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.
Would you buy a rMBP without a dedicated GPU?
- Thread starter johngwheeler
- Start date
- Sort by reaction score
Indeed it is!
Correct! Neither it is "OMG Battlefield 3 should be X-frames faster" machine.
And why it should be, when none of benched games (except Grid 2, which is faster on Intel hardware, with added special effects, but that's a glitch, 'cause it's not in line with "It's simply not capable enough") are optimized with Intel's graphics. Why? Because the most powerful up until recently HD 4000 is 40% the performance of HD 5200. Why would they bother?
Last edited:
Except BF3 performance and FCPX performance are related. The game benchmarks indicate the sort of performance you'll see in pro apps.
Do you think when it comes to Final Cut Pro X the Iris Pro is suddenly going to turn around and become an all star performer?
The same sort of apps that are leading Apple to put two high end GPUs in the Mac Pro.
Are you seriously suggesting that there are no pro apps which max out the 650m?
My point is the "Pro" label might mean something for graphic designers, musicians, gamers, or people who make a living with their notebook, but for me it doesn't mean anything, and more than ever after the Macbook died... So "Pro" compared to what? To an Air? So we have the "Air" and "Pro" users... Doesn't make sense for me...
The future of dGPU is likely the same as the cMBP one. Sooner or later, it will disappear... The iGPU is getting more and more powerful and the iGPU vs dGPU will likely have a happy end if Intel acquires nVidia...
I am likely getting a 13" model because I think it is the model which will mostly benefit from haswell architecture...
But Grid 2 and 3DMark are not? (to FCPX, obviously)
Is this that app that lists OpenCl compatible gpu as a requirement? If that's the case- nope, I don't see how it will help.
It isn't a regression - its a huge improvement over the HD4000, which is what most rMBPs today run most of the time. That's a day to day kind of improvement.
Its only a regression if you're intent on heavy gaming on a non optimized gaming platform. That's just not what most people use MBPs for.
Except that HD 4000 is enough (Mavericks proves this), so any "improvement" is not going to be immediately noticeable to users.
More heat and more power consumed (at higher TDP packages) would be noticeable, though.
And like I have written multiple times now, it's not just gaming. Anything that uses OpenGL (read: Photoshop, AutoCAD, Maya) that runs at Retina resolution will take a performance hit.
Of course. Because you quoted it out of context.
Number of processors = I meant shader processing units.
Number of execution units = could be anything. Hell, some GPUs have dedicated h.264 decoding and encoding units. Those don't do anything to benchmarks... though they do allow you to offload video decoding/encoding from CPU.
And you need to read more:
http://www.beyond3d.com/content/articles/38/1
1 game. Hardly the norm. And the latter version of the same engine (Crysis 3) does show a performance gap.
Those special effects can simply be turned off from the settings and the comparison is still fair.
It shows off what the architecture is capable of, sure... but Iris Pro is only available on laptops! Oh, and it's still no match to the fastest discrete GPUs on the market, which enthusiasts will most likely run Grid 2 on! So to have those effects, you'd have to give up super high resolution, antialiasing, and also some of the higher end settings.
So then... how do they "show off" what the hardware is capable of again?
How game developers work or how the drivers work don't necessarily underline the number one issue that I have been parroting for a while now: Iris Pro simply does not have the architecture to catch up to 650M in raw 3D performance (hardware) at high resolution.
And again, the Retina MacBook does have a high resolution screen.
Granted, you can argue that Intel may be able to optimize the drivers to "mask" those performance differences, but nVidia can also use the same "techniques" to "improve their drivers' performance", and the gap is still there.
It's silly if I have to run my 3D simulator at less than half the native resolution to match last year's computer in performance.
And synthetics reflect actual performance? Hmm... nope.
If you want big numbers, sure. But there are people who want real results.
But far more than regular HD 4000 or HD 4600 part, and likely would increase power consumption or heat under day to day desktop use when the dGPU does not kick in.
And when the dGPU does kick in, then it still achieves a higher level of performance. Even if just by 10%, it's still better performance.
Why not... AMD acquired ATI...
Cold you imagine an nVidia dGPU integrated with intel CPU...
