My thoughts on the upcoming Haswell rMBP's

[jafingi]

I have researched a little on the new Haswell CPU's for mobile, and here's my thoughts of what we're going to see in the next months with an rMBP update:

If we are to assume that the wait is because Apple waits for the Iris Pro CPU's, there are three choices (HQ models): i7-4950HQ (2.4-3.6GHz), i7-4850HQ (2.3-3.5GHz), i7-4750HQ (2-3.2GHz).

The i7-4950HQ runs same base- and turboboost-frequency as the current base-model, but I suspect that Apple will maybe make a cheaper base-rMBP if they discontinue the cMBP's.

If that's the case, the base model might be a i7-4750HQ (2-3.2GHz), which should deliver the same, if not better performance as the current base-model's CPU. The Iris Pro performs almost as good as the 650M in the current rMBP, so they could remove the dGPU from the base-model.

And that would actually not be a bad idea. According to Anandtech's benchmarks of Iris Pro versus 650M in rMBP, the two are actually pretty close!

Furthermore, the Iris Pro versus HD4000 in rMBP, Iris is a whole other league, that almost doubles the performance against HD4000. So running only with Iris Pro in a future rMBP base model, would give a huge performance boost in 99% of the time; I guess we'll se no more lags. When browsing and navigating the UI, it would be as if the user have forced 650M in the current rMBP, just with better battery life, and of course: less heat!

So this would be the logical way for Apple. Discontinue the cMBP, and then a base model of rMBP with only integrated GPU. That would performance-wise be like the current base-model, but have an improved battery life.

If course, as rMBP is Apple's flagship, they would also have to make a better version with discrete GPU. 750M would be a good guess. This model would have the i7-4950HQ, which have the same frequencies as the current base-model. Or maybe Apple would even go for one of the MQ models, which doesn't have Iris Pro, but instead HD4600. This could justify the small performance gains there would be with Iris Pro vs. 750M. That MQ model could be the 4900MQ (2.8-3.8GHz).

Actually, I think, that most people would benefit most of the base-model then. Although the frequency is smaller on paper, the performance would be same as the current base-model. So why would you buy a laptop with same performance as last year's IB rMBP? 1) Better battery life (no dGPU) 2) Better overall UI responsiveness (due to Iris Pro being way better than HD4000), 3) CHEAPER!

As a non-gamer, I think that I would be ready to go for a model without dGPU, and use that saved money on a larger SSD and more RAM. I think mainstream gamers would do that too. Pro's could then opt to the better model with discrete GPU to handle their games/video-editing/whatever.

What do you guys think?

 

[Scapal]

I have researched a little on the new Haswell CPU's for mobile, and here's my thoughts of what we're going to see in the next months with an rMBP update:

If we are to assume that the wait is because Apple waits for the Iris Pro CPU's, there are three choices (HQ models): i7-4950HQ (2.4-3.6GHz), i7-4850HQ (2.3-3.5GHz), i7-4750HQ (2-3.2GHz).

The i7-4950HQ runs same base- and turboboost-frequency as the current base-model, but I suspect that Apple will maybe make a cheaper base-rMBP if they discontinue the cMBP's.

If that's the case, the base model might be a i7-4750HQ (2-3.2GHz), which should deliver the same, if not better performance as the current base-model's CPU. The Iris Pro performs almost as good as the 650M in the current rMBP, so they could remove the dGPU from the base-model.

And that would actually not be a bad idea. According to Anandtech's benchmarks of Iris Pro versus 650M in rMBP, the two are actually pretty close!

Furthermore, the Iris Pro versus HD4000 in rMBP, Iris is a whole other league, that almost doubles the performance against HD4000. So running only with Iris Pro in a future rMBP base model, would give a huge performance boost in 99% of the time; I guess we'll se no more lags. When browsing and navigating the UI, it would be as if the user have forced 650M in the current rMBP, just with better battery life, and of course: less heat!

So this would be the logical way for Apple. Discontinue the cMBP, and then a base model of rMBP with only integrated GPU. That would performance-wise be like the current base-model, but have an improved battery life.

