AnandtechAMD Carrizo ExcavatorReview

[AtenRa]

Could Ian please confirm WHEN they went to Austin to test the Laptops ?? what date ??

 

[MrTeal]

Its more like AMD marketing has backfired:

http://www.notebookcheck.net/HP-Pavilion-15-ab052ng-Notebook-Review.152289.0.html

They even supplied a SODIMM so that Notebookcheck could test a 15W TDP Carrizo system in dual channel.

Yet,now all the systems they had on hand have issues running dual channel memory since most appear to be pre-production laptops and AMD is not aware it seems the production ones are fine!!

It wasn't all the systems they had on hand, it was the Lenovo Y700. It was populated with two DIMMs, but both were connected to a single memory controller and functioned in dual channel mode.

The other laptops should be capable of dual channel operation without issue (hopefully), but they were supplied with a single DIMM and thus were tested as supplied. Ian has said a more thorough review of Carrizo itself will be coming with the release of the X4 845.

 

[Dufus]

Yet,now all the systems they had on hand have issues running dual channel memory since most appear to be pre-production laptops and AMD is not aware it seems the production ones are fine!!

So the guy on NBR reviews his personal Y700-15ACZ (presumably working in dual channel according to CPU-z) back in October and several months later AT get a pre-production model to try!

I'm not sure whether there isn't a single/dual channel problem. Looking at several CPU-z validations show same Company board number and BIOS version with 2 DIMM's. Some DIMM brands showing SC some brands DC or maybe the manufacturer is just being overly generic for differences in systems. I give up trying to find out on here though.

 

[Phynaz]

Should had the reviewer at AT done some basic googling that he would had noticed that it s the Intel version that is single channel crippled, instead his article is used as a mean to spread fud on AMD using incorrect and even inverted realities:

The article wasn't about Intel.

You keep bringing up Intel in an AMD thread. Aren't you the one who always accuses Intel fans of derailing AMD threads?

 

[Arachnotronic]

The article wasn't about Intel.

You keep bringing up Intel in an AMD thread. Aren't you the one who always accuses Intel fans of derailing AMD threads?

It's true

 

[USER8000]

It wasn't all the systems they had on hand, it was the Lenovo Y700. It was populated with two DIMMs, but both were connected to a single memory controller and functioned in dual channel mode.

The other laptops should be capable of dual channel operation without issue (hopefully), but they were supplied with a single DIMM and thus were tested as supplied. Ian has said a more thorough review of Carrizo itself will be coming with the release of the X4 845.

So the guy on NBR reviews his personal Y700-15ACZ (presumably working in dual channel according to CPU-z) back in October and several months later AT get a pre-production model to try!

I'm not sure whether there isn't a single/dual channel problem. Looking at several CPU-z validations show same Company board number and BIOS version with 2 DIMM's. Some DIMM brands showing SC some brands DC or maybe the manufacturer is just being overly generic for differences in systems. I give up trying to find out on here though.

But the problem it is also said the 15w TDP Carrizo chips cannot run dual channel memory,except the Notebook check review has an A10 8700P running at 15W TDP but is running in dual channel with another SODIMM supplied by AMD LAST YEAR. Now AMD is saying that in 15w TDP mode you cannot run dual channel RAM.

The IGP results indicate the RAM is running in dual channel too.

Plus the Y700 mistake is very serious as it is the most premium Carrizo laptop out there with a 1920x1080 IPS display which supports FreeSync.

 

[MrTeal]

But the problem it is also said the 15w TDP Carrizo chips cannot run dual channel memory,except the Notebook check review has an A10 8700P running at 15W TDP but is running in dual channel with another SODIMM supplied by AMD LAST YEAR. Now AMD is saying that in 15w TDP mode you cannot run dual channel RAM.

The IGP results indicate the RAM is running in dual channel too.

Plus the Y700 mistake is very serious as it is the most premium Carrizo laptop out there with a 1920x1080 IPS display which supports FreeSync.

Where do you see AMD saying you can't run dual channel @ 15W? You can't run 2133, but that's not the same thing.

 

[coercitiv]

all i read are speculations but not a single commercial model is given as exemple

I already gave you an example: Dell Inspiron 15 5555 is listed as single channel even with with mutiple dimms installed.