Last edited:
Intel's execution units- fancy name for shaders:
Ah, thanks for the good read. A little bit old though.
From Wiki article:
And Imagination Tech's take on the matter, regarding PowerVR Series5XT:
http://withimagination.imgtec.com/i...wervr-tbdr-and-architecture-efficiency-part-4
Games indicate development effort more than anything else. 3DMark benches are made to actually reveal what hardware is capable off- it's not the perfect metric, but yes, it matters.
More silicon and lower clocks simply equals lower power consumption. Just like in the new Air. I guess it depends how the turbo will be implemented. L4 also consumes some power, but also helps with CPU performance, so let's not jump to conclusions.
MBP 13" has never gotten dGPU, so there's no discussion.
If rMBP 15" gets iGPU, sorry I'll buy the 2012 model, for a cheaper price plus overclocked GT 650M card. It just doesn't make sense.
I hope Apple will give us GTX 765M plus IGZO display. Holla holla take my dolla!
If rMBP 15" gets iGPU, sorry I'll buy the 2012 model, for a cheaper price plus overclocked GT 650M card. It just doesn't make sense.
I hope Apple will give us GTX 765M plus IGZO display. Holla holla take my dolla!
Regular "execution unit" as defined in computer science:
https://en.wikipedia.org/wiki/Execution_unit
It's not a GPU-specific term.
Deferred rendering is something else entirely.
http://en.wikipedia.org/wiki/Deferred_shading
And by "tiling is done by a piece of hardware", they really mean "instead of using CPU cache, we are now using on-die memory".
Here's Mali:
http://infocenter.arm.com/help/topic/com.arm.doc.dui0555a/CACDBIJF.html
And here's PowerVR:
http://www.imgtec.com/powervr/insider/powervr_presentations/GDC HardwareAndOptimisation.pdf
Yes, they do have embedded buffer memory that's super high bandwidth and super low latency on those GPUs to handle the tiling. Integrating GDDR5 onto the CPU die? Same thing.
In fact, Crystalwell is actually video memory embedded. It's used instead of GDDR5 because it's cheaper, and they can share it with the CPU as L4 cache to boost OpenCL performance. But there's a reason it's L4 cache and not L1 or L2 or L3. (read: higher latency than L1, L2, or L3)
It matters how?
Look at those 3DMarks results, and look at all of the game benchmarks. Aside from Grid 2, does it seem like those numbers indicate what the hardware is "capable of"?
And funny thing: if you break down those synthetic benchmarks, then you see this:
Clearly you haven't read the rest of Anand's analysis of Iris Pro...
http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/18
He already measured CPU performance. The 4950HQ (highest-end Haswell part that has Iris Pro) is barely that much faster than the old 3720QM (mid-end Ivy Bridge part) in the last generation rMBP.
Gains are as follow:
Cinebench 11.5 (single-threaded) - 8.1%
Cinebench 11.5 (multi-threaded) - 15.5%
POV Ray 3.7RC7 (single-threaded) - 24.8%
POV Ray 3.7RC7 (multi-threaded) - 13.8%
7-zip benchmark - 6.4%
7-zip benchmark (small) - 13.3%
x264 HD - 1st pass - 11.6%
x264 HD - 2nd pass - 8.5%
So that L4 cache isn't doing much for most tasks. Those gains are the same you'd get if you go with the highest-performance Ivy Bridge part that Apple offers for the old rMBP.
I'd bet that it's fabulous for OpenCL (well, that's why Iris Pro is so much better) and tasks that require a lot of CPU <-> GPU interaction, but it's not really a miracle worker elsewhere.
But yeah, you're right. Let's wait and see. Apple still has a choice on how much TDP they want to set the Haswell chip at. And 55W is already showing sizable gain.
I think they have 2 paths:
1) They can increase Haswell TDP to the point where it pretty much matches 650M in the current generation (and gain significantly with OpenCL). At 55W (+8W), it's pretty much a match for GT 640M already, so I'd bet that they can easily get there if they set it at 70W or 75W (+23 or 28W headroom). That's still 15-20W lower than maximum power consumption of last year's parts, and it's not like the chips would run at that power profile all the time. Though idle power would suffer, which means battery life would be on average equal or worse than last year's model, though battery life at load would be better.