If course, as rMBP is Apple's flagship, they would also have to make a better version with discrete GPU. 750M would be a good guess. This model would have the i7-4950HQ, which have the same frequencies as the current base-model. Or maybe Apple would even go for one of the MQ models, which doesn't have Iris Pro, but instead HD4600. This could justify the small performance gains there would be with Iris Pro vs. 750M. That MQ model could be the 4900MQ (2.8-3.8GHz).

Actually, I think, that most people would benefit most of the base-model then. Although the frequency is smaller on paper, the performance would be same as the current base-model. So why would you buy a laptop with same performance as last year's IB rMBP? 1) Better battery life (no dGPU) 2) Better overall UI responsiveness (due to Iris Pro being way better than HD4000), 3) CHEAPER!

As a non-gamer, I think that I would be ready to go for a model without dGPU, and use that saved money on a larger SSD and more RAM. I think mainstream gamers would do that too. Pro's could then opt to the better model with discrete GPU to handle their games/video-editing/whatever.

What do you guys think?

My point exactly! I would go for the all day batteries, silent and cold Iris Pro version without dGPU.

 

[Blue Sun]

That would be feasible, however I'm fairly sure the i7 with the Iris Pro is more expensive than a dGPU and an i7 with the HD4600.

 

[leman]

Well, I hope that you are wrong. The Iris Pro would be a big step up for the 13" model, but a significant step back for the 15" one. The benchmarks you quote show that its 30% and more slower than the 650M - and the industry does not stay still! There are now GPUs in the same thermal category as the 650M, but faster. It too Apple long enough to take graphics seriously, and now they are including pretty decent GPU options across the line. I mean, if the difference were 5%, I wouldn't care about having a dGPU. But we are talking about close to 40-50% difference compared to cards like 750M...

----------

That would be feasible, however I'm fairly sure the i7 with the Iris Pro is more expensive than a dGPU and an i7 with the HD4600.

This is another good point.

 

[devilcm3]

i'd rather not trust anand for intel benchmarks...

you see, rumor that they have a "special" relationship with Intel....

which explains why they are the first to benchmark their mobile processor



always wait until other websites start showing reviews then we can do a cross reference of the results

 

[zipur]

My point exactly! I would go for the all day batteries, silent and cold Iris Pro version without dGPU.

Is it not true the the display is the biggest power hog on the mbp? So are we going to see a big bump in battery life?

 

[cirus]

I have researched a little on the new Haswell CPU's for mobile, and here's my thoughts of what we're going to see in the next months with an rMBP update:

If we are to assume that the wait is because Apple waits for the Iris Pro CPU's, there are three choices (HQ models): i7-4950HQ (2.4-3.6GHz), i7-4850HQ (2.3-3.5GHz), i7-4750HQ (2-3.2GHz).

The i7-4950HQ runs same base- and turboboost-frequency as the current base-model, but I suspect that Apple will maybe make a cheaper base-rMBP if they discontinue the cMBP's.

If that's the case, the base model might be a i7-4750HQ (2-3.2GHz), which should deliver the same, if not better performance as the current base-model's CPU. The Iris Pro performs almost as good as the 650M in the current rMBP, so they could remove the dGPU from the base-model.

And that would actually not be a bad idea. According to Anandtech's benchmarks of Iris Pro versus 650M in rMBP, the two are actually pretty close!

Furthermore, the Iris Pro versus HD4000 in rMBP, Iris is a whole other league, that almost doubles the performance against HD4000. So running only with Iris Pro in a future rMBP base model, would give a huge performance boost in 99% of the time; I guess we'll se no more lags. When browsing and navigating the UI, it would be as if the user have forced 650M in the current rMBP, just with better battery life, and of course: less heat!

So this would be the logical way for Apple. Discontinue the cMBP, and then a base model of rMBP with only integrated GPU. That would performance-wise be like the current base-model, but have an improved battery life.

