How do you think Bristol Ridge will do as a mobile processor compared to Carrizo?

[The Stilt]

If they want to be competitive in power efficiency, they´ll implement VP9 in their UVD. Youtube uses VP9 for their HTML5 player and it is otherwise ramping up pretty quickly since it is completely free.

 

[monstercameron]

OEMS absolutely do want to carry AMD laptops, but they are faced with the challenge of making AMD laptops that fit somewhere in their product lineup.

I agree to an extent, I am just having difficulty seeing how they can't have a parallel product stack under a different brand or even literally stick it in the gaps lower down the product stack. PCs aren't really one size fits all and even having parallel products under the same brand can work. Then again I'm no psych or business major so maybe that isn't feasible.

 

[monstercameron]

If they want to be competitive in power efficiency, they´ll implement VP9 in their UVD. Youtube uses VP9 for their HTML5 player and it is otherwise ramping up pretty quickly since it is completely free.

Doesn't Google offer hardware design macros for a hardware vp9 codec?

http://www.webmproject.org/hardware/vp9/

Not sure about the license or costs though...

 

[cbn]

Unless you have actually tried a laptop which produces > 42W of heat, it is actually pretty hard to imagine how much noise and heat that actually is. On desktop standards that amount is basically nothing, but on a laptop it is something which I cannot personally accept.

Here is how Apple handled the 47W of the Haswell GT3e:

https://www.ifixit.com/Teardown/MacBook+Pro+15-Inch+Retina+Display+Late+2013+Teardown/18696

And here is an earlier comparison I did comparing the 28W Haswell to the 47W Haswell:

http://forums.anandtech.com/showpost.php?p=37734179&postcount=2521

As an example of different cooling gear for different TDPs I thought these teardowns of the 13.3" and 15" Apple Mac Pro Retina (late 2013) models were interesting:

https://www.ifixit.com/Teardown/MacBook+Pro+13-Inch+Retina+Display+Late+2013+Teardown/18695

https://www.ifixit.com/Teardown/MacBook+Pro+15-Inch+Retina+Display+Late+2013+Teardown/18696

The first laptop is the 13.3" model with a 28W Haswell dual core.

The second laptop is the 15" model with a 47W Haswell quad core with Iris Pro graphics.

Copper heatsink size is about the same for both, but the 47W model uses two fans (each with blowing air into a very small aluminum finned area). This compared to the 28W model which only uses one fan, but notice the aluminum finned area is much larger.

So overall, I would say both 28W and 47W set-ups have the following amount of material:

Copper heatsink material: About the same for both the 47W and 28W
heatpipe: twice as much for the 47W
fans: twice as much for the 47W
aluminum finned material (this is the part next to the fan): About the ~same total weight for both, the 47W has less of it for each fan.

So it appears the amount of metal used for cooling isn't really that different between the two TDPS, but difference in ventilation is much in favor of the 47W.

 

[cbn]

a thin 15W chassis can't support a 35W turbo.

Maybe a system where when the laptop is using AC power, the battery is removed and additional ventilation can be provided?

 

[The Stilt]

I believe Apple is using the thermal capacitance of their chassic much more effectively than the other manufacturers. The usual cheese grade laptops don´t have a alloy chassis either.

AMD STAPM (Mullins, Carrizo) relies on thermal capacitance too.

 

[DrMrLordX]

Does AMD need to worry about is how upset OEMs are over the gimping of their processors?

If said OEMs stop buying AMD processors because they can't just mash the chips into cheap craptops with parts from the leftover parts bin, then yes, AMD does need to worry about upsetting OEMs.

Considering how Intel bribed OEMs to blackball AMD in the past it seems like it would be foolish to give OEMs any opportunity to hamper performance since they could have an agreement with Intel to produce AMD products but only in gimped form.

We've uh . . . gone around about this issue before. All I can add is that we don't have any insider proof. But yeah AMD really has to be on the ball when it comes to not letting OEMs butcher products like Carrizo.

Eliminate DDR3 being used with discreet GPUs.

I don't think AMD should have allowed Carrizo to pair with dGPUs at all, period. Intel pretty much owns the CPU + dGPU laptop market, and it would be very difficult for AMD to break into that with Carrizo. The whole value proposition of Carrizo is to lean on the iGPU so that the OEM doesn't have to worry about the cost of including the dGPU, the cost of cooling the dGPU, or the cost of providing power for the dGPU, while still retaining respectable performance. The FX8800P should be in $599 "cheap" laptops built to allow cTDP to trend upwards to 35/42W mode with no dGPU. Such a product could carve out a nice niche as a value proposition.

