AMD Carrizo Pre-release thread

[Abwx]

It's a freaking 17" laptop, how many watts do you think go into driving that display?

Impact at the main of a CPU that actualy drain 15W from its VRMS is barely 19W, any people with a basic knowledge of EE will tell you the same thing.

There s reviews with other CPUs that are 15W and you wont see such high figures as 48W even if there s a 17" screen, here a 15W Beema with such a screen, there s 20W difference with said 4210U, that s roughly 15W more at the CPU level, the 4210U is used with a 30W+ TDP in said laptop.

http://www.notebookcheck.net/HP-Pavilion-17z-Notebook-Review.124669.0.html

Anyway i like how some people are systematicaly apologetic once it comes to a certain brand (not talking of you NTMBK) but if this 4210U was really a 15W item then with Beema AMD managed to create a 0W TDP APU...

Dont confuse him with facts. In his universe, every intel cpu draws 2x its TDP and AMD's draw 1/2.

I provide numbers while you re just threadcrapping with yet another ad hominem since your favourite firm overspecced perf/watt myth is bursted, try to contradict my saying with numbers if you can but i know that you wont, only thing you know is attacking the messenger..

Get used to it, mobile Haswell are power hogs and have lower perf/Watt than both Beema and Baytrail.

 

[coercitiv]

3D Mark is 100% right.

What is wrong is Intel sTDPs wich are obviously hugely underspecced, same as BDW 5600U, that is 15W TDP at ark.intel but roughly 30W+ once ypu test such a product.

44W at notebookcheck for a 4510U equiped laptop :

http://www.notebookcheck.net/Dell-Inspiron-17-5748-Notebook-Review.122422.0.html

People should definitly pay attention to measurement and stop using official specs that are obviously completely wrong.

It's a freaking 17" laptop, how many watts do you think go into driving that display?

Abwx brought this up in the Broadwell thread as well, and I thought I had explained it well enough there.

Intel ULV CPU have configurable TDPs. In the case of Haswell U hey have their base at 15W which ensures operation at base clocks, a lower one of 11.5W for 800Mhz, and a higher one of 25W for operating at maximum clocks.

In the case of the Haswell 4510U, which I happen to own in an ultraportable, the 15W TDP ensures only 2Ghz sustained speed in Prime 95. In order for the CPU in my 13" ultrabook to complete a Cinebench 11.5 run at 2.8Ghz as described in the quoted review it needs to be manually configured at 25W TDP.

Many OEMs use this cTDP and put ULV processors in machines with enough cooling for more than 15W (usually in 15" and 17" chassis). Then they configure the CPU to operate as they wish.

Hopefully Carrizo will bring idle power consumption down low enough to make battery life comparisons easier to make: with comparable idle power consumption, any difference in battery life will likely depend on CPU power draw under load.

 

[AtenRa]

Abwx brought this up in the Broadwell thread as well, and I thought I had explained it well enough there.

Intel ULV CPU have configurable TDPs. In the case of Haswell U hey have their base at 15W which ensures operation at base clocks, a lower one of 11.5W for 800Mhz, and a higher one of 25W for operating at maximum clocks.

In the case of the Haswell 4510U, which I happen to own in an ultraportable, the 15W TDP ensures only 2Ghz sustained speed in Prime 95. In order for the CPU in my 13" ultrabook to complete a Cinebench 11.5 run at 2.8Ghz as described in the quoted review it needs to be manually configured at 25W TDP.

Many OEMs use this cTDP and put ULV processors in machines with enough cooling for more than 15W (usually in 15" and 17" chassis). Then they configure the CPU to operate as they wish.

Hopefully Carrizo will bring idle power consumption down low enough to make battery life comparisons easier to make: with comparable idle power consumption, any difference in battery life will likely depend on CPU power draw under load.

Even if you enable the cTDP and increase the TDP to 25W the CPU (Haswell U) maximum Turbo frequency will not change.
What you gain is a prolong time of Turbo at maximum turbo frequency (depending of the Thread count).

So without increasing the CPU Frequency above the maximum Turbo and the CPU voltage, the power usage will remain partly the same (it will only increase if your temperature will also increase).

So, even with a better cooler and an increase in the cTDP the maximum power consumption will not change(unless you have higher temperatures but that is unlikely due to higher TDP heat-sink fun).

Have a look at the Intel 4th Generetion Mobile -U PDF page 61 paragraph 5.4.1 Configurable TDP

Configurable TDP availability may vary between the different SKUs.
With cTDP, the processor is now capable of altering the maximum sustained power with an alternate IA core base frequency. Configurable TDP allows operation in situations where extra cooling is available or situations where a cooler and quieter mode of operation is desired. Configurable TDP can be enabled using Intel's DPTF driver or through HW/EC firmware. Enabling cTDP using the DPTF driver is recommended as Intel does not provide specific application or EC source code.

In each mode, the Intel Turbo Boost Technology 2.0 power and frequency ranges are reprogrammed and the OS is given a new effective HFM operating point. The Intel DPTF driver assists in all these operations. The cTDP mode does not change the max per-core turbo frequency.

 

[coercitiv]

Even if you enable the cTDP and increase the TDP to 25W the CPU (Haswell U) maximum Turbo frequency will not change.

Indeed, but where did I state otherwise?

What you gain is a prolong time of Turbo at maximum turbo frequency (depending of the Thread count).

At 25W TDP you gain indefinite (close to) maximum turbo frequency, provided cooling capacity is adequate. At 15W TDP the turbo time window is limited (if any), usually under 30 seconds of sustained load.

 

[AtenRa]

Indeed, but where did I state otherwise?

At 25W TDP you gain indefinite (close to) maximum turbo frequency, provided cooling capacity is adequate. At 15W TDP the turbo time window is limited (if any), usually under 30 seconds of sustained load.

What im saying is that maximum power usage will be very close because maximum turbo frequency doesnt changed with a higher TDP.
What will eat more power is the platform(motherboard etc) and not the CPU.

 

[coercitiv]

What im saying is that maximum power usage will be very close because maximum turbo frequency doesnt changed with a higher TDP.

Maximum turbo frequency is the same, but sustained clock speed increases considerably.

For reference, an undervolted Haswell Ult requires ~21.5W to mantain 2.6Ghz in Prime 95.
Dropping turbo boost max power to ~19.5W will limit clocks to around 2.5Ghz (2.4-2.6). At 15W TDP it will stay at 2Ghz.

 

[burninatortech4]

Sounds good! I'll be more excited when AMD releases a APU + motherboard package with integrated HBM.

 

[maarten12100]

It's a freaking 17" laptop, how many watts do you think go into driving that display?

3 to 4W tops :hmm:

My 39 inch desktop VA screen uses a mere 40W and has a heavy backlight a heavy duty Tcon, heavy Tcon with a full blown ARM pc (smart tv) in it so yeah laptop screens don't consume a lot of power :awe:

The 3 to 4W is for the backlight and tcon what it takes to render the pixels every refresh rate is just the power consumed by the gpu/igp at that scenario.

 

[Enigmoid]

3 to 4W tops :hmm:

My 39 inch desktop VA screen uses a mere 40W and has a heavy backlight a heavy duty Tcon, heavy Tcon with a full blown ARM pc (smart tv) in it so yeah laptop screens don't consume a lot of power :awe:

The 3 to 4W is for the backlight and tcon what it takes to render the pixels every refresh rate is just the power consumed by the gpu/igp at that scenario.

Depends on the display. I can say for my laptop (15" 1080p 250 nit screen) going from minimum brightness to maximum brightness results in the power changing from 22W to 28W as measured on a typical power meter.

That panel is rated for 8.3W power consumption with a backlight power of 4.95/5.46W (typ/max) power.

http://www.panelook.com/N173HGE-E11_Innolux_17.3_LCM_parameter_20840.html

However that appears to be at 300 nits. Acer seems to have upped the brightness to ~350 nits (contrast is also significantly better) so its probably along the lines of nearly 9-10W power at max brightness.

 

[DrMrLordX]

Should improve touch interfaces and, if I'm reading correctly, better system responsiveness when running multiple GPU compute programs.

GPU context switching and interruptability are HSA hardware features, and are indeed lacking in Kaveri, along with hardware micro-task priority queues and dispatch. Of course, they are also lacking in every other chip made, and their absence has not been particularly "sore" thus far. The features would be "sorely" lacking if the intent was to maximize performance in very low level HSA coding. But the micro-task hardware also could have more conventional uses, such as real time coding.

Okay, that helps me to understand the matter a little better. I've been so used to desktop x86 CPUs context-switching near-seamlessly that it did not occur to me that a GPU would have problems doing the same when used as a compute device. Seems that Carrizo's "HSA Readiness" could help it with HSA programs that send work to the GPU from a large number of concurrent threads, not just when running multiple different HSA applications simulatenously.

 

[Abwx]

3 to 4W tops :hmm:

My 39 inch desktop VA screen uses a mere 40W and has a heavy backlight a heavy duty Tcon, heavy Tcon with a full blown ARM pc (smart tv) in it so yeah laptop screens don't consume a lot of power :awe:

For main supplied panels you can account 8W that are litteraly thrown away uselessly by primitive power factor corrections.

Depends on the display. I can say for my laptop (15" 1080p 250 nit screen) going from minimum brightness to maximum brightness results in the power changing from 22W to 28W as measured on a typical power meter.

That panel is rated for 8.3W power consumption with a backlight power of 4.95/5.46W (typ/max) power.

http://www.panelook.com/N173HGE-E11_Innolux_17.3_LCM_parameter_20840.html

However that appears to be at 300 nits. Acer seems to have upped the brightness to ~350 nits (contrast is also significantly better) so its probably along the lines of nearly 9-10W power at max brightness.

There can be power hogs in panels but you know well that Intel have their say in the OEMs designs such that Intel based items benefit from better components power comsumption wise, so the exemple above doesnt apply at all or eventualy for an AMD based laptop...

Indeed AMD have clearly stated that they will do the same as their competitor and ensure that power efficient components will be used in Carrizo based items.

 

[Enigmoid]

There can be power hogs in panels but you know well that Intel have their say in the OEMs designs such that Intel based items benefit from better components power comsumption wise, so the exemple above doesnt apply at all or eventualy for an AMD based laptop...

Indeed AMD have clearly stated that they will do the same as their competitor and ensure that power efficient components will be used in Carrizo based items.

I'm sorry. The same panel is the same panel. There may be better bins but that is up to the OEM and the manufacturer to decide and unless direct proof is obtained for a particular example cannot be measured.

Otherwise, please stop changing the subject. I have provided proof that the panel sucks an appreciable amount of power.

 

[Abwx]

I'm sorry. The same panel is the same panel. There may be better bins but that is up to the OEM and the manufacturer to decide and unless direct proof is obtained for a particular example cannot be measured.

Otherwise, please stop changing the subject. I have provided proof that the panel sucks an appreciable amount of power.

That s wrong, you have proved that one panel is power hungry, all panels doesnt consume the same, that s really a poor argument of yours.

As for the subject this is you that is talking of panel power footprints, yet another excuse for Intel s big TDP, indeed with the same accuracy as your other arguments in other discussions..

That said i rest my case that Intel has a big influence in designs and that the Intel besed items benefit from better components generaly, particularly the panels, just check the existing products...

Edit :

There s reviews with other CPUs that are 15W and you wont see such high figures as 48W even if there s a 17" screen, here a 15W Beema with such a screen, there s 20W difference with said 4210U, that s roughly 15W more at the CPU level, the 4210U is used with a 30W+ TDP in said laptop.

http://www.notebookcheck.net/HP-Pavilion-17z-Notebook-Review.124669.0.html

Anyway i like how some people are systematicaly apologetic once it comes to a certain brand (not talking of you NTMBK) but if this 4210U was really a 15W item then with Beema AMD managed to create a 0W TDP APU...

read what i edited, this will spare us pollution and your bad faith since the Beema is also using a 17" and is tested at max brighteness.

 

[Abwx]

AMD tells us Carrizo parts will fit into power envelopes ranging from 12 to 35W. That is the "breadth of the design space" for this chip, although, cryptically, the firm also says that "doesn't mean it doesn't scale beyond that." I'm not sure what to make of that statement. Either it's purely an engineering sentiment, or it might mean we could eventually see Carrizo-based products that push below 12W or above 35W, though presumably they wouldn't be entirely optimal implementations of the chip.

http://techreport.com/review/27853/amd-previews-carrizo-apu-offers-insights-into-power-savings/3

 

[krumme]

http://techreport.com/review/27853/amd-previews-carrizo-apu-offers-insights-into-power-savings/3

Then they for once can start to focus where the apu is strong and have a visible and solid cost/benefit advantage to oems and consumers.

Trying to make 5 or 50w parts of this is just a pain to watch and clouding the entire picture and a waste of ressources. Its doom to fail.

 

[Enigmoid]

That s wrong, you have proved that one panel is power hungry, all panels doesnt consume the same, that s really a poor argument of yours.

I don't know if this is deliberate or intentional but I was talking about that specific screen. No duh different screens may be different but the screen in question sucks down an insignificant amount of power.

I never said all screens in general. Even in the beginning I specifically said that that laptop has a 17" 350 nit, 1080p screen which sucks down quite a bit of power.

I don't know where you got that I was talking about all laptops.

Edit: The beema laptop is a 900p screen at 240 nits.

 

[maarten12100]

Then they for once can start to focus where the apu is strong and have a visible and solid cost/benefit advantage to oems and consumers.

Trying to make 5 or 50w parts of this is just a pain to watch and clouding the entire picture and a waste of ressources. Its doom to fail.

Going by the Carrizo slides a core pair aka a module takes up 2,5W at 0.4 times the frequency of regular Carrizo. So while those normilized scales tell us little it means that Carrizo can still run at low power with extremely good efficiency. Just that it won't go bellow that. (by a lot)

Also about the screens consuming a lot of power I have various laptops here it's literally a few watt for each of them I don't see how it is a big enough factor to consider.

 

[sm625]

The integrated PCH is going to have a huge impact on battery life. Right now there is no regulation of the PCH's power consumption on intel notebooks. It is a kind of phantom power draw; it isnt listed, just like RAM and screen draw isnt usually listed. That is why you cant trust the TDP numbers at face value. When you fully load the CPU, the PCH power consumption also rises, as does the various VRM power consumptions. It isnt all that hard to believe that a 15W intel notebook could idle at 9W yet jump up to 45W under load. That is only a 30W jump. Half of that jump comes from the CPU itself, while the other half comes from the PCH, the VRMs, RAM, and any other devices on the motherboard that will increase their power draw when the CPU is not idle. I'm betting that the cpu itself is probably going a bit over 15W though. I think intel is automatically compensating for their integrated VRMs in their TDP calculations. This would allow the actual power consumption of the entire chip to be higher than 15W even though it is a 15W TDP part. But that would only explain a few watts. Most of it has to be coming from the PCH, RAM, and various the motherboard components.

That's what makes Carrizo a really big deal. By getting rid of the PCH, they get rid of a huge source of "phantom" power draw, not just from the PCH but from the area of the motherboard around the PCH and the discrete components that support it as well. The AMD FCH (Bolton) TDP is 7.6W (max, definitely not typical). Getting rid of that is a bigger deal than a process node jump. Of course its not really gone, but it has at least been cut down by a very large amount. When you consider 5 Watts at the PCH probably equates to 6.5W at the battery terminals, it is even more significant.

 

[coercitiv]

The integrated PCH is going to have a huge impact on battery life. Right now there is no regulation of the PCH's power consumption on intel notebooks. It is a kind of phantom power draw; it isnt listed, just like RAM and screen draw isnt usually listed. That is why you cant trust the TDP numbers at face value.

I was under the impression that CPU Package Power was an estimate for power drawn by the entire package. Also RAM power draw estimate is listed in readings from software like HWInfo.

Nevertheless, the PCH move should help with idle power draw.

Any word on what "Integrated voltage regulation" means for AMD?

 

[cbn]

now this is interesting, http://www.amazon.com/Crucial-Ballis...&keywords=ddr4
8GB ddr4 2400 for ~$90

4GB DDR4 2400 has been $29.99 AR recently:

http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=9162548

However, according to the following, the DDR4 enabled version of Carrizo won't come with the initial release of Carrizo:

http://www.bitsandchips.it/hardware/52-english-news/5283-carrizo-die-shot-dual-ddr3-and-ddr4-imc

As we said some months ago, AMD integrated a dual IMC (DDR3 and DDR4) in Carrizo and Toronto, but the DDR4 IMC is disabled in Carrizo. The same die will be used in the consumer market (Carrizo) and in the enterprise market (Toronto), because AMD have to optimize the production.

This choice was made because the price of DDR4 memory will remain high till 2016, according to Intel sources, so the DDR3 memory will remain the leader in this market for a long time. The only real advantage of the DDR4 over the DDR3 is about the lower power consumption, but it's a useless advantage if we consider that in the high-density server market the LPDDR4 will dethrone the classical DDR4 chips.

Even Intel, for the first time, has taken a similar decision with SkyLake integrating a dual DDR3/DDR4 IMC, to promote its own new platform.

However, in the near future, AMD could commercialize a new Carrizo revision with the DDR4 IMC enabled, if the DDR4 memory will be cheaper (on the other hand, the GDRR5 IMC of Kaveri will never be enabled). Toronto will have the both IMC enabled from the beginning.

 

[Xpage]

how much room does have a DDR4 controller add to the chip?

 

[maarten12100]

how much room does have a DDR4 controller add to the chip?

It's already there however it is disabled. Doesn't matter as there is nothing to gain except lower power consumption as there is no increase in speed when effective clocks are the same. Delta colour compression should just be enough for the 512SPs and even that is a bit doubtful.

 

[Shehriazad]

Man...those Excavator modules are ridiculously small. They look like half the size of the previous generation (looking at it from a total space consumed perspective). Really looking forward to benchmarks of the fastest BGA models.

If they are any good I might grab one for a mini build if the price is right since the iGPU performance in the price category of AMDs APUs is generally impossible to beat with other iGPUs. To beat the price/performance ratio you generally have to take a cheap dGPU/CPU combo.

 

[NostaSeronx]

how much room does have a DDR4 controller add to the chip?

None over DDR3, what is labelled as DDR3/PHY is also DDR4/PHY. Carrizo and Carrizo-L aren't HEDT so DDR4 isn't planned for that type. Got to wait for Carrizo-E/BristolRidge for HEDT models and quad-channel DDR4.

 

[monstercameron]

http://www.planet3dnow.de/cms/14527-modellbezeichnungen-von-carrizo-l-bestaetigt/

carrizo-l skus