If course, as rMBP is Apple's flagship, they would also have to make a better version with discrete GPU. 750M would be a good guess. This model would have the i7-4950HQ, which have the same frequencies as the current base-model. Or maybe Apple would even go for one of the MQ models, which doesn't have Iris Pro, but instead HD4600. This could justify the small performance gains there would be with Iris Pro vs. 750M. That MQ model could be the 4900MQ (2.8-3.8GHz).

Actually, I think, that most people would benefit most of the base-model then. Although the frequency is smaller on paper, the performance would be same as the current base-model. So why would you buy a laptop with same performance as last year's IB rMBP? 1) Better battery life (no dGPU) 2) Better overall UI responsiveness (due to Iris Pro being way better than HD4000), 3) CHEAPER!

As a non-gamer, I think that I would be ready to go for a model without dGPU, and use that saved money on a larger SSD and more RAM. I think mainstream gamers would do that too. Pro's could then opt to the better model with discrete GPU to handle their games/video-editing/whatever.

What do you guys think?

Apple's battery measurements are already on the igp so you aren't going to see much improvement gettings rid of the dgpu.

For example taking a generic 15.6 inch laptop

i7 quad + igp has the same battery life as i7 quad + 650m + optimus.

Furthermore the edram of iris has an idle power use of 0.5-1 watt which combined with the beefier igp might put power use higher than HD 4600 + optimus and dgpu.

 

[dusk007]

But if 32MB already have such a high hitrate of 95% or something as Intel says, what happens to 128MB with idle workload.
Chances are you can shut the memory controller almost completely down and run only of the edram which then would be more power efficient then the alternative.
Only Intel really knows I suppose.

I would guess that the BGA cpus are likely more efficient than the socket ones. But that really is more a guess with based on as much facts as flipping a coin. It is where I would put my money if we were betting.

In medium to high load the edram probably pays for its power consumption by lowering latency and the need to go the long way to main memory as often.

I would really like to see some decent notebooks with both chips in a detailed review. Like a Samsung 770Z5E (currently 10h wlan battery life according to notebookcheck, compared to rMBP 6h in the same test). That Anandtech review doesn't really cover power consumption very well.

Apple's battery measurements are already on the igp so you aren't going to see much improvement gettings rid of the dgpu.

You are right in that is unlikely they can claim a higher battery life on the spec sheet. In real life the improvement is most likely there. No Chrome that you have to actively prevent from switching on the dGPU. Using it as a presentation device attached to a projector doesn't need a dGPU just to display some powerpoint slides. What you actually end up with and the convenience of not having to deal with this broken graphics switching would definitely be a significant improvement.

 

[jafingi]

Apple's battery measurements are already on the igp so you aren't going to see much improvement gettings rid of the dgpu.

Well.. The computer will not switch to dGPU when just using a little intensive apps, so that would save battery.

I think we'll see a little more battery-life. Maybe not in Apple's synthetic numbers (as they are with iGPU), but in real-world usage I'm positive we'll see improvements.

 

[ValSalva]

Well.. The computer will not switch to dGPU when just using a little intensive apps, so that would save battery.

I think we'll see a little more battery-life. Maybe not in Apple's synthetic numbers (as they are with iGPU), but in real-world usage I'm positive we'll see improvements.

You are right. Although the display uses a lot of power the dGPU uses a lot too. Just look at any MBP with a dGPU plugged into an external display which forces the dGPU to be used. The battery life is practically cut in half.

I wish Apple would offer the highest CPU with an option for excluding the dGPU for us non-gamers and non-3D renderers. But of course that won't happen...

 

[jafingi]

You are right. Although the display uses a lot of power the dGPU uses a lot too. Just look at any MBP with a dGPU plugged into an external display which forces the dGPU to be used. The battery life is practically cut in half.

I wish Apple would offer the highest CPU with an option for excluding the dGPU for us non-gamers and non-3D renderers. But of course that won't happen...

Yeah I know the backlight of the screen is the part that consumes most power. Just hoping we'll see a little better battery life

Other laptops with Iris Pro are shipping in the end of July. So I hope Apple will update it there!