They could still do it with mobile Bristol Ridge, though it would have been better for them to do it last year with Carrizo.

I'm not sure how representative the following list is, but you'll notice that all quad core Carrizos have dGPU (designated DX, for dual graphics). The only APUs without dGPU are the dual cores.

Yeah, that's . . . silly. But that's just my opinion.

In these situations I believe the single channel is fine for a 15W quad core with dGPU (but maybe not for the dual core with 256sp iGPU).

I don't. Carrizo in 15w mode is restricted to DDR3-1600. Single-channel DDR3-1600 provides less bandwidth than many DDR2 dual-channel configs, such as the one on my x2-3600+ which I overclocked to a whopping 3.2 GHz!!! Wow! So fast.

With that mentioned, one question is can a 35W quad core DDR4 APU with dual channel (2 x 4GB) DDR4 replace any of those 15W Quad core DDR3 APU with single channel (1 x 8GB) APU + dGPU?

It should. Overall, power dissipation between the two different configs should be the same. Power delivery would have to be handled differently. The OEM would have to use a board that was not built specifically for Carrizo-L.

Pretty sure he wouldn't do that. The 1st "fix" consists in removing any dGPU OEMs insist to add to the mix and using the extra TDP room for a proper cTDP implementation (15W while on battery, 20-35W TDP while plugged in). 2nd (optional) fix is to remove the optical disk drive and use the savings to ensure a decent battery and a somewhat slimmer unit.

Platform cost goes down, and from a marketing standpoint all they need to do is show the damn thing plays League or other MOBAs.

Yeah, that's the ticket.

btw, if AMD really wanted to push mobile dGPUs alongside Carrizo/mobile Bristol Ridge, all they'd have to do is pair them with XV chips that have failed dGPUs. I don't think the dual graphics/DX12 multigpu market is big enough that they're impressing anyone with APUs and dGPUs in the same laptop. Dual graphics is still buggy anyway.

 

[cbn]

I believe Apple is using the thermal capacitance of their chassic much more effectively than the other manufacturers. The usual cheese grade laptops don´t have a alloy chassis either.

AMD STAPM (Mullins, Carrizo) relies on thermal capacitance too.

So with Apple having 13.3" for 28W and 15" for 47W back in 2013 maybe a 35W Bristol Ridge would need a 14" aluminum chassis.

Then if the 35W Bristol Ridge chassis were made of plastic it should be 15.4" or larger?

P.S. How large is the AMD Carrizo reference laptop you were (or are) using? Is it plastic or aluminum?

 

[cbn]

I don't. Carrizo in 15w mode is restricted to DDR3-1600. Single-channel DDR3-1600 provides less bandwidth than many DDR2 dual-channel configs, such as the one on my x2-3600+ which I overclocked to a whopping 3.2 GHz!!! Wow! So fast.

I based that opinion on chips like C2Q Q9650 (3.0 Ghz Yorkfield) which came with dual channel DDR2 667.

So I figured DDR3 1600 single channel was at least equal to that....and therefore enough for a 15W Carrizo quad core without iGPU.

P.S Regarding my earlier stated opinion that single channel DDR3 1600 "maybe not enough for the dual core with 256sp". Let me change that to "it is definitely not enough" (re: 15W A6-8500 is basically ~one half of a 35W FX-8800P and we know dual channel DDR3 1600 is not enough for that 4C + 512sp processor when it is used at the higher TDP. So half that bandwidth (single channel DDR3 1600) is not enough for a processor (A6-8500) that is basically one half the 35W FX-8800P specification).

 

[cbn]

I don't think AMD should have allowed Carrizo to pair with dGPUs at all, period. Intel pretty much owns the CPU + dGPU laptop market, and it would be very difficult for AMD to break into that with Carrizo.

Some of them didn't have dGPUs.

But the others must have saw Carrizo @ 15W (using just the iGPU) as too slow for gaming. And so they went to AMD asking for the dGPU (and probably got a good deal on it as well).

P.S. For 15W, I'll bet AMD could have designed a 2M Excavator chip with smaller 256sp iGPU rather than a 512sp iGPU and not suffered any loss of performance per watt. (re: it appears according to Stilt's graphs in post #12 the performance per watt on AMD iGPUs doesn't degrade going from 350 Mhz to 750 Mhz. So smaller 256sp iGPU @ 750 Mhz is roughly equal to a larger 512sp iGPU @ 350Mhz for absolute performance and performance per watt, but at a lower cost)

 

[cbn]

The whole value proposition of Carrizo is to lean on the iGPU so that the OEM doesn't have to worry about the cost of including the dGPU, the cost of cooling the dGPU, or the cost of providing power for the dGPU, while still retaining respectable performance.

