Kabini Rumors

[Khato]

It depends what you actually intend to be measuring with load dynamic power. If the idle measurement doesn't actually have the CPUs at their lowest power states then should it be considered to be the start of the dynamic range instead of part way into it? We don't even know what's causing the difference but it's pretty safe to assume that it's at least partially due to the CPU not being driven to as low states as it could be, and in that case why should that not be considered partway into the dynamic range? Besides that, if the idle isn't even really very idle (I've seen 7% usage listed and that's pretty high) that's another factor.

Quite correct - using these numbers really represents a 'best case' sort of measurement in some cases (namely the multi-threaded) while in others it's questionable how accurate it is (aka the single-threaded.) Regardless, all we can guess at based on them is roughly how much dynamic power the SoC uses for the level of performance demonstrated in the review. Which implies that CPU efficiency is at least reasonable, but with the caveat that its single-threaded efficiency seems suspect. And that it may have respectable graphics performance, but only at a pretty heavy power cost.

People are using derivations like this to try to estimate how low the power consumption could be in a tablet. They're basically saying that the measurement with maximum performance profile while idle represents the SoC itself using essentially nothing and the big power consumption can be blamed on the design of the laptop.

Those sort of flawed conclusions and bias are what make these forums entertaining to read right? They add entertainment value to the actual discussion. I tend to take the opposite direction - these numbers just go to show that AMD is trying to shove a 10W+ TDP design into the tablet market. Yes they can coax it down into that range... and they're not the only one that can do that. (I find it amusing that the peak power draw is comparable to what they recorded in their review of the 18W TDP E-450 based Lenovo X130e despite their balanced idle with maximum brightness and WiFi off measurements being almost equal, 6.6W vs 6.8W.)

Their policy for clocking strictly based on active cores seems pretty rudimentary, and well behind what Intel is using (more similar to what ARM SoCs can do right now). You would think they would at least take into consideration if the 3D part of the GPU is in use or not.

Agreed. At least here's to hoping that it's a more rudimentary system that's dependent upon firmware/software since then it could just look pretty bad due to Acer botching the implementation.

 

[Piroko]

http://www.notebookcheck.com/Im-Test-AMD-A6-1450-APU-Temash.92206.0.html

The reviewer notes the decent improvement over Bobcat (+20% ipc), but also its still lackluster st performance compared to Intel ulvs. No benchmarking, but he assumes that Temash will need longer to render some websites. He also notes that playing two 1080p youtube streams works just fine and the system feels very responsive thanks to its mt performance.

Reading out GPU clocks wasn't possible, therefore no comment about the actual clock rates. The guy assumes some games are held back by st performance (demonstrated in the Dirt Showdown table) and that this might hold back some games which could otherwise be playable at lower graphics settings. He also notes that it's a huge improvement over current gen Atoms though, sometimes right at the heels of the HD4000 (ulv, I'm assuming).

No explanation between the big power consumption difference between balanced and high performance, he did note the not-TDP-related throttling when only taxing the CPU.

The notebook was cool and quiet while surfing, but did get hotter and louder while benchmarking. Power virus temperature was 80°C (I guess? Not precise here). He expects about 3h usage with the 30Wh battery while surfing with wlan and medium brightness.

His conclusion is mixed (thinking that two high clocked cores might have been preferrable in some tests) and he hints at followup tests.

 

[Cerb]

The reviewer notes the decent improvement over Bobcat (+20% ipc), but also its still lackluster st performance compared to Intel ulvs.

These chips won't sell in notebooks at typical Ultrabook prices. But, Ultrabooks with ULV Core CPUs already hardly sell at such prices, for the same reason. If they're $600+, they can expect mediocre sales. If they're $800+, they deserve to be laughed at. If they're $350+ (IE, like Zacate), with $600 being the price of a really nice one, not the entry-level, then they should do well, as long as the new Atoms follow typical Intel history (product limits if it's good, to prevent cannablism, too-high prices if it's bad, and/or an unwillingness to try to find out what consumers actually want, before trying to tell them what Intel thinks they aught to want).

 

[Piroko]

These chips won't sell in notebooks at typical Ultrabook prices. But, Ultrabooks with ULV Core CPUs already hardly sell at such prices, for the same reason. If they're $600+, they can expect mediocre sales. If they're $800+, they deserve to be laughed at. If they're $350+ (IE, like Zacate), with $600 being the price of a really nice one, not the entry-level, then they should do well, as long as the new Atoms follow typical Intel history (product limits if it's good, to prevent cannablism, too-high prices if it's bad, and/or an unwillingness to try to find out what consumers actually want, before trying to tell them what Intel thinks they aught to want).