 

[ValSalva]

Yeah I know the backlight of the screen is the part that consumes most power. Just hoping we'll see a little better battery life

Other laptops with Iris Pro are shipping in the end of July. So I hope Apple will update it there!

End of July? That would be great! I sure hope we don't have to wait too much longer than that. I was kind of counting on a new Haswell 13" rMBP at WWDC. My fear is that now it will be October or something

 

[Feed Me]

What could Apple use for a dGPU in the higher end models? (assuming they don't jump for iGPU-only throughout)

 

[jafingi]

End of July? That would be great! I sure hope we don't have to wait too much longer than that. I was kind of counting on a new Haswell 13" rMBP at WWDC. My fear is that now it will be October or something

Same with me. Already put money aside for the 15" rMBP.

This model have Iris Pro and a HQ processor:
https://www.system76.com/laptops/model/galu1

Ships mid/late July. So I think we could be lucky!

----------

What could Apple use for a dGPU in the higher end models? (assuming they don't jump for iGPU-only throughout)

Geforce 750M is my bet.

 

[cirus]

But if 32MB already have such a high hitrate of 95% or something as Intel says, what happens to 128MB with idle workload.
Chances are you can shut the memory controller almost completely down and run only of the edram which then would be more power efficient then the alternative.
Only Intel really knows I suppose.

I would guess that the BGA cpus are likely more efficient than the socket ones. But that really is more a guess with based on as much facts as flipping a coin. It is where I would put my money if we were betting.

In medium to high load the edram probably pays for its power consumption by lowering latency and the need to go the long way to main memory as often.

I would really like to see some decent notebooks with both chips in a detailed review. Like a Samsung 770Z5E (currently 10h wlan battery life according to notebookcheck, compared to rMBP 6h in the same test). That Anandtech review doesn't really cover power consumption very well.

You are right in that is unlikely they can claim a higher battery life on the spec sheet. In real life the improvement is most likely there. No Chrome that you have to actively prevent from switching on the dGPU. Using it as a presentation device attached to a projector doesn't need a dGPU just to display some powerpoint slides. What you actually end up with and the convenience of not having to deal with this broken graphics switching would definitely be a significant improvement.

Possibly on idle but any sort of light load is going to require the main memory to be active. 128 MB is simply not enough for anything but a last level cache.

In medium to high load the edram does not cut anything from main memory. 128 MB is simply not enough (ie in games the game exe will still use system RAM and 128 MB isn't enough vram for anything modern).

My bet is that it might bring a small gain. What would get a bigger gain is intel's dram display tech where they can store the last frame in a dram cache on the display and shut down the whole igp whenever the screen is not being refreshed.

 

[dusk007]

I think you are mistaking here how a cache works. 128MB are a lot in that case. The only data that needs to be in there is what the current active threads are actually using. In a browser Window that is mostly sitting there that can be a couple kilobytes or even less.
You cannot compare the amount of memory a program has in use with the memory it needs any given second. And don't forget Intel can shut of parts of the CPU or put the memory controller into a very low power state multiple times a second. If it doesn't need to access the main memory for 90 ms that is a big deal.
128MB can be easily enough to handle all the memory requests on repetitive workload which includes basically anything in desktop usage base load including browsing. It is just the same routines over and over again.

Even under load it saves many trips to main memory and cut the load on the DDR3 controllers a lot. Anand said Intel said the hit rate hardly ever drops below 95%. That means 95% of trips that would go to main memory don't need to happen. That is who much it cuts from system memory. For a cache you just cannot compare size the only thing that matters is hit rate.
The entire memory subsystem consumes quite a few Watts.
One has to wonder when this cpu ever needs two 64bit memory controllers. Part of the reason IGPs are so much more efficient than entry level dedicated GPUs is because they don't have to maintain extra memory controllers but only scrape of what the cpu doesn't need which is a lot in most cases.

Display self refresh saves about 500 mW according to an Intel presentation a few months back. That is a lot for Tablets but not so much for 10W idling big notebooks.