 

[The Stilt]

Core i5 2400S:   35945 samples/s/GHz/core
15/25W/DDR-1600: 34475 samples/s/GHz/module
35/42W/DDR-2133: 34030 samples/s/GHz/module

Just ran this (2.76b) on Core i5-6200U (9m 15s) and FX8350 (5m 40s):

Core i5-6200U:   69189 samples/s/GHz/core
FX8350:    37187 samples/s/GHz/module

I don't think the sustained clockspeeds on the Carrizo machines are correct. Excavator should have higher IPC than Piledriver.

You´re right, these are not proper numbers :\

Like I said, I wasn´t able to install any display driver under Ubuntu. After switching back to Windows I noticed something which I had forgotten. On Steamroller / Carrizo APUs the NB frequency will be locked to the minimum PState when display driver is not installed. On Carrizo it means that the NB is locked to 700MHz instead of being able to reach 1.3GHz.

In Windows Blender 2.76 appears to be completely bugged. It doesn´t seem to be able to utilize the cores properly no matter what configuration is used. The performance is extremely bad, the clocks stay pretty low and the cooling is barely inaudible despite 42W TDP limit. In this system the cooling is never inaudible in the first place.

 

[superstition]

You're at least the third person to come out and say this, and of course, you're correct. That would have solved a lot of problems.

But I've already gone on too much about what they should have done with Carrizo (and/or what they should do with mobile Bristol Ridge). Either they wise up or they don't . . . everything they need to know has already been said.

Maybe AMD is controlled by Intel more directly than just through OEMs.

 

[Abwx]

I already gave you an example: Dell Inspiron 15 5555 is listed as single channel even with with mutiple dimms installed.

That s the only exemple, and beside the same page is used for both Carrizo and Carrizo-L, not sure that marketing guys are aware of the subtilities between the two chips, isnt it :

Dell Inspiron 15" Laptop with AMD A10 Quad Core (i5555-2843SLV)
Processor: AMD A10-8700P APU
Memory: 12GB Dual Channel DDR3L 1600MHz (8GBx1 + 4GBx1)

https://www.theshoppingchannel.com/...oductdetails?nav=R:648072,N:131339,E:11926287

Laptop DELL Inspiron 5555 Linux

DELL Inspiron 5555 AMD Quad-Core A10-8700P
404,26 €

8192MB Dual Channel 1600MHz (4GBx2)

http://order.komaks.com/dell-inspiron-5555-5397063762460.html

So, what is your conclusion now..?..

That i did the homework that you neglected doing because it could had
killed your "exemple" s "relevancy"..?.

 

[The Stilt]

Dell Inspiron 5x55 systems have motherboards which are designed for both Carrizo-L & Carrizo. It can use two memory slots with Carrizo-L, meaning the system is actually restricted to single channel on Carrizo too :sneaky:

The bios updates released by Dell all contain multiple images, for the same motherboard design. One for "CarrizoLite" (Carrizo-L) and one for Carrizo.

Geekbench memory results clearly show that the system is running in single channel, despite having two memory modules installed.

At DDR-1600 dual channel (+ DR modules), 15W TDP Carrizo scores ~2000 (ST) & 2900 (MT) in memory test.

http://browser.primatelabs.com/geekbench3/search?utf8=✓&q=Inspiron+5555

 

[Abwx]

Dell Inspiron 5x55 systems have motherboards which are designed for both Carrizo-L & Carrizo. It can use two memory slots with Carrizo-L, meaning the system is actually restricted to single channel on Carrizo too :sneaky:

The bios updates released by Dell all contain multiple images, for the same motherboard design. One for "CarrizoLite" (Carrizo-L) and one for Carrizo.

Geekbench memory results clearly show that the system is running in single channel, despite having two memory modules installed.

At DDR-1600 dual channel (+ DR modules), 15W TDP Carrizo scores ~2000 (ST) & 2900 (MT) in memory test.

http://browser.primatelabs.com/geekbench3/search?utf8=✓&q=Inspiron+5555

The MB have different references..

http://browser.primatelabs.com/geekbench3/compare/5095163?baseline=5006460

 

[dark zero]

I already gave you an example: Dell Inspiron 15 5555 is listed as single channel even with with mutiple dimms installed.

