Haswell Prediction Poll: Sept? dGPU?

[zI INFINITY Iz]

This is my first post here, I've been reading this forum for a few weeks now.

Personally, I just can't imagine the new rMBP without a dedicated GPU. Not this year. Haswell is great for portable devices and an iGPU like Iris Pro is fantastic, bringing almost the same power as a mid-range dedicated GPU from 2012. It's the first (big) step of iGPU's showing they can replace dGPU's in the future.

But come on, this is the Retina MacBook Pro, Apple's top of the line portable computer, a $2000+ device. If it comes with only an iGPU like Iris Pro, I would find it insane. I have been waiting for months to finally see what my first Mac is going to look like. Paying so much money for something with the GPU power of last year's, mid-range devices is just stupid. Especially since the competition offers more for sometimes half the price.

I mean Apple has always "had big balls" and kind of "ignored" the competition so to speak, not following certain trends while focusing only on what it does itself and doing that as good as possible. But in the end, they are HUGE now. Such a company does a lot of research on the market, on their competitors and what they offer for a certain price.

Look at this new dell (dutch site): http://www.dell.com/nl/p/inspiron-15-7537/fs

Both the €799, €999 and €1199 version come with standard NVIDIA® GeForce® GT 750 M 2 GB DDR5.

I mean, seriously, these come with it and the $2000-$3000 MacBook Pro's with Retina Displays don't? A device so expensive should be fantastic in every category, not just some. In the end that's why it's so expensive, right?

To sum it up, I personally can't imagine that the new rMBP will come with only an iGPU. And if it does, it better be a lot cheaper, because if it isn't I (sadly) won't be buying my first Mac

 

[Mr MM]

There will be a dgpu, it will be the GeForce 256. And you will be forever grateful

Oh milord Im forever grateful!

Napoleon! bring me back my bunny flippers!

 

[ezekielrage_99]

depends on your definition of professional and what it does deliver to your workload the absence of the gpu

Agreed, there would be users out there keen with the battery improvement and wouldn't notice the difference for very generalised usage such as surfing the inter webs, office, facebook and twitter...

Though working in broadcast I can't see too many of the people I would be working with buying a rMBP due to needing the graphical grunt for applications such as AE, C4D and Photoshop. And not to mention if you're interested is a decent game or two.

 

[actuallyinaus]

Though working in broadcast I can't see too many of the people I would be working with buying a rMBP due to needing the graphical grunt for applications such as AE, C4D and Photoshop. And not to mention if you're interested is a decent game or two.

- those programs use opencl
- apple has requested binned iris pro chips
- 750M is just an overclocked 650M

those programs use opencl - so it would actually make sense for apple to pick the binnediris pro over the 750M as the iris pro performs better in programs that a real 'pro' user would want - the 750M is worse in 'pro' programs and only better for gaming (the rmbp is not a gaming laptop its a pro user laptop)

using iris pro will allow apple to show much larger 'major graphics improvements' using charts of final cut pro and photoshop performance numbers than they would be able to get with a 750M

mallLuxGPU is an OpenCL accelerated ray tracer that is allowing artists to render scenes much more quickly than with CPUs alone:

Sony Vegas Pro 12, an OpenGL and OpenCL video editing and authoring package:

final cut pro uses opencl link - link2
photoshop uses opencl link - link2
maya uses opencl link
blender uses opencl link
after effects ...

the iris pro only falls to the 650M in actual gaming:

and

(the 650M can't even get to 1920x1080 at 30fps so its not particularly good for gaming anyways)

although the article shows that the iris does not suffer from a memory bandwidth limitation like the dgpus sometimes can:

increase the quality and ...

 

[B...]

autocad / photosphop / .. programs use opencl - iris pro blows the 650M out of the water in opencl, the 750M is just an overclocked 650M and apple has requested binned iris pros (so the same margin should exist)

so it would actually make sense for them to pick the binned iris pro over the 750M as the iris pro performs better in programs that a real 'pro' user would want - the 750M is worse in 'pro' programs and only better for gaming (the rmbp is not a gaming laptop its a pro user laptop)

mallLuxGPU is an OpenCL accelerated ray tracer that is allowing artists to render scenes much more quickly than with CPUs alone:
Image

Sony Vegas Pro 12, an OpenGL and OpenCL video editing and authoring package:
Image

the iris pro only falls to match the 650M in actual gaming:
Image

One of those benchmarks has an HD 4000 outperforming the 650M.

 

[dusk007]

One of those benchmarks has an HD 4000 outperforming the 650M.

That is probably because that HD 4000 is part of a 77W desktop Quad Core and CPU performance is a big part in those benches. The 650M is paired with just a 45W mobile CPU.

Thing most people don't seem to get is that the GPU is actually not all that important for most of these professional applications. They are still heavily CPU dependent. GPU acceleration isn't really much of a reality still. The brunt of the load is still on the CPU.

Cinebench has a new Bench out for C4D. It will be interesting how that will do in OSX. Chances are Iris Pro will beat or get very close to a 650M in the OpenGL test under OSX. A 750M will probably stay slightly ahead.
Maybe Anand will get one Iris Pro IMac and test the new cinebench under a Mavericks Beta. The 65W R chip could even beat a 750M.

 

[ezekielrage_99]

- those programs use opencl
- apple has requested binned iris pro chips
- 750M is just an overclocked 650M

those programs use opencl - so it would actually make sense for apple to pick the binnediris pro over the 750M as the iris pro performs better in programs that a real 'pro' user would want - the 750M is worse in 'pro' programs and only better for gaming (the rmbp is not a gaming laptop its a pro user laptop)

using iris pro will allow apple to show much larger 'major graphics improvements' using charts of final cut pro and photoshop performance numbers than they would be able to get with a 750M

mallLuxGPU is an OpenCL accelerated ray tracer that is allowing artists to render scenes much more quickly than with CPUs alone:
Image

Sony Vegas Pro 12, an OpenGL and OpenCL video editing and authoring package:
Image

final cut pro uses opencl link - link2
photoshop uses opencl link - link2
maya uses opencl link
blender uses opencl link
after effects ...

the iris pro only falls to the 650M in actual gaming:
Image
and
Image
(the 650M can't even get to 1920x1080 at 30fps so its not particularly good for gaming anyways)

although the article shows that the iris does not suffer from a memory bandwidth limitation like the dgpus sometimes can:
Image
increase the quality and ...
Image

I both understand and agree there will be benefits for applications running under OpenCL via the newer Iris Pro.

I'm coming more from the point, if apple includes both then the performance across the board is going to be rather good in Haswell rMBP. The issue I am having if they drop the discreet graphics there will be users both needing and wanting this inclusion who will be penalised with just integrated graphics.

For examples there are specialised software that won't run on anything by nVidia or FireGL graphics, Metacast, ORAD and Weatherscape won't run on integrated graphics cards due to the software being 100% built around the nVidia libraries.

There are plenty of professional with stacked out laptops ready to update for the Haswell rMBP and if the discreet cards are to be drop well there are going to be some disappointed people out there who may look for different laptops.

Again I say if apple drops the discreet card then those savings should be forwarded on the customer (but I doubt this will be the case).

One of those benchmarks has an HD 4000 outperforming the 650M.

My thoughts, it makes it a little more dubious...

 

[Serban]

if Iris Pro was only for pro users, and Iris pro is better for app and OpenCL, why apple only put Iris pro only in the base model 21.5" iMac? the iMac is still under apple logo and still a pro user, its not just portable like a laptop, is for office/home use

 

[Atomic Walrus]

One of those benchmarks has an HD 4000 outperforming the 650M.

The 650m is a Kepler chip, and with Kepler unless you're looking at a GK110 without FP units disabled (Titan) you're going to see terrible compute performance. So yes, in compute the 650m can sometimes lose to last year's Intel iGPU. This was a huge deal when the Kepler cards originally came out; GTX 680 produced very impressive gaming results and terrible compute.

So here's the dilemma: If you believe that the "low" end model will have HD 5200 and the "high" end model will have HD 4600 + 750m you've created a very odd performance split. The low end model will have much better compute performance than the high end, which is more than a little confusing. I'm not saying it won't happen, it just seems like the only people demanding a 750m are gamers at this point.

 

[dusk007]

if Iris Pro was only for pro users, and Iris pro is better for app and OpenCL, why apple only put Iris pro only in the base model 21.5" iMac? the iMac is still under apple logo and still a pro user, its not just portable like a laptop, is for office/home use

The iMac primarily sells to people that use it as home desktop family PCs. Pro users are probably the much smaller group and even among pros there is only a fraction that benefit anything at all from GPUs of any kind.
For the casual home pc user the gaming capability of a 750M is worth everything while what OSX apps of any sort need really doesn't need a strong GPU at all. There is no benefit of not having to deal with the broken graphics switching on the desktop. So very little the Iris Pro has going for it.
On a mobile notebook it is different.

Also we should remember that the primary reason why many believe an Iris Pro MBP is coming is because of the Geekbench leek with the 4950hq (too expensive for an entry level option). In the past those leaks have always been accurate.
From that piece of evidence it is mostly a fight between those that prefer not having to deal with graphics switching and getting good battery life at medium high utilization versus the ones that just want max gpu capability. Most of the latter probably want it for gaming even they though some pretend it is for pro use.

Maybe Apple leaked those Geekbenches on purpose and just wanted to see how people would react. See what the fallout would be. Intel still has a bad reputation especially among semi-tech nerds. Given how long ago that leak has been I am beginning to loose confidence.
Anyway I doubt the iMac compares to well to what is needed in the mobile lineup. Different problems and different considerations.
Given the dGPU competition I think Apple has the best chance of switching right now with the least speed difference. All the current dGPUs are just tuned fairly old ones on a 28nm tech that has been around quite long now. There is new architectures and 20nm showing in 2014. Intel claims only a 40% speed increase target for braodwell, I have my doubts margins will shrink. Intel will most likely have a significant efficiency lead on 14nm but on speed not until they get at least DDR4 or better HMC going. HMC would be awesome.

 

[GSPice]

At this point, I definitely respect the best arguments for an iGPU-only as well as dGPU rMBPs.

All we know for sure is that there was much silence on apple.com, but no refresh.

We march toward:

Oct 8 - the last hope for a silent refresh before:
Oct 15 - the last hope for any refresh before:
Q1 2014.

 

[B...]

At this point, I definitely respect the best arguments for an iGPU-only as well as dGPU rMBPs.

All we know for sure is that there was much silence on apple.com, but no refresh.

We march toward:

Oct 8 - the last hope for a silent refresh before:
Oct 15 - the last hope for any refresh before:
Q1 2014.

The reasoning that because Apple often does silent refreshes on Tuesday, they would do a silent rMBP refresh on a Tuesday, is flawed. They could do it on a Monday, Wednesday, Thursday or Friday for all we know.

 

[powerslave12r]

The reasoning that because Apple often does silent refreshes on Tuesday, they would do a silent rMBP refresh on a Tuesday, is flawed. They could do it on a Monday, Wednesday, Thursday or Friday for all we know.

Woah woah woah there.. don't be bringing your fancy 'reasoning' in here.

 

[john123]

Prepare yourself for a feast my good man!

I'm gonna love this.

You guys are seriously cracked out. I mean, like I said, I really hope I'm wrong, but every bit of evidence and logic suggests I'm not.

----------

The reasoning that because Apple often does silent refreshes on Tuesday, they would do a silent rMBP refresh on a Tuesday, is flawed. They could do it on a Monday, Wednesday, Thursday or Friday for all we know.

Actually, no, the original reasoning is not flawed. It's a logical inference. Tuesday is a much higher probability day than any other day. Just because it could happen on any other day of the week does not make all those outcomes equally likely. These are not random stochastic events.

----------

We march toward:

Oct 8 - the last hope for a silent refresh before:
Oct 15 - the last hope for any refresh before:
Q1 2014.

Err, no. 2014 Q1 already began on September 29. Source: http://investor.apple.com/faq.cfm?FaqSetID=3

 

[B...]

It's a logical inference. Tuesday is a much higher probability day than any other day. Just because it could happen on any other day of the week does not make all those outcomes equally likely. These are not random stochastic event.

I agree with that. But saying that a Tuesday (October 8) is the last hope for a silent refresh is completely false. Having a higher probability of something happening on one day does not make every other day completely impossible.

 

[ppone]

I just want to know if any combinations of gpu's mentioned in the thread will fix the lag issue.

I hope it just the transition is like from iPad 3 -> iPad 4. They fixed the lag issues with the 3 by upgrading the gpu.

 

[GSPice]

The reasoning that because Apple often does silent refreshes on Tuesday, they would do a silent rMBP refresh on a Tuesday, is flawed. They could do it on a Monday, Wednesday, Thursday or Friday for all we know.

Oops.. see below..

Woah woah woah there.. don't be bringing your fancy 'reasoning' in here.

You guys are seriously cracked out. I mean, like I said, I really hope I'm wrong, but every bit of evidence and logic suggests I'm not.

----------



Actually, no, the original reasoning is not flawed. It's a logical inference. Tuesday is a much higher probability day than any other day. Just because it could happen on any other day of the week does not make all those outcomes equally likely. These are not random stochastic events.

----------



Err, no. 2014 Q1 already began on September 29. Source: http://investor.apple.com/faq.cfm?FaqSetID=3

Er, did you even think for a split second that I was implying calendar year? Why in the world would I say October is before Q1 if I was talking about the fiscal year? I'll tell you. Because you want to assume I'm a dolt.

It's as if you enjoy hunting and pecking to construe anything you like for the sake of disagreement. There's other ways to contribute intellectually besides the venue of disagreement, you know?

I agree with that. But saying that a Tuesday (October 8) is the last hope for a silent refresh is completely false. Having a higher probability of something happening on one day does not make every other day completely impossible.

There there... let me introduce you to the world of reasonable generalizations. Much less stressful.

 

[john123]

I agree with that. But saying that a Tuesday (October 8) is the last hope for a silent refresh is completely false. Having a higher probability of something happening on one day does not make every other day completely impossible.

Oh. Well yeah. I mean, as I pointed out in my other comment, his entire statement is wrong, because Q1 already began. And if he meant that October 15 is the last hope before the end of the calendar year...well, that's just silly. We've seen Apple laptop refreshes in both late October and November before.

----------

Er, did you even think for a split second that I was implying calendar year? Why in the world would I say October is before Q1 if I was talking about the fiscal year? I'll tell you. Because you want to assume I'm a dolt.

You said Q1. That's a fiscal term. Maybe you were trying to sound fancy but tripped over yourself? If you meant "by the end of the year," then that's what you should have said. I legitimately thought you were perhaps confused about the dates for the fiscal year. And anyway, as stated above, the whole "last hope" by the end of the "calendar year" is equally silly.

It's as if you enjoy hunting and pecking to construe anything you like for the sake of disagreement. There's other ways to contribute intellectually besides the venue of disagreement, you know?

When people go off half-cocked, and the potential impact of those silly comments is to confuse others, then yeah, I derive some pleasure in correcting the record.

 

[Atomic Walrus]

I just want to know if any combinations of gpu's mentioned in the thread will fix the lag issue.

