I've heard that Kepler will be a big update for laptops, but how significant is the update exactly? In which area does it improve on? (Performance, power usage?)
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Laptop Kepler graphics - how good is it exactly?
- Thread starter rekhyt
- Start date
- Sort by reaction score
In performance mostly.
Whatever they save on power usage they just up the clocks until they hit their old TDP specs. It usually is almost always the same.
There are no benchmarks with GDDR5 mobile Kepler out yet. They are all based on DDR3. I guess Apple can afford to put some GDDR5 in and clock them lower. A 650M with GDDR5 could be an over clocking beast if there is enough cooling headroom.
In the big desktop GPUs Kepler made a big Perf/Watt leap and is a little ahead of AMDs GCN now, while they trailed AMD for the last couple of years. Little is known about the 640-660M cards. They don't publich TDP specs. You need to wait and see in what kind of Notebooks they crop up to know what TDP they really have. Nvidia just switches around to often.
Usually 560M was a gaming GPU way too hot for anything but huge thick gaming notebooks. If a 660M is the same GPU grade it seems GDDR5 makes it a very hot GPU.
640M seems to be a direct 540M replacement.
650M comes in DDR3 flavors but can also come with GDDR5. Until there is a GDDR5 card with optimal clock settings I am not sure what that GPU is really capable of.
Also the whole AMD 28nm GPUs are still missing. 6770M used to be cooler than a 550M and only slightly hotter than a 540M. The only numbers we know of the 7770M puts it slightly ahead of the 650M but at only 30W TDP. Either the 650M is lower TDP than a 550M or the 7770M is still better and faster. I guess the former.
Unfortunately there is no data that really tells anything specific. You can only infer later on after most notebooks have been released an benched. Also Nivida likes to switch around numbers. AMD usually has more consistent naming.
Whatever they save on power usage they just up the clocks until they hit their old TDP specs. It usually is almost always the same.
There are no benchmarks with GDDR5 mobile Kepler out yet. They are all based on DDR3. I guess Apple can afford to put some GDDR5 in and clock them lower. A 650M with GDDR5 could be an over clocking beast if there is enough cooling headroom.
In the big desktop GPUs Kepler made a big Perf/Watt leap and is a little ahead of AMDs GCN now, while they trailed AMD for the last couple of years. Little is known about the 640-660M cards. They don't publich TDP specs. You need to wait and see in what kind of Notebooks they crop up to know what TDP they really have. Nvidia just switches around to often.
Usually 560M was a gaming GPU way too hot for anything but huge thick gaming notebooks. If a 660M is the same GPU grade it seems GDDR5 makes it a very hot GPU.
640M seems to be a direct 540M replacement.
650M comes in DDR3 flavors but can also come with GDDR5. Until there is a GDDR5 card with optimal clock settings I am not sure what that GPU is really capable of.
Also the whole AMD 28nm GPUs are still missing. 6770M used to be cooler than a 550M and only slightly hotter than a 540M. The only numbers we know of the 7770M puts it slightly ahead of the 650M but at only 30W TDP. Either the 650M is lower TDP than a 550M or the 7770M is still better and faster. I guess the former.
Unfortunately there is no data that really tells anything specific. You can only infer later on after most notebooks have been released an benched. Also Nivida likes to switch around numbers. AMD usually has more consistent naming.
AMD's 6770M (currently used in MBP) is believed to have a TDP of between 30W and 32W. Its replacement (the 7770M) has a TDP of 32W [source].
Nvidia's - Fermi based - GT 550M/555M has a TDP of between 30W and 35W [source], making it comparable (in TDP only) with the 6770M/7770M. The newer - Kepler based - GT 650M appears (and I'm dubious about these numbers) to have a TDP of 45W [source]. If this is truly the case, the new MBP is more likely to have the 32W GT 640M (despite the rumours of a GT 650M).
There are 'processing power' numbers listed in the tables I've linked; from which you can infer ballpark-performance. Direct comparison of GFLOP numbers is not possible as Fermi was a very GPGPU focused architecture, whereas Kepler is gaming orientated (with relativly poor GPGPU performance). AMD's new GCN architecture (7770M and others) is more balanced (in terms of it's focus) than Nvidia's Kepler. GCN appears to have slightly worse gaming performance than Kepler, but better GPGPU performance (particularly OpenCL, which Apple are pushing).
Which GPU is best (and by how much) is very subjective. Five years ago, gaming was insignificant on OSX, but that's not true today. If you're after gaming performance, the GT 640M (with GDDR5) will likely be on par (if not slightly better) than the present AMD 6770M, whilst the GT 650M would provide a large increase in performance. Conversely, if you're GPGPU focused, it seems the GT 640M will be a large step backwards from AMD's 6770M, and even the GT 650M will struggle to match the present MBP's GPGPU performance.
Most of these wikipedia TDP ratings are really just guestimates afaik.
I think the initial 6750M or was 35W with GDDR5 and the DDR3 6730 was 24W.