2) They keep TDP at stock 47W, and the chip ends up being far more power efficient than last year's part, but graphics performance takes a hit. They can probably get 10-12 hours of battery life this way, though. And that's a sizable boost from the 7 hours of last year's model.
You know there are a HEAP of professionals who do actual work that don't need (or want) a power hungry, hot discrete GPU in their PORTABLE machine (that chews battery), right?
All laptop GPUs are crap (relatively speaking). If you want maximum OpenGL or OpenCL performance, use a desktop. It's not the designer's fault - there's only so much you can achieve in 45w or less.
A portable machine is intended to be usable on battery. I'd much rather have 1.5-2x battery life and same/slightly less GPU performance than a 10% improvement in GPU and the same battery life as last year.
And no, turning the discrete GPU off doesn't work very well, unless apple can fix the software to be a LOT more intelligent with dGPU use (and mavericks, whilst better is still retarded in this respect - for the sake of what apple is going to release, I'm not going to consider gfxstatus for hard-switching GPUs in this analysis).
Given the writing is on the wall for the entire discrete GPU market within 5 years (IMHO - both mobile and desktop except for the extreme high end of desktops), Apple may consider the additional R&D on discrete GPU switching required to not be worth it and just kill the discrete GPU off in mobile machines early. It will certainly simplify and streamline their GPU code and eliminate potential bugs.
That said, I think it will be next update, rather than this one - with Broadwell.
To be honest, I'd actually just appreciate it if the dGPU was entirely turned off when on battery.
Last edited:
I wasn't suggesting anything, just wondering if you were actually making full use of the 650M at this point.
Anything to write home about? It is really close which means you can run almost exactly the same settings in most games as on a 650M. You won't be able to really tell the difference between 20% while doing anything with it. If you read reviews of new GPUs, that is the kind of difference that was quite usual between AMD and Nvidia on different games. That why performance was averaged and nobody said oh complete fail it is soo much slower. You are really being applying different standards.
The question is, is whether there is anything that you cannot do anymore because of the regression in performance. Than it was nothing to write home about. Here it mostly comes down to good drivers and that Iris Pro will be quite capable of handling anything people used to throw at the 650M.
What I tried to point out is that moving memory accesses closer saves power over wide far away access.
One could also say that dedicated is a GPU that has its own memory access. Architecturally that is what makes the difference. Whether it would be on die or not.
Intel couldn't use GDDR5 because it sucks for the Cache nature which they use. Xbox uses SRAM which has very low latency. DRAM is in between. GDDR5 is just bad in that respect. It is much better to just made a very wide access as they did with 50GB/s DRAM. AMD/Nvidia had their share of problems to get GDDR5 memory controllers power efficient. Intel putting that power hog into the die would just not work out for power efficiency. GDDR5 isn't worth it.
Just because stuff is on die, doesn't make its power consumption disappear. A big die size is even easier to cool but the GDDR5 the high clock rate memory controller will remain to be an bad choice for power consumption.
I have about 20C difference at the same fan speed between the two, doing nothing. A dGPU even if it does nothing needs a lot more power than the iGPU in active mode, because it has to power its own memory controller, memory chips, lots of low clocked but not completely power gated transistors.
You are not reading properly. I did remark on the topic of lowered performance at higher settings. (it is not just higher resolutions but also settings, we don't have any benchmarks yet of higher resolutions without higher settings. I play MW usually on rather low settings but native res).
And in all the other benchmarks the difference between Iris Pro and 650M in lower settings doesn't really grow on the higher settings. The difference appears to be relatively the same. Except for Battlefield and there it could be all kinds of things. It could be some feature that is turned on at high settings that the Intel GPU doesn't cope well and requires some driver tuning. These benchmarks don't suggest a data starved GPU.
And where did you find these benchmarks.