It's a 450€ Notebook with IPS touchscreen and a good build quality (according to the review). With electronics you can do a 1:1 conversion to dollars, so expect similar Notebooks in between 400 and 500$.

 

[Gideon]

(I find it amusing that the peak power draw is comparable to what they recorded in their review of the 18W TDP E-450 based Lenovo X130e despite their balanced idle with maximum brightness and WiFi off measurements being almost equal, 6.6W vs 6.8W.)

I'd hold off any conclusions here, till we get the full laptop review as we don't know yet what's the brightness of the Acer IPS display. I suspect it's considerably better than that, as Lenovos max-brightness is hideously bad:

The LED backlight has a brightness of 173 cd/m², which falls to 142 cd/m² in the bottom right corner. Despite the matte display surface, the laptop is not suited for use in bright surroundings, as even indoors use requires the maximum brightness

The Z-60 base Toshiba got a ton of flak for it's display, which was considered unusable in daylight, and even that has a max-brightness of 270 cd/m²:

http://www.notebookcheck.net/Review-Fujitsu-Stylistic-Q572-Tablet.91078.0.html

 

[monstercameron]

Idle update from chippy

www.umpcportal.com/2013/05/acer-aspire-v5-122p-system-idle-power-and-temash-tablet-thoughts/?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+carrypadfullfeed+(UMPCPortal+Full+Feed)

 

[Exophase]

How bright can this display be if the difference between minimum and maximum brightness only results in at most 1.5W difference in power consumption? Unless the minimum is excessively bright, which seems unlikely.

 

[Khato]

I'd hold off any conclusions here, till we get the full laptop review as we don't know yet what's the brightness of the Acer IPS display.

What does that matter? Their idle power number with WiFi off and display at maximum brightness was 6.6W for the Lenovo and 6.8W for the Acer. The only differences in the delta between that measurement and the peak load measurement of 21.8W for the Lenovo and 21.9W for the Acer are from the SoC and WiFi. Which means that in this particular system the 'tablet-oriented' A6-1450 SoC is drawing roughly as much dynamic load power as the 18W TDP E-450.

 

[Olikan]

Which means that in this particular system the 'tablet-oriented' A6-1450 SoC is drawing roughly as much dynamic load power as the 18W TDP E-450.

wasn't the dual core version for tablets?

 

[Exophase]

wasn't the dual core version for tablets?

A6-1450 was promoted as available for AMD's turbo dock technology, so you could say it's for tablets too. But it'd be restricted to 1GHz there.

I'm not sure what difference that really makes though, apparently it's restricted to 1GHz here too when all four cores are active. And since it uses less power with one core active at 1.4GHz there's no reason to allow 4x1GHz and not 1x1.4GHz.

 

[Khato]

wasn't the dual core version for tablets?

From Anandtech's coverage of AMD's CES 2013 Press Event: "Temash: dual/quad-core, less than 5W, goes into fanless designs - full Windows 8 compatibility"

Not to mention the tablet reference designs that they demonstrated last month were both using quad core SoCs - http://www.tomshardware.com/news/Cape-Temash-Jaguar-Kabini-Tablet,21893.html

 

[sm625]

Screen off: 4.04W

Screen 100%: 5.1W

Screen 50%: (enough for living-room usage) , 4.5W

This is a fail right out of the gate. 4 Watts with the screen off presumably idling? What a bad joke.

 

[Arkadrel]

This is a fail right out of the gate. 4 Watts with the screen off presumably idling? What a bad joke.

this is worse:

Keyboard backlight adds about 1W to system drain which is more than I had expected.

Low powered ultra mobile units wasteing 1 watt of power, to pretty up the keyboard with backlight.

Maybe its just acers design thats not very well made.
Or maybe the Temash here, just doesnt idle very well (I hope thats not the case).


I want to know how much power the chipset for the thunderbolt uses, and why they wasted $ on that, instead of putting in a better battery instead.

 

[monstercameron]

in idle max brightness and wifi on @10W
1w keyboard backlight
1.5w max display
3.5W wifi?
4W for the rest of the system [hdd, ddr3L(?), on chip controller hub, uncertain c-states for cpu/gpu]

 

[Exophase]

3.5W wifi?

