AnandtechAMD Carrizo ExcavatorReview

[frozentundra123456]

While I agree a 35 watt version would be nice as another option, I am not sure how it would fit in. It would require a bigger chassis and pretty robust cooling, ruling out thin and light. And it still is not going to be a gaming powerhouse, especially with the tdp shared between cpu and igp. So again, you are running into the problems that apus have had since the beginning, aimed at gaming, but not quite there performance wise except for less demanding games at lower settings. Someone who doesnt want to game will most likely go with a 15 watt intel solution, and for serious gaming, you will still need a quad intel, and a dgpu. So there is a niche there, granted, but rather small i think.

 

[Dave2150]

Yes, DDR4 has up to 16GB per stick. (Even DDR3 also has 16GB stick, but those parts are going to be rare to find)

DDR4 can have 32GB, 64GB and even 128GB per stick and even more in the future

 

[el etro]

How about this Carrizo bench? Could anyone find the tester to he gets his results here to we discuss?

http://browser.primatelabs.com/geekbench3/4654211

From what i know is a Full locked 3.4Ghz all cores bench, power limit topping the 42W on the AC, a thing that didn't happen in any other notebook test. Memory is Dual-Channel here.


Can The Stlit explain this good bench for us?




Edit: Updated the pick with a comparison with the best GB3 15W bench with the best GB3 35W bench.

 

[Shaun_Brannen]

The problem with heavy laptops (at least IMO) isn't the weight itself, it's the difference in how they're used. If you have a laptop for work or school and it goes between desk, bag, table to desk whether it weighs 3 or 6 lbs doesn't matter much. If the laptop is something you use as your main computing device and you want to lounge with it on the couch, those extra lbs add up quick. My Dell E6540 weighs over 6.5lbs with the 9 cell battery, and I have no trouble carrying it and moving it in and out of a bag. Sitting on the couch and holding it out at 90 degree with my arm at full extension to pass it to someone sitting on the loveseat isn't going to make me drop it, but after a second or two you sure notice the difference between that and our 2.6lbs UX305FA.

:thumbsup:

 

[Shaun_Brannen]

Dude, it's right in the review.

I usually try to take it upon myself to get a detailed knowledge of the brand/idea/whatever I'm defending/promoting/passionate about... tends to save me some embarrassment.

 

[deasd]

How about this Carrizo bench? Could anyone find the tester to he gets his results here to we discuss?

http://browser.primatelabs.com/geekbench3/4654211

From what i know is a Full locked 3.4Ghz all cores bench, power limit topping the 42W on the AC, a thing that didn't happen in any other notebook test. Memory is Dual-Channel here.


Can The Stlit explain this good bench for us?




Edit: Updated the pick with a comparison with the best GB3 15W bench with the best GB3 35W bench.

You don't know? SOME GUYS in this forum have agreed Geekbench is a CRAP. Period.:sneaky:

 

[el etro]

You don't know? SOME GUYS in this forum have agreed Geekbench is a CRAP. Period.:sneaky:

Is not a crap if you consider/compare subtest to subtest(being at same S.O.). Many GB tests are pretty good to measure performance at common kinds of codes. The best thing about GB is that it still don't favors one or other CPU design, being the best(that we have today) multi-platform system benchmarking tool.





PS: Now i got the tone of your post.

 

[Shaun_Brannen]

You don't know? SOME GUYS in this forum have agreed Geekbench is a CRAP. Period.:sneaky:

Frankly, it'll be a very long time (never?) before the general public and even informed people truly trust synthetic benchmarks.

 

[Blitzvogel]

First solution - when using commonality platforms, DO NOT design them in such a way to cripple your best products. If anything it should be designed around the best performing product in terms of processing capability, with the capability to run lower end processors.

Second solution - Contractual obligations. "Systems built with dual-channel processors must be shipped with memory occupying both memory channels"

Third solution - Design your processors to be less RAM dependent (L3 cache, HBM).

I do think Bristol Ridge is supposed to ship with more stringent cTDP, in that the configurability is limited in specific processors, probably as a way to prevent the Carrizo debacle from happening again.

AMD has essentially screwed themselves over by not imparting some regulation into how their products are used when it comes to OEMs. The other issue is whether or not they have any real power to sway OEMs in the direction they need them to go to put AMD APUs in a better light. I too, like others here, would love to have a 35W Carrizo in a 14" chassis, full dual channel memory, a 1080p screen, and an SSD. Of course any current options are both limited and downright badly configured in one way or another to really damn the potential of what is actually a pretty nice chip.

 

[The Stilt]

How about this Carrizo bench? Could anyone find the tester to he gets his results here to we discuss?

http://browser.primatelabs.com/geekbench3/4654211

From what i know is a Full locked 3.4Ghz all cores bench, power limit topping the 42W on the AC, a thing that didn't happen in any other notebook test. Memory is Dual-Channel here.

Can The Stlit explain this good bench for us?

Edit: Updated the pick with a comparison with the best GB3 15W bench with the best GB3 35W bench.

This score is around what you expect to see with the maximum cTDP setting (35W base, 42W boost), with DDR-1600 dual channel memory. It is within few percent of the score I can do with these settings, using the reference AMD FX-8800P system.

Even the 35W base / 42W boost cTDP limit allows significantly higher clocks than the default 15W base / 25W boost configuration, it is by no means locked to the maximum frequency (3.4GHz). 35W / 42W is not sufficient to run the CPU cores constantly at their maximum frequency, however the average clocks will remain quite close to the maximum. Yet there will be multithreaded workloads in which the clocks cannot remain at 3.4GHz. In single threaded workloads the default TDP is obviously sufficient to keep a single compute unit at the maximum frequency, unless the GPU is utilized.

FX-8800P needs around 50W power limit in order to run all of the cores at static 3.4GHz frequency, regardless of the workload. If the iGPU is used, then around 75W is needed in order to be able to max the whole chip out (CPU, GPU, NB and MEMCLK at their maximum frequency, constantly).

 

[William Gaatjes]

First solution - when using commonality platforms, DO NOT design them in such a way to cripple your best products. If anything it should be designed around the best performing product in terms of processing capability, with the capability to run lower end processors.

Second solution - Contractual obligations. "Systems built with dual-channel processors must be shipped with memory occupying both memory channels"

Third solution - Design your processors to be less RAM dependent (L3 cache, HBM).

I do think Bristol Ridge is supposed to ship with more stringent cTDP, in that the configurability is limited in specific processors, probably as a way to prevent the Carrizo debacle from happening again.

AMD has essentially screwed themselves over by not imparting some regulation into how their products are used when it comes to OEMs. The other issue is whether or not they have any real power to sway OEMs in the direction they need them to go to put AMD APUs in a better light. I too, like others here, would love to have a 35W Carrizo in a 14" chassis, full dual channel memory, a 1080p screen, and an SSD. Of course any current options are both limited and downright badly configured in one way or another to really damn the potential of what is actually a pretty nice chip.

I agree and i can only hope for this :
When Zen is out and after that AMD has some money to spend, they should negotiate a deal with Quanta Computers and/or Foxxcon to create a laptop range where AMD components look good and not castrated.
I stopped reading about a given laptop when i read that it has a TN panel...

https://en.wikipedia.org/wiki/Quanta_Computer

https://en.wikipedia.org/wiki/Foxconn

 

[beginner99]

I think the analysis is wrong.
The purpose of carizo is to move some old gf 28nm capacity nobody else wants.
Its about safeguarding mubadala investment in gf.
Nomatter what happens it needs to move it.
Lowend always sell. Its safe strategy. Low risk.

Look what is awailable to amd (effectively mubadala)
A crappy arch intended for everything else but this. But its the only option.

It wouldnt matter if it was dual channel. Still weak. Its not something you build a brand on.

What is going to happen is a lot of dirt cheap 28nm capacity dropped on segment that would otherwise have kabini soc.
From that perspective what is not to like as a consumer?
Its this or kabini/atom

The capacity is there. Its going to get produced no matter what - its politics and oil money. Prices will just go way down. For those segments single channel is fine. So is 15w. They are both requirements.

Even if this is true, user buys laptop, laptop runs like crap, users sees red AMD sticker on laptop. User never buys laptop with said sticker again.

This is the main issue at hand here. So yeah allowing Carrizo single-channel designs was a huge mistake. loosing to intel 5200u/6200U in graphics load is an utter joke.

 

[Abwx]

So yeah allowing Carrizo single-channel designs was a huge mistake.

Put a second RAM module and it will be dual channel.

 

[William Gaatjes]

Put a second RAM module and it will be dual channel.

Most people do not know and read reviews. Reviews tested with a single RAM module.
And trying to get a second Ram module with similar timing specs can be a problem. Marketing thinking dictates that it will be sold by the laptop vendor. For a premium of course.
Then trying to get a pair of modules from another brand (and perhaps cheaper) that work may also be a problem when there is no list of approved RAM modules. Especially when the laptop vendor sells RAM modules.
That second RAM module should be present from the beginning. With a single RAM module, AMD is always identified as the low budget cpu/gpu for a craptop.

 

[The Stilt]

Put a second RAM module and it will be dual channel.

I think he refers to the fact that Carrizo-L and Carrizo / Bristol Ridge / Stoney Ridge are package compatible. Carrizo-L & Stoney Ridge have the two memory slots connected in parallel, since there is only a single channel memory controller available. On Carrizo & Bristol Ridge the memory slots must be connected to their own signals for the dual channel memory to work.

It is completely idiotic that while the package itself is identical, you still cannot use the same motherboard on both of them. Or you can, but such design would be pretty expensive. A motherboard which supports both Carrizo-L / Stoney Ridge & Carrizo / Bristol Ridge fully must have three memory slots:

Two slots which are connected in parallel (share MA signals) and single slot which is connected to MB signals (not available on Carrizo-L and Stoney Ridge). On Carrizo-L & Stoney Ridge the memory modules would be installed to the slot(s) connected to MA signals and on Carrizo & Bristol Ridge one module would be installed to slot connected to MA signals and another one to the slot connected to MB signals.

MA = Memory Channel A
MB = Memory Channel B (only available on Carrizo & Bristol Ridge)

MA + MB = 128-bit bus / dual channel.

EDIT: Or alternatively you could just design the motherboard for Carrizo / Bristol Ridge and connect the two slots for MA & MB signals. That way Carrizo & Bristol Ridge compatibility would be guaranteed, however on Carrizo-L and Stoney Ridge the system would be restricted to a single memory module. The single module limitation wouldn´t be a major issue either, since DDR3 SO-DIMMs are available in 8GB size and DDR4 in 16GB size. The only issue is that the manufacturers could no longer dump their 2GB & 4GB module stocks to these systems :sneaky:

 

[BigDaveX]

Put a second RAM module and it will be dual channel.

Out of the four Carrizo notebooks in the review, one of them (the Toshiba) only had one memory slot full-stop, and another one (the Lenovo) had two slots, but both connected to the same memory channel. The manufacturers are being so cheap with AMD's line-up that they're shaving off the costs of the PCB space and connectors for a second memory channel.

 

[USER8000]

That's a great find.
Take a look at the board shots of the AT review vs the gaminglaptopsjunky review.

Besides the obvious differences that Ian noted such as no WiFi or Battery, you can also see the partial paint job on the heatsink assembly.
Additionally, examine the PCB itself between the DIMM slots and the HDD. The AT picture is pretty bad, but the different silkscreen with the missing white squares on the PCB is obvious. Less obvious is that you can see many traces running in the area by the edge of the board just below the dashed copper line on the AT PCB, but on the GLJ review picture there is just a single wide trace just below that dashed line at board edge.
The PCB itself is a different revision.

Given the general unfinished state of it, the single channel issues, and the different PCB, I almost wonder if AMD just provided Ian with a early rev or prototype of a Carrizo Y700 they had laying around.

LOL,the Y700 is probably the most premium AMD Carrizo lappy as it has a 1080P IPS panel which is Freesync validated:

http://hexus.net/media/uploaded/2015/12/42f0ff6d-4ad5-42b9-acc6-c20b72504211.PNG

Yet,AMD sends out a faulty non-production version with dual channel memory not working??

Plus that LCD display does not look a cheap part to make or validate!

 

[The Stilt]

I don´t think AMD has sent anything to the review. At best they would have asked Lenovo to send one. If these 3rd party system infact came directly from AMD, then someone will get their tails slapped for that. I also would imagine AMD would have sent their reference system to the review, after all it is fully validated design.

 

[Abwx]

I think he refers to the fact that Carrizo-L and Carrizo / Bristol Ridge / Stoney Ridge are package compatible. Carrizo-L & Stoney Ridge have the two memory slots connected in parallel, since there is only a single channel memory controller available. On Carrizo & Bristol Ridge the memory slots must be connected to their own signals for the dual channel memory to work.

It is completely idiotic that while the package itself is identical, you still cannot use the same motherboard on both of them. Or you can, but such design would be pretty expensive. A motherboard which supports both Carrizo-L / Stoney Ridge & Carrizo / Bristol Ridge fully must have three memory slots:

Two slots which are connected in parallel (share MA signals) and single slot which is connected to MB signals (not available on Carrizo-L and Stoney Ridge). On Carrizo-L & Stoney Ridge the memory modules would be installed to the slot(s) connected to MA signals and on Carrizo & Bristol Ridge one module would be installed to slot connected to MA signals and another one to the slot connected to MB signals.

MA = Memory Channel A
MB = Memory Channel B (only available on Carrizo & Bristol Ridge)

MA + MB = 128-bit bus / dual channel.

EDIT: Or alternatively you could just design the motherboard for Carrizo / Bristol Ridge and connect the two slots for MA & MB signals. That way Carrizo & Bristol Ridge compatibility would be guaranteed, however on Carrizo-L and Stoney Ridge the system would be restricted to a single memory module. The single module limitation wouldn´t be a major issue either, since DDR3 SO-DIMMs are available in 8GB size and DDR4 in 16GB size. The only issue is that the manufacturers could no longer dump their 2GB & 4GB module stocks to these systems :sneaky:

It s a little more complicated than this and that s why i dont believe that there s Carrizo s single channel MBs as they couldnt use all kind of RAM modules with Carrizo-L, dual ranked RAM sticks would be incompatible with this latter APU...

721 pins wouldnt be necessary on AM1 if the RAM modules where simply parraleled, Kabini has 128 pins for the two RAM data buses that are then dispatched on an actual 64 bit data bus within the APU.

It wouldnt make sense to design a specific MB that not only wouldnt support all kind of RAM for Carrizo-L but would also be incompatible with a dual channeled Carrizo, no firm with enginers in their right mind would design something that could be only marginaly used while providing not a single cts saving, quite the contrary since another design would be necessary for dual channeled Carrizo....

 

[Abwx]

Out of the four Carrizo notebooks in the review, one of them (the Toshiba) only had one memory slot full-stop, and another one (the Lenovo) had two slots, but both connected to the same memory channel. The manufacturers are being so cheap with AMD's line-up that they're shaving off the costs of the PCB space and connectors for a second memory channel.

These are pre production samples, you really think that Toshiba would save the cost of a dimm slot with a Satellite FX8800P laptop while implementing said second slot on the Carrizo-L Satellite..??..

http://www.notebookcheck.net/Toshiba-Satellite-C55D-C-10P-Carrizo-L-Notebook-Review.148290.0.html

And here the same Carrizo Toshiba laptop as the one tested by AT but in the regular commercial version :

http://forum.notebookreview.com/threads/toshiba-radius-14-amd-e45dw-c4210.781305/

 

[Dresdenboy]

Frankly, it'll be a very long time (never?) before the general public and even informed people truly trust synthetic benchmarks.

I remember times, where I saw lots of Sandra results here and in many reviews. There is currently a new thread collecting CPU-Z benchmark results, and SuperPi, wPrime, 3DMark are still in broad use.

 

[The Stilt]

It s a little more complicated than this and that s why i dont believe that there s Carrizo s single channel MBs as they couldnt use all kind of RAM modules with Carrizo-L, dual ranked RAM sticks would be incompatible with this latter APU...

721 pins wouldnt be necessary on AM1 if the RAM modules where simply parraleled, Kabini has 128 pins for the two RAM data buses that are then dispatched on an actual 64 bit data bus within the APU.

It wouldnt make sense to design a specific MB that not only wouldnt support all kind of RAM for Carrizo-L but would also be incompatible with a dual channeled Carrizo, no firm with enginers in their right mind would design something that could be only marginaly used while providing not a single cts saving, quite the contrary since another design would be necessary for dual channeled Carrizo....

The memory interface on Jaguar, Mullins / Carrizo-L requires 144-pins due ECC. Stoney Ridge requires 133-pins.

They both can support two dual rank DIMMs even there are 64 data signals.

The FP4 design goes exactly as I said. The designer must choose if they design it to Carrizo / Bristol Ridge or for Carrizo-L / Stoney Ridge, or add the third memory slot. The contacts for the second memory channel are NC on Carrizo-L / Stoney Ridge, so if the designer wants to allow Carrizo-L / Stoney Ridge to use two memory slots, they must be connected together. In this case dual channel will not be available.

 

[Burpo]

Put a second RAM module and it will be dual channel.

From your link..
"One 8 GB RAM module is plugged in onto the motherboard. A second module could be added. Due to a single-channel memory controller adding a second RAM module does not bring a performance gain for the graphics card. Only moving to the fast memory might bring a small performance gain."

 

[Dresdenboy]

Dude, it's right in the review.

Mea culpa. Now I see, what you meant. Yeah, it looks like the reference design was better then. Or this Kaveri system wasn't the worst one in efficency.

 

[Shivansps]

EDIT: Or alternatively you could just design the motherboard for Carrizo / Bristol Ridge and connect the two slots for MA & MB signals. That way Carrizo & Bristol Ridge compatibility would be guaranteed, however on Carrizo-L and Stoney Ridge the system would be restricted to a single memory module. The single module limitation wouldn´t be a major issue either, since DDR3 SO-DIMMs are available in 8GB size and DDR4 in 16GB size. The only issue is that the manufacturers could no longer dump their 2GB & 4GB module stocks to these systems :sneaky:

That was the right thing to do, use 2 slots, and if you use Carrizo-L only one whould work, they could have added a warning about it inside in case they where worried about clients complaining about the non working slot.

I think that wa AMD idea, but hell, AMD should have added a boot block on Carrizo so it whould only boot up with DC... one whould think they should know what OEMs do by now.