Yes, At 35W Carrizo would have competed almost purely against CPU + dGPU. (but we know the story about the memory so that didn't happen).

P.S. Another benefit (besides the ones you mentioned) should be a larger battery because of the extra integration.

 

[The Stilt]

P.S. How large is the AMD Carrizo reference laptop you were (or are) using? Is it plastic or aluminum?

I think it is a 15.4" chassis. The chassis is basically completely made out of plastic. The whole system is bending while you are prying the lid open. One of the worst quality chassis and displays I´ve ever seen. Probably the only parts made out of metal are the hinges. I dunno what´s wrong with the display since it is a AUO IPS LED panel, but I´ve never seen as much backlight leakage on any panel in the past.

It is a barebone system after all.

 

[cbn]

I think it is a 15.4" chassis. The chassis is basically completely made out of plastic. The whole system is bending while you are prying the lid open. One of the worst quality chassis and displays I´ve ever seen. Probably the only parts made out of metal are the hinges. I dunno what´s wrong with the display since it is a AUO IPS LED panel, but I´ve never seen as much backlight leakage on any panel in the past.

It is a barebone system after all.

When you use it at 35W, rather than 42W AC boost does it quiet down noticeably?

 

[The Stilt]

When you use it at 35W, rather than 42W AC boost does it quiet down noticeably?

There is no difference. The fan is controlled by the APU power management directly. Fully tied to the die temperature.

If cTDP is set to 35W, the boost power limit will always be 42W. It is possible to be configured separately for AC/DC, however at default such configuration is not used.

Anything beyond the stock 15/25W configuration is too hot, too loud and drains the battery too quickly for my taste

 

[monstercameron]

There is no difference. The fan is controlled by the APU power management directly. Fully tied to the die temperature.

If cTDP is set to 35W, the boost power limit will always be 42W. It is possible to be configured separately for AC/DC, however at default such configuration is not used.

Anything beyond the stock 15/25W configuration is too hot, too loud and drains the battery too quickly for my taste

Doesnt windows power profile adjust clocks to prolong battery life, I dont see where the battery drain would be an issue if you can just manually adjust the power profile. Heat is a problem but there are thin laptops with dgpus that run just fine, look at the macbooks etc. I dont see where a little engineering can't solve some of its rough edges.

 

[The Stilt]

Of course it does, at idle. Obviously under load it makes no difference.

 

[cbn]

Heat is a problem but there are thin laptops with dgpus that run just fine, look at the macbooks etc. I dont see where a little engineering can't solve some of its rough edges.

That 47W MacBook has two decent sized fans. I'll bet this makes heavy workloads simultaneously stressing both the CPU and GPU fairly quiet though. This at the expense of some battery size.

However, with external dGPUs becoming available (via usb-c or Thunderbolt III) I do wonder how important having two fans (or even a large iGPU) is if the average laptop is only getting heavy usage while on AC power. (re: having a dGPU outside of the laptop means only CPU part should be stressed potentially reducing noise if max turbo is controlled).

This brings up at least the following question (for a Hypothetical 35W Bristol Ridge laptop):

How should fan and battery size be balanced for any given size laptop PCB?

1.) Should the fan or fans be large at the expense of room for battery

2.) Should the battery be larger at the expense of room for fan(s) or fan size.

I would hope #2 could work. Then the laptop has a niche of relatively "long" gaming run time while on battery.

Then if #2 works, what is the best way of doing the cooling with the least amount of area spent? Two smaller fans side by side in front of a single heatsink vs. one larger fan in front of a single heatsink? Which yields a better cooling effect per volume?

 

[cbn]

Heat is a problem but there are thin laptops with dgpus that run just fine, look at the macbooks etc. I dont see where a little engineering can't solve some of its rough edges.

That 47W MacBook has two decent sized fans. I'll bet this makes heavy workloads simultaneously stressing both the CPU and GPU fairly quiet though. This at the expense of some battery size.

However, with external dGPUs becoming available (via usb-c or Thunderbolt III) I do wonder how important having two fans (or even a large iGPU) is if the average laptop is only getting heavy usage while on AC power. (re: having a dGPU outside of the laptop means only CPU part should be stressed potentially reducing noise if max turbo is controlled).