I've been trying to say already there's no way it's 3.5W just for the wifi, you know what I've been saying... But read the latest report. It's 0.5W wifi on vs off. Which is actually already much higher than I thought it would be.

I looked at some DDR3L datasheets, and power consumption for 8 256Mx8 DRAM chips @ 800MHz is about 1.5W max. That corresponds to one 64-bit DIMM. Standby is around 100mW. So two SODIMMs where only one is accessed at a time would probably use under 2W while under full load. The Prime95 test may stress this, the Cinebench tests probably don't come anywhere close.

This particular HDD only has a 0.6W idle power consumption, 0.35W parked:

http://www.storagereview.com/western_digital_scorpio_blue_review_500gb_7mm_wd5000lpvt

Basically, the idle numbers we're seeing now are at least in line with other laptops and netbooks (and not tablets). But not with max performance mode on. Display is going to be way more than just the difference between max and min brightness - first, min brightness won't be turning the backlight totally off, second, the unlit display uses power. I'm guessing something like 2.5W for the whole deal.

So at idle we have 0.5W wifi + 0.2W DDR3L + 0.6W HDD (maybe less) + 2.5W display + 1W keyboard lighting (geez wtf how is this even possible) = 4.8W of stuff outside the SoC while it's idling but with that stuff like full display brightness and KB backlight turned on. Problem is that we don't have a reading of how much power this takes when the SoC is in power saver profile. But it does suggest that the SoC uses at least a few W while idling in maximum performance mode.

 

[Gideon]

What does that matter? Their idle power number with WiFi off and display at maximum brightness was 6.6W for the Lenovo and 6.8W for the Acer. The only differences in the delta between that measurement and the peak load measurement of 21.8W for the Lenovo and 21.9W for the Acer are from the SoC and WiFi. Which means that in this particular system the 'tablet-oriented' A6-1450 SoC is drawing roughly as much dynamic load power as the 18W TDP E-450.

Agreed. To truly be competitive Temash would need cosiderably more clock speed in a given power envelope (IPC by itself is pretty decent, maybe even too much given the power-draw). GCN GPU also seems to be particularily power hungry, hopefully they manage to improve it in the next version.

This is a fail right out of the gate. 4 Watts with the screen off presumably idling? What a bad joke.

Keyboard backlight adds about 1W to system drain which is more than I had expected.

About the overall system power-draw ... it's indeed really, bad.

I guess some blame can be put on the fact that this is the first Temash design out there. Therefore the choice of HW components, BIOS etc must be rushed. Probably by both sides (AMD and Acer). Anand had an interesting paragraph in a Haswell article. It was about some component and driver mismatch, on a certain suppliers notebook motherboard, eating up a lot of energy pointlessly.

I have no doubt that some Temash designs will do better. However it would seem that low-usage power draw will probably still be pretty bad compared to ARM and Silvermont tablets.

 

[monstercameron]

I've been trying to say already there's no way it's 3.5W just for the wifi, you know what I've been saying... But read the latest report. It's 0.5W wifi on vs off. Which is actually already much higher than I thought it would be.

I looked at some DDR3L datasheets, and power consumption for 8 256Mx8 DRAM chips @ 800MHz is about 1.5W max. That corresponds to one 64-bit DIMM. Standby is around 100mW. So two SODIMMs where only one is accessed at a time would probably use under 2W while under full load. The Prime95 test may stress this, the Cinebench tests probably don't come anywhere close.

This particular HDD only has a 0.6W idle power consumption, 0.35W parked:

http://www.storagereview.com/western_digital_scorpio_blue_review_500gb_7mm_wd5000lpvt

Basically, the idle numbers we're seeing now are at least in line with other laptops and netbooks (and not tablets). But not with max performance mode on. Display is going to be way more than just the difference between max and min brightness - first, min brightness won't be turning the backlight totally off, second, the unlit display uses power. I'm guessing something like 2.5W for the whole deal.

So at idle we have 0.5W wifi + 0.2W DDR3L + 0.6W HDD (maybe less) + 2.5W display + 1W keyboard lighting (geez wtf how is this even possible) = 4.8W of stuff outside the SoC while it's idling but with that stuff like full display brightness and KB backlight turned on. Problem is that we don't have a reading of how much power this takes when the SoC is in power saver profile. But it does suggest that the SoC uses at least a few W while idling in maximum performance mode.