I very much doubt that memory performance is responsible for the difference between performance of the Iris Pro and 650M. If you actually compare the DDR3 and GDDR5 650M on notebookcheck benchmarks they are within a very tight range of on average maybe 5%. There are clock speed differences but if a 650M @ 835Mhz would be data starved in Battlefield 3 why does it yield very similar performance on playable framerates.
http://www.notebookcheck.com/Test-Acer-Aspire-V3-771G-Notebook.74162.0.html
http://www.notebookcheck.com/Test-Deviltech-Fire-DTX-MSI-MS-16GA-Notebook.76511.0.html
Battlefield doesn't appear to have memory issues. Keep in mind that the Iris Pro still has way more bandwidth available than a 650M with DDR3. I doubt very much Battlefield 3 problems have anything to do with memory access.
As for the clock rate differences keep in mind that many of the last TDP numbers released were significantly higher for GDDR5 versions. The GDDR5 memory controller has gotten better but it is still a power hog compared to DDR3.
You shouldn't overestimate it either. The difference between a HD 5200 and a 650M isn't big at all. And clearly the 650M can still keep up with just DDR3 in almost any benchmark without being starved out. You also mentioned in many posts that 128MB is too little while it actually is more than big enough. It is just a cache and a cache doesn't have to hold only enough data to reach a hit rate of some 90%. It is not comparable to a dedicated big VRAM size. The DDR3 memory is in latency still closer than GDDR5 dedicated memory.
At that time all they had was the GMA stuff. Atom was a new product and they didn't have many parts down. Merryfield will be the first actual product where they think they have all parts down. The older Smartphone Atoms where just a first try where they put their core in and stuff the rest from thrid parties in. It is not just the GPU that isn't from Intel in there and integration still sucked. It is just pointless to argue about that. Now they put the Gen7 graphics and and when it is out one can argue about where Intel graphics are.
It is like complaining about memory performance in those Atoms as if Intel couldn't do low latency integrated memory well, just because they didn't in those chips.
We have very different ideas of considerable. IMO considerable is when you don't need a benchmark to notice a big difference or when a small setting tweak doesn't fix an annoying problem.
Indeed. It is weird how people can be so fussed up about some 0-40% difference when a mainstream desktop GPU is far away and iGPU used to be less than a third the performance.throAU said:
Laptop GPUs are all crap from a Desktop perspective. iGPUs used to be all crap from dGPU perspective. Smartphone GPUs (even the hyped ones) are still all crap from an iGPU perspective. Now the new breed of iGPUs with eDRAM is still all crap from a gaming notebook 780M perspective but so is a 650M. It is somewhat ridiculous the whole discussion.
Looking at games like call of duty that says plays on 650M
Would these actually play on Iris Pro
Would these actually play on Iris Pro
U
Please read those charts again.
Difference at low resolution is indeed on average around 20-30%.
But at high resolution? As high as 70%!
That's not "really close". That's far off the mark.
"Really close" is how the GT 640 (desktop part) and 650M (laptop part) trade blows at around 10-20% difference on average.
If it's just about "can do" or "can't do", then why bother getting a MacBook Pro with a quad-core CPU to begin with?
Just get a netbook. Is there anything you "can do" on a MacBook Pro that you can't do on a $100 netbook?
And what prevented Intel from implementing faster memory than what they used? Oh yes... cost...
Because if they tried to make it any faster, it would cost more than regular CPU + dGPU. As it is right now, they're already cutting pretty close. So from a cost perspective, there are already eye brows being raised.
Nope. "Integrated" just means the GPU is joined with the CPU inside a single package. That's the standard definition of "integrated" in the English language.
Otherwise it's still a GPU.
If Intel pulled Iris Pro out and put it on the board on its own bus, it would be "dedicated" as well.
Same thing if Intel decided to integrate a 650M GPU inside their CPUs. In that case, the 650M would be "integrated" into the CPU.
GDDR5 has 80GB/s bandwidth.
For video memory (texture cache), GPUs typically need bigger bandwidth (to push everything all at once) rather than low latency (to keep pushing small things back and forth continuously).
When it's integrated on the package, it'll use less power as there doesn't have to be a dedicated power supply circuit for it. That's why desktop counterparts of laptop GPUs typically use more power, because the power supply circuit is built into the board and it wastes power... as any power supply circuit does.
