leman]

Originally Posted by prfrma

Hmmmmm....

Its the SSD kicking in. Nothing to do with CPU or GPU performance.

[prfrma]

Originally Posted by leman

Its the SSD kicking in. Nothing to do with CPU or GPU performance.

Point being, the benchmarks offer little control. No info on systems used, resolutions tested at, test duration/repetition etc.

[luigi.lauro]

Originally Posted by leman

Its the SSD kicking in. Nothing to do with CPU or GPU performance.

Definitely, yeah.

[prfrma]

Basically, does anyone know if this still happens to the uMBP?

[leman]

Originally Posted by prfrma

Point being, the benchmarks offer little control. No info on systems used, resolutions tested at, test duration/repetition etc.

I agree that we need more benchmarks, including more and different models.