This brings up at least the following question (for a Hypothetical 35W Bristol Ridge laptop):

How should fan and battery size be balanced for any given size laptop PCB?

1.) Should the fan or fans be large at the expense of room for battery

2.) Should the battery be larger at the expense of room for fan(s) or fan size.

I would hope #2 could work. Then the laptop has a niche of relatively "long" gaming run time while on battery.

Then if #2 works, what is the best way of doing the cooling with the least amount of area spent? Two smaller fans side by side in front of a single heatsink vs. one larger fan in front of a single heatsink? Which yields a better cooling effect per volume?

Perhaps the heatsink finned area could be made larger too? This to assist any methods to provide extra ventilation when the laptop is on AC power with the battery removed.

So in this way perhaps a balance can struck where the laptop is optimized for the largest possible battery for mobile, but when on AC power it can be optimized for the most quiet operation?

 

[cbn]

When you use it at 35W, rather than 42W AC boost does it quiet down noticeably?

There is no difference. The fan is controlled by the APU power management directly. Fully tied to the die temperature.

If cTDP is set to 35W, the boost power limit will always be 42W. It is possible to be configured separately for AC/DC, however at default such configuration is not used.

So configuring separately for AC power would mean limiting to 35W while on battery? And then only getting the 42W when plugged in.

If that is true, then I would think AMD could allow any degree of TDP split to happen if they wanted to? For example: 15W while on battery and 35W or 42W while on AC power.

With that in mind, for a 35W or 42W laptop does the VRM implementation take up much more room than a 15W VRM implementation? (I'm trying to imagine a relatively small laptop that normally runs on 15W, but would be capable of 35W or 42W when on AC power.....and thus I am wondering if the extra VRMs would have a significant negative effect on laptop PCB size and thus maximum battery size)

 

[The Stilt]

So configuring separately for AC power would mean limiting to 35W while on battery? And then only getting the 42W when plugged in.

If that is true, then I would think AMD could allow any degree of TDP split to happen if they wanted to? For example: 15W while on battery and 35W or 42W while on AC power.

With that in mind, for a 35W or 42W laptop does the VRM implementation take up much more room than a 15W VRM implementation? (I'm trying to imagine a relatively small laptop that normally runs on 15W, but would be capable of 35W or 42W when on AC power.....and thus I am wondering if the extra VRMs would have a significant negative effect on laptop PCB size and thus maximum battery size)

I´ll use the current high-end FX-8800P as an example. FX-8800P has TDP of 15W as do all of the other Carrizos currently available.

TDP Limit = 15W
PPT Limit AC = 25W
PPT Limit DC = 25W
STAPM = Enabled
STAPM Duration = 200 seconds
STAPM Scalar = 100%
BBT = Enabled
BBT Scalar = 100% (of TDP)

That´s the default configuration for all 15W Carrizo APUs broken down. TDP limit is the maximum allowed, sustained power draw. STAPM stands for "Skin Temperature Aware Power Management". It is used to increase the performance by allowing the APU to exceed it´s nominal power limit, while still keeping the skin temperature of the system below the manufacturer specifications.

Since different systems have different kinds of cooling, chassis size and chassis materials used, the STAMP parameters should be calibrated by the system manufacturers to maximise the performance.

At default settings, when STAPM is enabled the power limit is effectively 25W for the first 200 seconds. Once the 200 seconds have elapsed, the power limit will again become 15W. On physically larger, high-end systems with chassic made of alloy the manufacturer could increase the STAPM duration to 400 seconds to improve the performance, while keeping the absolute TDP limitations at the default levels. The STAPM duration basically tells the APU how long it is allowed to exceed it´s nominal TDP before the skin temperature of the chassis exceeds the specifications of the manufacturer. Longer the STAPM duration is, longer the PPT limit will be available for use.

Despite the power management supports adjusting the AC / DC limits for PPT separate, by default they are both configured to the same value (25W). Additionally it is possible to configure the TDP while on battery (DC) separately. By default BBT is set to 100% which makes it equal to the nominal TDP (15W). Valid range for BBT is 50-200% of the nominal TDP.

To maximise the performance on AC and to maximise the battery life while on DC you could do following configuration:

TDP Limit = 35W
PPT Limit AC = 42W
PPT Limit DC = 20W
STAPM = Enabled
STAPM Duration = 200 seconds
STAPM Scalar = 100%
BBT = Enabled
BBT Scalar = 50% (of TDP) // 17.5W

You would have the maximum officially supported cTDP configuration (35W TDP, 42W PPT) available when plugged in to AC and 17.5W power limit enforced while on battery power.

Systems which support 35W TDP / cTDP APUs need to be built significantly stronger. Carrizo APUs need three main VRM circuits and VDDCR & VDDGFX circuits need to be built significantly stronger on 35W supporting systems. Such changes have only a minor effect on BOM and require a very moderate amount of additional area. The main issue is the cooling of the VRM. The dissipation of the VRM directly correlates with the power drawn by the load (APU). The additional phases added to the VRM will do no good, unless the dissipated heat can be removed. As the cooling in laptops in general, cooling the VRM on a laptop effectively is extremely hard (and expensive).

Both the AMD reference Carrizo system and Lenovo Y700 throttle due VRM overheating after ~ 10 minutes of proper CPU or GPU intensive use, eventhou both of them were designed to handle 35W TDP APUs. When both of the main circuits are stressed together it shouldn´t happen as easily. That´s because the power limit will be shared between the two and the power in each circuit will be effectively cut in ~ half.

 

[cbn]

To maximise the performance on AC and to maximise the battery life while on DC you could do following configuration:

TDP Limit = 35W
PPT Limit AC = 42W
PPT Limit DC = 20W
STAPM = Enabled
STAPM Duration = 200 seconds
STAPM Scalar = 100%
BBT = Enabled
BBT Scalar = 50% (of TDP) // 17.5W

You would have the maximum officially supported cTDP configuration (35W TDP, 42W PPT) available when plugged in to AC and 17.5W power limit enforced while on battery power.

I am glad to hear this is already possible with Carrizo.

Now if only the right supplemental cooling system could be be created to go go along with that.

I am thinking a modular cooling system that would snap in the notebook where the battery normally is could be one way to make this work. This supplemental cooling would add to the existing CPU cooler and fan(s). A simple example could be a plastic chassis (shaped like the battery) with a set of fans that provides a positive pressure ventilation effect in the laptop.

This would allow the following configurations:

1. Laptop with supplemental modular cooling (replacing internal battery) running at 35W/42W with AC power. This would be good for someone that mostly uses their laptop as a desktop, but wants more performance than what they would normally get.

2. Laptop with supplemental modular cooling (replacing internal battery) running at 35W/42W with portable Power bank (ie, external battery). This is how a person could get quiet operation (in a relatively small laptop) at 35W/42W on battery power.

3. Laptop with battery running at 17.5W.

 

[cbn]

The main issue is the cooling of the VRM. The dissipation of the VRM directly correlates with the power drawn by the load (APU). The additional phases added to the VRM will do no good, unless the dissipated heat can be removed. As the cooling in laptops in general, cooling the VRM on a laptop effectively is extremely hard (and expensive).

Both the AMD reference Carrizo system and Lenovo Y700 throttle due VRM overheating after ~ 10 minutes of proper CPU or GPU intensive use, eventhou both of them were designed to handle 35W TDP APUs. When both of the main circuits are stressed together it shouldn´t happen as easily. That´s because the power limit will be shared between the two and the power in each circuit will be effectively cut in ~ half.

With enough VRM cooling would be possible to get the TDP higher than 42W?

For Bristol Ridge this would show some benefit*, but I'm mainly thinking about Raven Ridge (assuming it comes with HBM).

*Benefit of going beyond 42W should be greater on the SKUs with 384sp iGPUs compared to 512sp.

 

[The Stilt]

I believe there will be models with > 35W TDP too (for FP4)

 

[cbn]

Acer's frostcore (which replaces the optical drive with an Intake fan that injects air into the laptop);

https://www.youtube.com/watch?v=VWA6gKOPEX4

So that would be another way to make a 35W laptop run cooler.

However, with this pointed out, I still like the idea of replacing the laptop's battery with a casing having two or more thin fans injecting air (coupled to a CPU heatsink with perhaps a bit more surface area on the finned area) better.

 

[MrTeal]

So that would be another way to make a 35W laptop run cooler.

However, with this pointed out, I still like the idea of replacing the laptop's battery with a casing having two or more thin fans injecting air (coupled to a CPU heatsink with perhaps a bit more surface area on the finned area) better.

Doesn't running it sans battery kind of defeat the purpose of the laptop? I get ditching the ODD, but running corded all the time or constantly swapping the battery in and out seems like a pretty severe limitation.