AnandtechAMD Carrizo ExcavatorReview

[superstition]

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.

I'm not sure how it relates to iGPU performance but ExtremeTech's chart showed poor performance for DDR4 2133 with Skylake as compared to 2400 and above with Dirt 3. The 2133 RAM was a bottleneck.

DDR3 2133 should be significantly better due to the lower latency.

(SLI)

 

[cbn]

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).

That is how I think of it also in basic terms (ie, 35W APU replaces 15W CPU + 20W/25W dGPU).

But then none of the high volume laptop makers use DDR3 2133 RAM.

So that would leave only low volume laptop makers (or perhaps some kind of white box or barebone) as the only realistic option.

But can these low volume laptop maker/ whitebox/ barebone compete against the higher efficiency of a high volume big name laptop maker (using more expensive parts including dGPU)?

I think it would be very difficult.

P.S. Maybe if the 35W APU barebones was shipped so Windows 10 would activate with a 3rd party supplied SSD that would help though (technically this should be possible as long as Windows 10 was activated to the barebones laptop prior to leaving the factory).

 

[superstition]

Given the maturity of DDR3 it shouldn't be difficult to get 2133 RAM, eh?

 

[cbn]

Given the maturity of DDR3 it shouldn't be difficult to get 2133 RAM, eh?

2 x 4GB DDR3 2133 SO-DIMMs are $21 more than 2 x 4GB DDR3 1600 SO-DIMMs at Newegg.

($54 vs $33).

So 64% more expensive.

In contrast, 2 x 4GB DDR4 2133 SO-DIMMs start at $42.

 

[AtenRa]

I'm not sure how it relates to iGPU performance but ExtremeTech's chart showed poor performance for DDR4 2133 with Skylake as compared to 2400 and above with Dirt 3. The 2133 RAM was a bottleneck.

DDR3 2133 should be significantly better due to the lower latency.

(SLI)

Im talking about the iGPU, not the CPU + dGPU like the test above.

BristolRidge iGPU with 2133MHz DDR-4 will be as fast as with DDR-3. But it seams that 35W TDP BristolRidge support up to 2400MHz DDR-4.

Since 2133MHz DDR-4 is a jedec standard, OEMs will use at least a single module in Laptops unlike DDR-3 that official jedec standard is up to 1600MHz.

 

[cbn]

I wonder at what point does the Carrizo become cheap enough that maybe some OEM/ODM decides to make a 35W barebone or white box type dual channel laptop with some of the leftover chips?

Maybe they can supply the SSD and then let us the consumers decide on the RAM? (Or maybe make it activate with Windows 10 even if not shipped with SSD)

35W Carrizo would be slower than 35W Bristol Ridge (re: lower clocks and DDR3 2133 rather than DDR4 2400), but if the price was right maybe it would be interesting to the enthusiast community while also being the basis for a 35W Bristol laptop design later on as well?

P.S. Maybe even ship the barebones without a battery too, but make it compatible with some existing battery design (if such a strategy is possible).

 

[DrMrLordX]

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

Ummmm maybe? It wouldn't be hard for Intel to do that, considering how cheap is their stock price and all that. But you know, that would probably show up on their books somewhere. So. I dunno?

 

[ShintaiDK]

Ummmm maybe? It wouldn't be hard for Intel to do that, considering how cheap is their stock price and all that. But you know, that would probably show up on their books somewhere. So. I dunno?

No, just no...

 

[mrmt]

Ummmm maybe? It wouldn't be hard for Intel to do that, considering how cheap is their stock price and all that. But you know, that would probably show up on their books somewhere. So. I dunno?

Why is it harder for you to accept AMD management is utterly incompetent or that there isn't much value in AMD offers to OEM than to believe in a pie-in-the-sky conspiracy involving Intel?