If the 640M really is as high as 32W it seems logical to assume the 650M at 45 with GDDR5 and the given clock speeds.
I would have bet on a 650M simple because the 640M really isn't enough of a trade up from the 6770M and a GDDR5 version with lowered clock speed while more expensive is probably the most efficient design.
I would have thought the 550 was higher with the 555 at 45W or so. The 540M probably at 25-28. If this was where the 640M really is the 650M could be in the <35W group and viable.
Maybe it will be a 640M with GDDR5 and some 600Mhz clock speed. We will see.
More importantly I think Apple should clean up their terrible switching technology and ask for a Mac compatible optimus driver. So much more useful as there really is little need for the dGPU in most use cases.
I think the initial 6750M or was 35W with GDDR5 and the DDR3 6730 was 24W.
If the 640M really is as high as 32W it seems logical to assume the 650M at 45 with GDDR5 and the given clock speeds.
I would have bet on a 650M simple because the 640M really isn't enough of a trade up from the 6770M and a GDDR5 version with lowered clock speed while more expensive is probably the most efficient design.
I would have thought the 550 was higher with the 555 at 45W or so. The 540M probably at 25-28. If this was where the 640M really is the 650M could be in the <35W group and viable.
Maybe it will be a 640M with GDDR5 and some 600Mhz clock speed. We will see.
More importantly I think Apple should clean up their terrible switching technology and ask for a Mac compatible optimus driver. So much more useful as there really is little need for the dGPU in most use cases.
Some Nvidia GT 650M (Kepler) benchmarks.
Edit: This is the DDR3 variant; hopefully Apple would use the GDDR5 variant.
Edit: This is the DDR3 variant; hopefully Apple would use the GDDR5 variant.
Last edited:
I wasn't aware of a fiasco with those two GPUs.
The 8600m was big trouble, but the 9600m and 330m were both OK mid-range GPUs with no real problems except for price-performance issues.
Wondering about this as well.
Aside from the lacklustre performance compared to PC discrete graphics (as expected, not really a criticism but an obvious statement, due to power consumption and heat), I don't remember hearing about any fiasco with the GT 330M.
Also had a MacBook Pro 15" with the 9600M and was not aware of any problems with that either.
http://www.game-debate.com/hardware/index.php?gid=832&graphics=GeForce GT 650M GDDR5
"Very Demanding games might be playable at high settings while less demanding games should be maxed out easily...."
http://www.game-debate.com/hardware...ics=GeForce+GT+650M+GDDR5&checkSubmit=#system requirements
go about 3/4ths the way down and you'll see a comparison with a Radeon HD 6770M 1gb. This card is 21% better
"Very Demanding games might be playable at high settings while less demanding games should be maxed out easily...."
http://www.game-debate.com/hardware...ics=GeForce+GT+650M+GDDR5&checkSubmit=#system requirements
go about 3/4ths the way down and you'll see a comparison with a Radeon HD 6770M 1gb. This card is 21% better
Where on this page are any links to actual benchmarks.
All I have seen to date is 650m DDR3 benchmarks. What notebooks even offer a GDDR5 version yet that aren't thick gaming notebooks and where are the benchmarks.
I don't see any on the links you posted. Maybe one needs an account.
I found one well two.
DDR3 2GB in an 11.6" clevo
http://www.notebookcheck.com/Test-Schenker-XMG-A102-Clevo-W110ER-Subnotebook.74505.0.html
GDDR5 1GB in an 15" clevo 47mm thick not exactly mobile
http://www.notebookcheck.com/Test-Schenker-XMG-A502-Clevo-W150ER-Notebook.74449.0.html
In fact if you look closely the DDR3 version is actually faster. Possible because the GDDR5 memory controller produces so much heat that the Turbo is less effective and the GPU just doesn't seem to need the bandwidth.
Benchmarks are very close and almost identical. Absolutely no sign of them being any faster than a 7770M. There isn't much on notebookcheck except one 3dmark11 bench and that seems rather close either better or equal to the 650M. No sign of 21% difference.
All I have seen to date is 650m DDR3 benchmarks. What notebooks even offer a GDDR5 version yet that aren't thick gaming notebooks and where are the benchmarks.
I don't see any on the links you posted. Maybe one needs an account.
I found one well two.
DDR3 2GB in an 11.6" clevo
http://www.notebookcheck.com/Test-Schenker-XMG-A102-Clevo-W110ER-Subnotebook.74505.0.html
GDDR5 1GB in an 15" clevo 47mm thick not exactly mobile
http://www.notebookcheck.com/Test-Schenker-XMG-A502-Clevo-W150ER-Notebook.74449.0.html
In fact if you look closely the DDR3 version is actually faster. Possible because the GDDR5 memory controller produces so much heat that the Turbo is less effective and the GPU just doesn't seem to need the bandwidth.
Benchmarks are very close and almost identical. Absolutely no sign of them being any faster than a 7770M. There isn't much on notebookcheck except one 3dmark11 bench and that seems rather close either better or equal to the 650M. No sign of 21% difference.
Last edited: