• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

New PS4 model (CUH-1200) tore down - consumes 26W less at load, 5 dB quieter on avg

Eurogamer (via pocketnews) has posted an article covering the new PS4 model that debuted in Japan first. It covers the motherboard changes as well as new power measurements and ambient noise measurements.

There are many significant changes, the most important going from clamshell mode GDDR5 (16 modules, sharing address/data lines in pairs) to non-clamshell with 8 modules. Power consumption has also dropped 26W from the launch model while playing a game.

The move to 8 8Gb GDDR5 modules was expected, but the drop in power consumption was perhaps not, especially since the reports noted that the package/die measurements stayed constant. Both sites correctly theorize that the reduction in GDDR5 modules plays into it, but that alone cannot explain the 26W load drop. Samsung's product offering charts (pdf warning) show that the previous GDDR5 modules and the new ones run at the same voltage, meaning mW per bit has likely not changed drastically. The articles theorize that the APU may be on an alternate 28nm node, but that seems far-fetched to explain the large gulf between the power numbers. Assuming power supply efficiency has stayed relatively constant, it likely only accounts for ~3W of the savings (88% efficiency at peak).

We can see from anandtech and this chart that a 18% drop in power consumption in between revisions is equal to those seen in revisions with die shrinks. It is completely possible that the chip designers would elect to keep the same die padout to retain package characteristics, and potentially port more existing design work from one motherboard revision to the next, if the smaller die (and thus a smaller reticle for more dies per wafer) was not the primary objective of the shrink. Less power consumption allows everything to shrink, decreasing cost and increasing reliability. On the other hand, it makes sense to maintain the physical case size as molds are fairly expensive. Microsoft elected to keep the same case until the Valhalla revision of the Xbox 360, for example. Given the cost of new nodes rising, a die with same dimensions would go up in cost compared to 28nm, which is a detracting factor to a possible process only shrink.

The question of a die shrink becomes potentially problematic in that we know TSMC produced the first APU for the PS4. AMD, along with Nvidia, are skipping the 20nm node for graphics products, meaning this shrink would have been a custom job with no ability to re-use shrunken assets from their graphics products. In fact, AMD confirmed today that they are skipping 20nm entirely. TSMC's 16nm FinFET is only in the initial stages of production, with volume likely only just starting up. They are behind Samsung in that regard, whose 14/16 FinFET are already in the latest Galaxy flagship phone. It could be that AMD has thrown a curveball and transitioned to the shared Samsung/Glofo process, but that also seems unlikely given GPU makers' strong reliance on TSMC node variants for their graphics products.

Another important point to note is that standby power has decreased by as much as 50%. This strongly indicates a process change of some sort as leakage has reduced significantly. Perhaps different variants of the 28nm offerings could create such a significant difference, but there's no product on the market that was known to go through such a transition to compare. It seems likely that some power gating would need to be involved to achieve the lower standby power, and a significant nominal voltage reduction (greater than 10%) to achieve drop in consumption at peak load. Unfortunately, I don't know enough about TSMC's 28nm process variants to make further educated comments on the matter.

Also, there does not appear to be a fix of the companion processor that turned out to be insufficient for standby downloads. Standby download power is still nearly 60W. It seems we may need something along the lines of Chipworks' gate pitch analysis to determine the process used.

Image of the new motherboard (Top side)
20150628_11.jpg

Image of the old motherboard (Top side, from iFixIt)

note: I can post more images upon request to break up the text, but wanted to encourage people to click through to the links to give the content creators their pageviews.
 
I've been waiting for a quieter model like this one. Will give my current PS4 to my brother and buy that CUH-1200 once it's available in EU.
 

Andrefpvs

Member
Almost everything sounds great. I've been waiting for a more efficient and quieter system.

This, however...:
Also, there does not appear to be a fix of the companion processor that turned out to be insufficient for standby downloads. Standby download power is still nearly 60W. It seems we may need something along the lines of Chipworks' gate pitch analysis to determine the process used.

...is a shame. So much power used for standby downloads. One would think this would be a priority fix for them.
 
Is that the whole article?

Don't quote all.

No, it's commentary I created. Everything is linked save the last two images.

...is a shame. So much power used for standby downloads. One would think this would be a priority fix for them.

Perhaps the licensing and design in of a more powerful ARM IP is just too epensive a proposition for something that doesn't help lower their ASP or reliability.
 

Caayn

Member
This unit being quieter is a nice thing and more power effecient is a nice thing.
Almost everything sounds great. I've been waiting for a more efficient and quieter system.

This, however...:


...is a shame. So much power used for standby downloads. One would think this would be a priority fix for them.
Indeed, but I'd expect such a fix (and backside USB ports) for a new SKU more in line with a slim version or a 360 elite kind of version. If Sony intends to fix that at all.
 

USC-fan

Banned
That's for the net only standby mode (See here). It indicates a significant change in process, power gating or some other change.
No. It because the ps4 has to keep the ram powered. While they are likely some other changes that help. This is the biggest change.
50% ram chips almost 50% less power in standby.
 

DJ_Lae

Member
Standby download power usage is still a bit crazy.

When did the second PS4 revision fall? I didn't realize we were already on 3, though I assume the 1100 model was snuck out quietly
 

Lethe82

Banned
No. It because the ps4 has to keep the ram powered. While they are likely some other changes that help. This is the biggest change.
50% ram chips almost 50% less power in standby.

what sort of performance difference could there be by going from Clamshell to less chips with a higher capacity?
 
Do you think we will see a PS4 version of the Xbox Elite controller, with a higher build quality and similar levels of cutstomization?
 
Former process engineer/ current test/product engineer.

The power consumption reduction is likely not due to a process change. The leakage improvement is primarily due to more efficient chips using better clock and power gating techniques. Looks like Sony is getting rev01 chips from their suppliers who did minor redesigns to improve standby current.

Generally, high quality dc/dc converters will consume less than 1uA of current in standby mode. This means the internal references on these chips are "turned" off under some form of crude UVLO (undervoltage lockout) circuitry that is lower that the true UVLO for the chip.

As for package size, you can in fact put a smaller chip in a package (most of these are chip scale or quad flat no lead packages). The concern is that while the die gets smaller, the copper/gold wires from the chip bond pads increase in length in order to be correctly bonded to the leadframes. This can cause reliability issues if process is not carefully controlled (typically shorts between wires).

Sony must have not wanted to do a full blown fresh HTOL or characterization to get their hw to the market faster. I expect the next revision will be a pretty significant shrink.
 

Bowler

Member
Wish they went all in with the hdmi. 2.0 with hdcp2.2 this way we could start getting 4k content like Amazon and netflix
 

ElTorro

I wanted to dominate the living room. Then I took an ESRAM in the knee.
I wonder if they have fixed the secondary ARM chip in the southbridge. According to an interview on a Japanese website, the secondary chip turned out to be too weak to handle the workload associated with background downloads. That's why the PS4 has to use the APU to actually download stuff. And that's why the system consumes so much power and has to run the fan at low RPM, even when downloading in standby mode.
 

Theonik

Member
Former process engineer/ current test/product engineer.

The power consumption reduction is likely not due to a process change. The leakage improvement is primarily due to more efficient chips using better clock and power gating techniques. Looks like Sony is getting rev01 chips from their suppliers who did minor redesigns to improve standby current.

Generally, high quality dc/dc converters will consume less than 1uA of current in standby mode. This means the internal references on these chips are "turned" off under some form of crude UVLO (undervoltage lockout) circuitry that is lower that the true UVLO for the chip.

As for package size, you can in fact put a smaller chip in a package (most of these are chip scale or quad flat no lead packages). The concern is that while the die gets smaller, the copper/gold wires from the chip bond pads increase in length in order to be correctly bonded to the leadframes. This can cause reliability issues if process is not carefully controlled (typically shorts between wires).

Sony must have not wanted to do a full blown fresh HTOL or characterization to get their hw to the market faster. I expect the next revision will be a pretty significant shrink.
Next revision is probably a slim next year on a 16nm finfet process.

Edit:
I wonder if they have fixed the secondary ARM chip in the southbridge. According to an interview on a Japanese website, the secondary chip turned out to be too weak to handle the workload associated with background downloads. That's why the PS4 has to use the APU to actually download stuff. And that's why the system consumes so much power and has to run the fan at low RPM, even when downloading in standby mode.
This has not been rectified in this revision.
 

ps3ud0

Member
Shame they couldnt improve the ARM chip - looking forward to when they do so that suspend/standby works as they originally imagined. Power usage for downloads in standby will become significantly lower when that happens.

ps3ud0 8)
 
Next revision is probably a slim next year on a 16nm finfet process.

Edit:

This has not been rectified in this revision.

Yes which makes more sense. This is like a revision 1.1 change where some minor spin revisions of existing chips have been made. There should be a dramatic change in the next hardware revision.
 

Blanquito

Member
Nice, will read more when I get the chance.

[Edit] Can someone link to the article about the ARM chip being underpowered? I hadn't heard that before. [/Edit]
 

ElTorro

I wanted to dominate the living room. Then I took an ESRAM in the knee.
This has not been rectified in this revision.

Sure? The markings on the southbridge have changed.

20150629_02.jpg


On CUH-1000 it is the "88EC120-BNS2" (a Marvell code), on CUH-1200 it is the "88EC128-BNS2"
 

Occam

Member
I wonder if they have fixed the secondary ARM chip in the southbridge. According to an interview on a Japanese website, the secondary chip turned out to be too weak to handle the workload associated with background downloads. That's why the PS4 has to use the APU to actually download stuff. And that's why the system consumes so much power and has to run the fan at low RPM, even when downloading in standby mode.

No, it hasn't, according to the article. 26W less power use is nice, but I certainly won't buy a new one until that's resolved, too.
 

ps3ud0

Member
Sure? The markings on the southbridge have changed.

20150629_02.jpg


On CUH-1000 it is the "88EC120-BNS2" (a Marvell code), on CUH-1200 it is the "88EC128-BNS2"
I think that guy on twitter confirmed that the power usage while downloading on standby wasnt considerably improved over the last 2 revisions. Ill try and find it

EDIT:
20150627_01.png


ps3ud0 8)
 

ElTorro

I wanted to dominate the living room. Then I took an ESRAM in the knee.
I am still amazed that an ARM chip with 256MB Ram can't handle a HTTP(S) download. How far into the past do you have to go to find a chip that is that weak? Even the Cortex-M0 can do that.
 

Creaking

He touched the black heart of a mod
Cool. I hope they do another Taco Bell promotion this year with the new model or some kind of bundle. I don't think I'd buy a brand new non-slim system, even with slight improvements, but I'm always up for playing the TB game.
 

Occam

Member
I think that guy on twitter confirmed that the power usage while downloading on standby wasnt considerably improved over the last 2 revisions. Ill try and find it

ps3ud0 8)

Also, there does not appear to be a fix of the companion processor that turned out to be insufficient for standby downloads. Standby download power is still nearly 60W.

Occam 8)
 

ElTorro

I wanted to dominate the living room. Then I took an ESRAM in the knee.
No, it hasn't, according to the article. 26W less power use is nice, but I certainly won't buy a new one until that's resolved, too.

OP claims they didn't. If they made other revisions is not known or perhaps it is fixed and they will do a FW update.

Missed that part in the OP. That's really a shame.
 

hesido

Member
Wow, what does the secondary chip do at ALL? If it's not sufficient for download, than it's not offloading background download / upload activity during gameplay, meaning Sony may have a hard time releasing some of those extra cores for games (2 cores being used for system tasks, even with that secondary chip) even when they fix that problem because of older PS4's with that problem. Talk about design oversight.
 

Theonik

Member
Wow, what does the secondary chip do at ALL? If it's not sufficient for download, than it's not offloading background download / upload activity during gameplay, meaning Sony may have a hard time releasing some of those extra cores for games (2 cores being used for system tasks, even with that secondary chip) even when they fix that problem because of older PS4's with that problem. Talk about design oversight.
Chances are they were trying to cut costs on that chip and then realised they cut too much and had to abandon many of their ideas.
 

ElTorro

I wanted to dominate the living room. Then I took an ESRAM in the knee.
Wow, what does the secondary chip do at ALL? If it's not sufficient for download, than it's not offloading background download / upload activity during gameplay, meaning Sony may have a hard time releasing some of those extra cores for games (2 cores being used for system tasks, even with that secondary chip) even when they fix that problem because of older PS4's with that problem. Talk about design oversight.

One thing it does for sure is to turn on the PS4 if there is a new download available. It also turns it on for things like Remote Play.