I hope it just the transition is like from iPad 3 -> iPad 4. They fixed the lag issues with the 3 by upgrading the gpu.

After recently being corrected on this I had to go do some research. Basically the problem with the rMBP's frame rates (lag means latency and is not the correct term for what happens on the rMBP) is a CPU limitation. The HD 4000 is more than capable of handling the basic 2D rendering tasks assigned to it (remember that this is a GPU that can run a game like Skyrim... not well, but it can).

The short version is that the CPU is doing too much of the work that should belong to the GPU. This is from a lack of optimization at the OS level, but up until now it basically didn't matter. Anandtech's discussion of the issue expresses it better than I could:

http://www.anandtech.com/show/6409/13inch-retina-macbook-pro-review/5

The point is that the limitation right now is at the CPU level, and the OS needs to make changes to move more of the rendering work to the GPU. Mavericks supposedly does this (according to the reports of many beta testers), and having more powerful GPUs certainly couldn't hurt, at least once more work is being sent to them.

Without OS-level rendering optimizations the best way to improve scrolling speed is actually with a higher CPU clock speed, as these kinds of rendering operations are generally limited to single-threaded processing.

 

[GSPice]

Oh. Well yeah. I mean, as I pointed out in my other comment, his entire statement is wrong, because Q1 already began. And if he meant that October 15 is the last hope before the end of the calendar year...well, that's just silly. We've seen Apple laptop refreshes in both late October and November before.

----------


You said Q1. That's a fiscal term. Maybe you were trying to sound fancy but tripped over yourself? If you meant "by the end of the year," then that's what you should have said. I legitimately thought you were perhaps confused about the dates for the fiscal year. And anyway, as stated above, the whole "last hope" by the end of the "calendar year" is equally silly.


When people go off half-cocked, and the potential impact of those silly comments is to confuse others, then yeah, I derive some pleasure in correcting the record.

Q1? Only a fiscal term? Oh wow.

Carry on in your noble quest to correct all that is half-cocked, silly, and confusing here on MR.

Anyway, I will salute you if there's no dGPU.

 

[john123]

Q1? Only a fiscal term? Oh wow.

Correct. Glad to be of service.

 

[GSPice]

Correct. Glad to be of service.

Quite interesting indeed.

 

[john123]

Quite interesting indeed.

Semantics aside, you sort of glossed over the whole (and actually important) point: there's no evidence to substantiate your claim that a release between 10/16 and 12/31 isn't possible—and, indeed, history suggests late October and November releases are quite possible. (If you meant something else by the term "last hope," feel free to clarify.)

 

[GSPice]

Semantics aside, you sort of glossed over the whole (and actually important) point: there's no evidence to substantiate your claim that a release between 10/16 and 12/31 isn't possible—and, indeed, history suggests late October and November releases are quite possible. (If you meant something else by the term "last hope," feel free to clarify.)

Well there's always a chance, isn't there...

 

[Vanilla35]

For the casual home pc user the gaming capability of a 750M is worth everything while what OSX apps of any sort need really doesn't need a strong GPU at all. There is no benefit of not having to deal with the broken graphics switching on the desktop. So very little the Iris Pro has going for it.
On a mobile notebook it is different.

So then why did they include Iris Pro in the iMac?

Maybe Apple leaked those Geekbenches on purpose and just wanted to see how people would react.

Of course they did.

Anyway I doubt the iMac compares to well to what is needed in the mobile lineup. Different problems and different considerations.

I think that is false hope. The rMBP and cMBP have seen the same GPU (including amount of dRAM) as the 21.5 inch iMac for the past 2 years.

You said Q1. That's a fiscal term.

My class started at 1pm today and ended at 2pm. During Q1 of this class period, I sat at my desk figuring out what my schedule for tomorrow would be. For Q2 I was chatting to the person next to me. During Q3 of the class I actually paid attention to the teacher. Q4 was cut short by early dismissal.