What a retarded way to ruin an uARCH... AMD did screwed BIG time with their failure called Carrizo-L

 

[Phynaz]

The MB have different references..

http://browser.primatelabs.com/geekbench3/compare/5095163?baseline=5006460

You're going to use a website that says the memory is running at 1mhz as some kind of authority?

 

[The Stilt]

What a retarded way to ruin an uARCH... AMD did screwed BIG time with their failure called Carrizo-L

The only screw up from AMD with FP4 was that they underestimated the greed of the system manufacturers. The idea of a single package for multiple designs is just great (for this type of products), however there should be a way to ensure that the higher-end design always gets served first.

If you leave a back-door to the design which allows the ODM to save a penny in their design, these people will exploit it without even blinking. No matter how harsh the performance penalty is or how tiny the cost saving is. The ODMs are like pikeys, never let one out of sight and hold on your wallet D:

 

[The Stilt]

The AMD reference system I used has been pretty neglected too. Since the first stable bios release there has not been any further releases, which would update the coprocessor (SMU, PMU, PSP) firmwares / AGESA, VBIOS or even the CPU microcode. Since there has been several updates (fixing various bugs and errata) after the bios was compiled, I went ahead and replaced all of the individual components myself. Everything on these things rely completely on software and if the SMU firmware for example has bugs or issues, they definitely can affect the performance. SMU is the master of puppets on all APUs / SoCs never than Richland (Steamroller, Excavator, Kabini, Mullins) and NOTHING work without it. The APU is a brick, a piece of silicon and substrate without it.

Updating the components improved the system behavior slightly and seemed to give some additional performance.

Before I was able to enjoy from my newly cooked bios, I had to do some additional code though. The laptop uses AMI Aptio V UEFI bios, which means that the bios is RSA2048 protected. While the unmodified bioses can be written just fine by using the standard tools, that´s obviously not the case with modified bioses. The internal tools which allow you to override the protections do not support Aptio V which meant I had to find another way around. The two options would have been using an ISP programmer or Coreboot based Flashrom. Obviously having the ISP connector present on the motherboard was rather wishful thinking, so I had to go with Flashrom. Unfortunately Flashrom didn´t support Carrizo at all... before I patched in the support for all Excavator based designs :sneaky:

Anyway the system is now updated and runs on a clean and fully updated (excl. the telemetry updates) Windows 7 x64.

Did few 3DMark runs on various settings.

Note: I use two DDR-2133 rated dual rank modules from Kingston. On recent APUs (=> Steamroller) dual rank modules provide roughly one speed class (e.g. 2133MHz vs. 1866MHz) advantage over single rank modules, at the same frequency. This means that running dual rank modules at 2133MHz provides almost exactly the same bandwidth as DDR-2400 with single rank modules.

Drivers: Crimson 16.1.1 Hotfix

Default 15/25W TDP configuration - DDR-1600

3DMark Cloud Gate: 5840
3DMark Sky Diver: 4757
3DMark Fire Strike: 1346

cTDP 35/42W TDP configuration - DDR-1600

3DMark Cloud Gate: 6599
3DMark Sky Diver: 5557
3DMark Fire Strike: 1513

cTDP 35/42W TDP configuration - DDR-2133

3DMark Cloud Gate: 7163
3DMark Sky Diver: 6040
3DMark Fire Strike: 1708

 

[AtenRa]

cTDP 35/42W TDP configuration - DDR-2133

3DMark Cloud Gate: 7163
3DMark Sky Diver: 6040
3DMark Fire Strike: 1708

Hell, i get the same score with the A10-7870K 95W TDP on a Gigabyte mini-Itx board.

 

[The Stilt]

Hell, i get the same score with the A10-7870K 95W TDP on a Gigabyte mini-Itx board.

Yeah, Carrizo has around 35% higher relative bandwidth for the GPU at the same MEMCLK than Kaveri / Godavari.

 

[AtenRa]

Yeah, Carrizo has around 35% higher relative bandwidth for the GPU at the same MEMCLK than Kaveri / Godavari.

65W TDP BristolRidge with 2400MHz DDR-4 could be 20-30% faster than 95W TDP 2133MHz Kaveri then , in iGPU performance.

 

[The Stilt]

65W TDP BristolRidge with 2400MHz DDR-4 could be 20-30% faster than 95W TDP 2133MHz Kaveri then , in iGPU performance.

On AM4 definitely. Despite the improved compression algorithm, the fastest BR part on AM4 will be more bandwidth bottle necked than ever... AMD went bat s*it crazy with the GPU clock on AM4 BR.

 

[AtenRa]

Im pretty sure no OEM will ever make a Laptop with DDR-3 at 2133MHz since this is not a jedec standard.

On the other hand, DDR-4 2133MHz is a jedec standard so im expecting some BR Laptop models to have at least a single 2133MHz memory dimm.

And all of this really show the incompetence of those inside AMD that involved for the AT review.
Really a 35W TDP Carrizo Laptop with dual channel 2133MHz memory would be extremely competitive if not faster than any Intel 15W TDP paired with NVDIA dGPU at the same price and power consumption. Not to mention with lower BOM (single APU vs CPU + dGPU) thus cheaper to produce = higher margins.

 

[The Stilt]

A proper Carrizo laptop isn´t exactly cheap to build. It´s not THAT expensive, but a properly built system cost will exceed the price range which Carrizo was designed for (<600$ IIRC). IMO already at 700$ it is already questionable if there is any point in Carrizo. Regardless of the configuration. In my own use I need much more CPU performance than top notch GPU performance (excl. video features). Since on Carrizo the CPU performance is just not there, I would most likely choose a Skylake-U based system.

In Europe Lenovo Y700-15ISK (i5-6300HQ, 8GB, GTX 960M, 1TB HDD, 1080P IPS) can be had for 899&#8364;, which is 20&#8364; more than the equivalent Y700 variant with Carrizo FX-8800P in it. Needless to say the Skylake system for 20&#8364; more will just annihilate the Carrizo system

 

[cbn]

In Europe Lenovo Y700-15ISK (i5-6300HQ, 8GB, GTX 960M, 1TB HDD, 1080P IPS) can be had for 899&#8364;, which is 20&#8364; more than the equivalent Y700 variant with Carrizo FX-8800P in it. Needless to say the Skylake system for 20&#8364; more will just annihilate the Carrizo system

When we see component spec differences like that at X price point I have to think that AMD is just not supplying enough chips to the ODM/OEM.

And the resulting "low volume" AMD laptop runs are passing too much of the fixed cost down to the consumer.

Maybe if AMD can increase the volume of chips next round, they will not need processors like Carrizo-L (or in the case of Bristol Ridge, Stoney Ridge) so much to help lower the per unit cost. And thus we may not see as many compromises in the AMD laptop designs. I'm crossing my fingers that happens.

 

[AtenRa]

A proper Carrizo laptop isn´t exactly cheap to build. It´s not THAT expensive, but a properly built system cost will exceed the price range which Carrizo was designed for (<600$ IIRC). IMO already at 700$ it is already questionable if there is any point in Carrizo. Regardless of the configuration. In my own use I need much more CPU performance than top notch GPU performance (excl. video features). Since on Carrizo the CPU performance is just not there, I would most likely choose a Skylake-U based system.

In Europe Lenovo Y700-15ISK (i5-6300HQ, 8GB, GTX 960M, 1TB HDD, 1080P IPS) can be had for 899€, which is 20€ more than the equivalent Y700 variant with Carrizo FX-8800P in it. Needless to say the Skylake system for 20€ more will just annihilate the Carrizo system

Using the APU only without any dGPUs will decrease the cost significantly.

Take a 15W TDP Intel CPU + NVIDIA dGPU (GT 920/930M)

NVIDIA dGPU will cost $100-150 (including the GPU + Memory)
Bigger Heat-sink fan for both CPU + dGPU
Motherboard design and manufacturing cost to include the dGPU is higher

For the Carrizo APU alone your only cost adder is the Dual Channel 2133MHz Memory.

Im pretty sure the APU Carrizo Laptop could be $100-150 cheaper than the Intel 15W TDP + NVIDIA dGPU (for the same performance).