Intel simply decided not to use it for cost-cutting reasons as outlined above.
Also because they'd blow up die size even more than it is now.
Not sure if you are being sarcastic.
Doing some maths on the differences
Grid 2 at 1366 x 768, no AA, High Settings: 27%
Grid 2 at 1600 x 900, 4xMSAA, High Settings: 67%
The difference in high resolution performance shows even on the "optimized" Grid 2.
And in other benchmarks, the differences don't grow? Seriously!? Let's disregard Battlefield 3 for a second, and look at 2 others:
Bioshock Infinite at 1366 x 768, Medium Settings: 37%
Bioshock Infinite at 1600 x 900, Very High Settings: 47%
Sleeping Dogs at 1366 x 768, Medium Settings: 7.9% (game capped at 60fps)
Sleeping Dogs at 1600 x 900, High Settings: 47%
Oh yeah, they DO grow at higher settings and higher resolution. Seriously, Iris Pro is bandwidth starved.
http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested
Read back a page. I already posted a benchmark where the 650M may lose up to 50% of its performance just by using DDR3.
But far less available high-speed memory than 650M with DDR3. You're comparing 128MB vs 1GB+.
No, but it has problems with low amount of memory.
There are 2 issues with Iris Pro that I highlighted:
1) It doesn't have that much "fast" memory. The 50GB/s access is only up to 128MB.
2) 50GB/s is far short of the 80GB/s GDDR5 achieves.
Battlefield 3 and many other games likely don't have any problem with low memory bandwidth, but they'd surely have problems with only 128MB of video memory, and sharing from DDR3 means Iris Pro is bandwidth-starved.
Perhaps Crystalwell has just enough bandwidth that it's comparable to GDDR5, but it's still not enough in capacity... because 128MB is a joke. There's a reason workstation graphics cards have massive amount of video memory (up to 4GB).
At 1GB, the 650M from last generation is already strained enough in Maya that I'd dread having to deal with the same scene with only 128MB of fast memory.
It's not all about gaming. There are still pro applications that deal a lot with 3D rendering, and rely a lot on the graphics processor.
Uh... it's a cache for the CPU, but it's still video memory for the Iris Pro GPU.
And as above, DDR3 does starve the 650M. It loses (in actual games, not in synthetic benchmarks that don't stress high resolution assets) on average around 20-50% of its performance level, and that's a lot.
It's also more in line with the differences we're seeing between Iris Pro and 650M.
I'm sure that anyone would notice a difference when running 3D applications even without those benchmark numbers... especially since it's consistently around 50% at higher resolutions.
GeForce GT 640 (desktop part) is consistently on par with GeForce GT 650M (laptop part) in Anandtech's benchmarks.
And the GT 640 is mainstream enough, so no, it's not "far away".
And again, for the umpteenth time, this is not just about gaming.
There are professional applications that make heavy use of 3D performance, and those are going to take a serious hit to performance because most of them have to run at native resolution on the Retina MacBook. (or even scaled to 3840 x 2400)
Please read those charts again.
Difference at low resolution is indeed on average around 20-30%.
But at high resolution? As high as 70%!
That's not "really close". That's far off the mark.
"Really close" is how the GT 640 (desktop part) and 650M (laptop part) trade blows at around 10-20% difference on average.
If it's just about "can do" or "can't do", then why bother getting a MacBook Pro with a quad-core CPU to begin with?
Just get a netbook. Is there anything you "can do" on a MacBook Pro that you can't do on a $100 netbook?
And what prevented Intel from implementing faster memory than what they used? Oh yes... cost...
Because if they tried to make it any faster, it would cost more than regular CPU + dGPU. As it is right now, they're already cutting pretty close. So from a cost perspective, there are already eye brows being raised.
Nope. "Integrated" just means the GPU is joined with the CPU inside a single package. That's the standard definition of "integrated" in the English language.
Otherwise it's still a GPU.
If Intel pulled Iris Pro out and put it on the board on its own bus, it would be "dedicated" as well.
Same thing if Intel decided to integrate a 650M GPU inside their CPUs. In that case, the 650M would be "integrated" into the CPU.