It is still hard to believe that the chip is just too weak. Maybe there is some sort of software issue that prevents the chip to access the filesystem. In general, having two heterogeneous chips sharing the same resources (in this case I/O) sounds tricky. I know operating systems that can run two synchronizing kernels on different chips and cooperate, but these are research. So it's likely that the ARM chips is only active while the APU is shutdown.
 

10k

Banned
The most important improvement is the 5dB drop in noise. One thing I love about the Xbox One compared to the PS4 was now quiet it was.
 

hesido

Member
One thing it does for sure is to turn on the PS4 if there is a new download available. It also turns it on for things like Remote Play.

It is still hard to believe that the chip is just too weak. Maybe there is some sort of software issue that prevents the chip to access the filesystem. In general, having two heterogeneous chips sharing the same resources (in this case I/O) sounds tricky. I know operating systems that can run two synchronizing kernels on different chips and cooperate, but these are research. So it's likely that the ARM chips is only active while the APU is shutdown.

Without having any knowledge in hardware design, I was wondering how they'd go about dealing with having to share the harddrive, I came to the conclusion that all hard-drive i/o could go through the secondary chip so the secondary chip would read data and write to RAM, where requests from the game would be prioritized, the simplest solution with respect to harddrive but this would also mean they would have to race for RAM access, so this time you would have three separate entities accessing RAM (cpu, gpu, secondary chip) but since you could only read so much from a harddrive with respect to the RAM bandwidth, I thought this wouldn't be much of a problem.

It turns out none of this was true..

If this is just a software problem, Microsoft made backwards compatibility a reality by the time Sony couldn't fix this, so I doubt it. But then again this is Microsoft, they can spare a lot of programmers for many tasks, they have a gigantic pool of programmers.

If Sony gives up on fixing this, we may see an even simpler secondary chip that just does what you say (a glorified wake on WAN chip)
 
Wow, what does the secondary chip do at ALL?

Someone correct me if I'm wrong, but I think the secondary ARM chip does everything they said it'd do except the actual background downloading on its own. So it still runs while the system is in standby and only wakes up the APU during the actual downloading. So it's only in the 60w mode a short time.

I believe the ARM chip is also responsible for all of the game DVR functionality, encoding h.264 on the fly, and live streaming.
 

hesido

Member
Someone correct me if I'm wrong, but I think the secondary ARM chip does everything they said it'd do except the actual background downloading on its own. So it still runs while the system is in standby and only wakes up the APU during the actual downloading. So it's only in the 60w mode a short time.

I believe the ARM chip is also responsible for all of the game DVR functionality, encoding h.264 on the fly, and live streaming.

Oh, I actually forgot about those DVR functionality for a while. The problem then could be having to play nicely with the file-system encryption, then.. Enough speculation from me for now though.
 

ElTorro

I wanted to dominate the living room. Then I took an ESRAM in the knee.
Without having any knowledge in hardware design, I was wondering how they'd go about dealing with having to share the harddrive, I came to the conclusion that all hard-drive i/o could go through the secondary chip so the secondary chip would read data and write to RAM, where requests from the game would be prioritized, the simplest solution with respect to harddrive but this would also mean they would have to race for RAM access, so this time you would have three separate entities accessing RAM (cpu, gpu, secondary chip) but since you could only read so much from a harddrive with respect to the RAM bandwidth, I thought this wouldn't be much of a problem.

It shouldn't be hard to use an ARM chip as a programmable I/O controller that manages access to hard drive, blu-ray and network on a block level and lower network level respectively. However, in order to actually write files autonomously, the chip would have to be in charge of managing the file system. And here begin issues with OS design, since the Kernel on the APU would have to delegate the file system to the secondary chip.

It would have been cool if they had managed to use low-end, smartphone-level hardware in the secondary chip. Something like a dual-core Cortex-A7 with 512MB ram. That would have been enough to run all background tasks except the game on the secondary chip. Then again, that would have likely required a multikernel operating system, since both chips are heterogeneous. But such OS are only research, for instance barrelfish [1][2].

[1] http://www.barrelfish.org
[2] http://sfma14.cs.washington.edu/wp-content/uploads/2013/03/sfma2014_submission_2.pdf

I believe the ARM chip is also responsible for all of the game DVR functionality, encoding h.264 on the fly, and live streaming.

No, that is done by the hardware encoder in the APU.
 
Top Bottom