Good analysis, and going by amds own numbers for there dual core in idle of 1.2w, I thinks its safe to say that the quad will will 50-100% higher power idling - I am thinking two more cores, faster memory and higher GPU clock.

 

[Vesku]

The 1 watt on the back lit LEDs pretty much tells the story on how much power optimization Acer did on this product. =S

 

[inf64]

Yep we need more solutions apart from Acer's. Let's see what else gets launched.

 

[IntelUser2000]

Jaguar IPC for Cinebench 11.5 is just a bit behind Core2/Phenom II.

Don't confuse IPC with performance. Remember Jaguar has 4 cores, while a SU7300 Core 2 with 1.2GHz clocks get 0.8 points with only 2 cores. That means Jaguar has little better than 0.5x the IPC of Core 2.

 

[SammichPG]

Given AMD's past binning history with mobile parts I'd say that enthusiasts will be able to run every core at the max multiplier even under load and/or use a 15-25% reduced vcore.


I undervolted my 45nm mobile phenom by 15% and got between -8°C and -10°C out of it.

 

[strata8]

Don't confuse IPC with performance. Remember Jaguar has 4 cores, while a SU7300 Core 2 with 1.2GHz clocks get 0.8 points with only 2 cores. That means Jaguar has little better than 0.5x the IPC of Core 2.

The A6-1450 was running at 1.1 Ghz during the Cinebench test, so 70% of the Core 2 IPC, not 50%.

 

[krumme]

Yep we need more solutions apart from Acer's. Let's see what else gets launched.

When you cut down your technical marketing to next to zero thats what you get. An cheap Acer product reviewed by Notebookcheck. The absolute worse combination for AMD.

Whatever the product here, it will probably resemble by far most of the product of Temash and Kabini. Dirt cheap, wrapped fast in plastic with a mechanical hdd and a bettery the size of a phone. Probably there will be a handfull of good Temash design, perhaps 1 or 2 good kabini designs but thats it. This is meant to be dirt cheap. Nothing wrong with that, on the contrary, it will sell like hotcakes.

What we will probably see the OEM, using the cost reduction, to make product with features like here; an ips touchscreen. Its quite an visible and tangible advantage - and one the consumers prefer to good furmark benchmarks. It plays tons of high res youtube on a "good" screen, is quad core, and does everything fine in the shop.

(written on a Samsung series 9, 13.3 ib ultrabook, with high-res screen and upgraded to faster msata ssd - lol)

 

[Dresdenboy]

Don't confuse IPC with performance. Remember Jaguar has 4 cores, while a SU7300 Core 2 with 1.2GHz clocks get 0.8 points with only 2 cores. That means Jaguar has little better than 0.5x the IPC of Core 2.

Taking into account strata8's remark on clock speeds, this means 70% Core 2 Floating Point IPC. While Jaguar's 32b scalar/packed FP throughput is good (and also the 2c fmul latency!), the 64b/80b FP throughput is much lower.

Given AMD's past binning history with mobile parts I'd say that enthusiasts will be able to run every core at the max multiplier even under load and/or use a 15-25% reduced vcore.


I undervolted my 45nm mobile phenom by 15% and got between -8°C and -10°C out of it.

I'm not sure whether past AMD/GF fab performance can be used to predict TSMC's Temash binning characteristics.

 

[inf64]

Don't confuse IPC with performance. Remember Jaguar has 4 cores, while a SU7300 Core 2 with 1.2GHz clocks get 0.8 points with only 2 cores. That means Jaguar has little better than 0.5x the IPC of Core 2.

SU7300 is 1.3Ghz 45nm Penryn dual core. It scores 0.75pts in MTed portion of C11.5.
QC 1450 Temash scores 1pt @ 1Ghz in MTed portion of the benchmark. Dual core Temash at ~1.3Ghz would score : 1x1.3/1.95=0.67pts (I divided with 1.95 factor since scaling is not perfect). This is ~12% slower than DC Penryn at the same clock. Basically Temash is close per clock to 45nm Penryn here.

Now take a look at Phenom II 920 45nm and its score of 3Pts . It's a Deneb QC 45nm @ 2.8Ghz. Deneb dual core at 1.3Ghz would score : 3x1.3/2.8/1.95=0.71pts. This makes Temash at same core and clock ~6% slower, 2way Vs 3way core and 128bit fp unit vs 128bit fp unit. You can call them practically equal in this benchmark.

So 12% slower than Penryn and 6% slower than Deneb per core and per clock in C11.5. That's amazing performance per core and per watt there.