GDDR5 has 80GB/s bandwidth.
For video memory (texture cache), GPUs typically need bigger bandwidth (to push everything all at once) rather than low latency (to keep pushing small things back and forth continuously).
When it's integrated on the package, it'll use less power as there doesn't have to be a dedicated power supply circuit for it. That's why desktop counterparts of laptop GPUs typically use more power, because the power supply circuit is built into the board and it wastes power... as any power supply circuit does.
Intel simply decided not to use it for cost-cutting reasons as outlined above.
Also because they'd blow up die size even more than it is now.
Not sure if you are being sarcastic.
Doing some maths on the differences
Grid 2 at 1366 x 768, no AA, High Settings: 27%
Grid 2 at 1600 x 900, 4xMSAA, High Settings: 67%
The difference in high resolution performance shows even on the "optimized" Grid 2.
And in other benchmarks, the differences don't grow? Seriously!? Let's disregard Battlefield 3 for a second, and look at 2 others:
Bioshock Infinite at 1366 x 768, Medium Settings: 37%
Bioshock Infinite at 1600 x 900, Very High Settings: 47%
Sleeping Dogs at 1366 x 768, Medium Settings: 7.9% (game capped at 60fps)
Sleeping Dogs at 1600 x 900, High Settings: 47%
Oh yeah, they DO grow at higher settings and higher resolution. Seriously, Iris Pro is bandwidth starved.
http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested
Read back a page. I already posted a benchmark where the 650M may lose up to 50% of its performance just by using DDR3.
But far less available high-speed memory than 650M with DDR3. You're comparing 128MB vs 1GB+.
No, but it has problems with low amount of memory.
There are 2 issues with Iris Pro that I highlighted:
1) It doesn't have that much "fast" memory. The 50GB/s access is only up to 128MB.
2) 50GB/s is far short of the 80GB/s GDDR5 achieves.
Battlefield 3 and many other games likely don't have any problem with low memory bandwidth, but they'd surely have problems with only 128MB of video memory, and sharing from DDR3 means Iris Pro is bandwidth-starved.
Perhaps Crystalwell has just enough bandwidth that it's comparable to GDDR5, but it's still not enough in capacity... because 128MB is a joke. There's a reason workstation graphics cards have massive amount of video memory (up to 4GB).
At 1GB, the 650M from last generation is already strained enough in Maya that I'd dread having to deal with the same scene with only 128MB of fast memory.
It's not all about gaming. There are still pro applications that deal a lot with 3D rendering, and rely a lot on the graphics processor.
Uh... it's a cache for the CPU, but it's still video memory for the Iris Pro GPU.
And as above, DDR3 does starve the 650M. It loses (in actual games, not in synthetic benchmarks that don't stress high resolution assets) on average around 20-50% of its performance level, and that's a lot.
It's also more in line with the differences we're seeing between Iris Pro and 650M.
I'm sure that anyone would notice a difference when running 3D applications even without those benchmark numbers... especially since it's consistently around 50% at higher resolutions.
GeForce GT 640 (desktop part) is consistently on par with GeForce GT 650M (laptop part) in Anandtech's benchmarks.
And the GT 640 is mainstream enough, so no, it's not "far away".
And again, for the umpteenth time, this is not just about gaming.
There are professional applications that make heavy use of 3D performance, and those are going to take a serious hit to performance because most of them have to run at native resolution on the Retina MacBook. (or even scaled to 3840 x 2400)
Some off-topic ranting first:
I would like to state for the record that I cannot stand FPS games so I do not play games such as Crysis, Battlefield, etc. they look nice and all and I will watch people play them who are good at them. However, overall they are boring as hell in my view.
I play RPGs such as Diablo although after over 10 years, I felt Diablo III was kind of a disappointment. However, if I am going to play a PC game (or if I want to), it must be a hack and slash RPG (or even a turn by turn is acceptable such as Dragon Quest) such as Diablo or if you've ever played Gauntlet Legends/Dark Legacy in the arcade or on consoles. I like collecting gold, buying weapons and armor, and building up a powerful character. Add to that playing some Bathory in the background.
----------------------------------- End of rant
I agree with people that discrete cards are not just for gaming but still I want something solid. I care about power efficiency, heat, and everything else though dGPUs are improving just as iGPUs are improving. I mean you're telling me that with Maxwell and beyond, that Apple isn't going to take advantage? Why not? Intel I certainly give credit to where it's deserved. They aren't at the gold level yet. To me they're fighting for the silver but if the Iris Pro works well enough, the bronze medal is surely theirs. That probably doesn't make much sense but oh well.
I would like to state for the record that I cannot stand FPS games so I do not play games such as Crysis, Battlefield, etc. they look nice and all and I will watch people play them who are good at them. However, overall they are boring as hell in my view.
I play RPGs such as Diablo although after over 10 years, I felt Diablo III was kind of a disappointment. However, if I am going to play a PC game (or if I want to), it must be a hack and slash RPG (or even a turn by turn is acceptable such as Dragon Quest) such as Diablo or if you've ever played Gauntlet Legends/Dark Legacy in the arcade or on consoles. I like collecting gold, buying weapons and armor, and building up a powerful character. Add to that playing some Bathory in the background.
----------------------------------- End of rant
I agree with people that discrete cards are not just for gaming but still I want something solid. I care about power efficiency, heat, and everything else though dGPUs are improving just as iGPUs are improving. I mean you're telling me that with Maxwell and beyond, that Apple isn't going to take advantage? Why not? Intel I certainly give credit to where it's deserved. They aren't at the gold level yet. To me they're fighting for the silver but if the Iris Pro works well enough, the bronze medal is surely theirs. That probably doesn't make much sense but oh well.
Yep, for sure GT650M is slower than HD4000 in this tasks...
.Im 100% sure that some benchmarks were "tinkered" to show better score than normally...
Which of course are hardly relevant to OS X platform.
I mean, OpenCL?!
It's not like Apple invented the freakin' thing, and it's sure as hell not like they will put some pressure on developers to use it more? Has anybody heard Phil Schiller, at this year's WWDC said something like, hmm... " You all know you should use it" ?! 'Cause I didn't! Or maybe he did, but he meant something else?
So why bother giving them (developers, developers, developers
) a compelling mobile platform to target for gpu accelerated work (great compute-oriented gpu's in all of Apple's notebook lineup, top to bottom) ?Everything is possible*...however, Anand would be hard pressed to do more "tinkering" than Nvidia already did.
* ...but nothing is real ...
Last edited:
Maxwell is when? Next year? Because next year Intel will be on 14nm (at least for mobile parts), and so far every rumor points out they will have completely different gpu architecture. No reason to assume they wouldn't be competitive.
Which of course are hardly relevant to OS X platform.
I mean, OpenCL?!
It's not like Apple invented the freakin' thing, and it's sure as hell not like they will put some pressure on developers to use it more? Has anybody heard Phil Schiller, at this year's WWDC said something like, hmm... " You all know you should use it" ?! 'Cause I didn't! Or maybe he did, but he meant something else?
So why bother giving them (developers, developers, developers
Apple does market OpenCL as a selling feature of OSX Lion (10.7).
But they don't push developers to implement it.
And it's pretty clear not a lot of applications use it.
But all that aside, OpenCL is only slow on Kepler because nVidia purposefully limited the capabilities of the platform in order to sell more of their Tesla GPU computing platforms.
Just like they purposefully limited the performance and capabilities of "identical" GeForce GPUs to upwell their Quadro GPUs.
If Apple was going with an AMD GPU instead, then OpenCL performance would have been way up (and would leave Iris Pro far in the dust). But OpenGL performance would have gone down.
They can't please everyone, I guess.
Personally, I do game a little bit sometimes, and I work heavily with tasks that require decent OpenGL performance (like the iOS Simulator), so I do want more OpenGL performance than the rest.
Here's hoping that if Apple is going to go Iris Pro, then they do the same thing that they did to the 650M and overclock Iris Pro (or increase TDP profile) so that at least it matches the performance of the last generation.
If that is the case, I'd consider upgrading for the battery life. Otherwise, next year it is.