 

[cbn]

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 Fire Strike: 1346

cTDP 35/42W TDP configuration - DDR-1600

3DMark Fire Strike: 1513

cTDP 35/42W TDP configuration - DDR-2133

3DMark Fire Strike: 1708

Firestrike scores (from various processors) I gleaned from notebookcheck for comparison:

Intel Iris graphics 540 (15W Skylake GT3e): 1395 ---> http://www.notebookcheck.net/Intel-Iris-Graphics-540.149939.0.html

Intel Iris Pro 6200 (47W Broadwell GT3e): 1685 ---> http://www.notebookcheck.net/Intel-Iris-Pro-Graphics-6200.125593.0.html

GT 940M (64 bit DDR3 2000): 1424 ---> http://www.notebookcheck.com/NVIDIA-GeForce-940M.134809.0.html

GT 940M with GDDR5 (64 bit memory bus): ---> 1866 http://www.notebookcheck.com/NVIDIA-Maxwell-GPU-940M-GDDR5.153639.0.html

So the FX-8800P with dual channel DDR3 2133 (dual rank) at 42W boost is in between GM108 (384 Maxwell v1 cores) with 64 bit DDR3 2000 and GM108 (384 Maxwell v1 cores) with 64 bit GDDR5 5000....and actually closer to the GDDR5 score than it is to the DDR3 2000 score (1424 vs. 1708 vs. 1866).

 

[Abwx]

Firestrike scores (from various processors) I gleaned from notebookcheck for comparison:

Intel Iris graphics 540 (15W Skylake GT3e): 1395 ---> http://www.notebookcheck.net/Intel-Iris-Graphics-540.149939.0.html

Intel Iris Pro 6200 (47W Broadwell GT3e): 1685 ---> http://www.notebookcheck.net/Intel-Iris-Pro-Graphics-6200.125593.0.html

GT 940M (64 bit DDR3 2000): 1424 ---> http://www.notebookcheck.com/NVIDIA-GeForce-940M.134809.0.html

GT 940M with GDDR5 (64 bit memory bus): ---> 1866 http://www.notebookcheck.com/NVIDIA-Maxwell-GPU-940M-GDDR5.153639.0.html

So the FX-8800P with dual channel DDR3 2133 (dual rank) at 42W boost is in between GM108 (384 Maxwell v1 cores) with 64 bit DDR3 2000 and GM108 (384 Maxwell v1 cores) with 64 bit GDDR5 5000....and actually closer to the GDDR5 score than it is to the DDR3 2000 score (1424 vs. 1708 vs. 1866).

That s all good to do the comparisons but The Stilt numbers are achieved on a TDP controled environment while on the other hand we must be content with marketing numbers for the competition, for the record the exemple i bolded in your list is achived with at least 25W permanently if not more given the GPU allocated power, the GPU alone can use about 20W :

So much for the so called "15W" SKLs GPU "scores" ,and btw, very well gleaned number....

 

[cbn]

The GT 940M is listed at 25W according to Notebook check (would be interesting to see how much power it pulls).

P.S. These GT940M are typically paired with an ULV Core i5, so the CPU side on these Nvidia notebooks is strong....but glad to see the 512sp Carrizo iGPU (with dual rank DDR3 2133) doing so well in iGPU (about 20% faster in Firestrike than a GT940M with 64 bit DDR3 2000).

Here is the cheapest ($499.99) GT 940M listing I could find on Newegg:

http://www.newegg.com/Product/Product.aspx?Item=N82E16834315131

15.6" 1920 x 1080, i5-5200U, 8GB RAM, 1TB HDD, GT 940M 2GB GDDR3, Windows 10.

 

[cbn]

Okay, so with DDR3 2133 not being used....what would a DDR3 1600 35W Carrizo A10/A8 (ie, 384sp iGPU APUs) compete against? (FX-8800P @ 35W/42W with DDR3 1600 has a firestrike score of 1513....so how much does having the smaller 384sp iGPU lower the score and capability).

Could it even be that AMD is best off saving all their current 512sp dies for Bristol Ridge (where the DDR4 actually can help the performance of the highest bin iGPU)? And then just use the 384sp dies at 35W dual channel from here on out in some revised fashion? Perhaps with higher CPU clocks? (perhaps even at FX-8800P level CPU clocks for some new 384sp Carrizo SKU)

 

[The Stilt]

Results with 384SP, 35/42W cTDP and DDR-1600:

3DMark Sky Diver: 4897
3DMark Fire Strike: 1325

 

[DrMrLordX]

No, just no...

Why is it harder for you to accept AMD management is utterly incompetent or that there isn't much value in AMD offers to OEM than to believe in a pie-in-the-sky conspiracy involving Intel?

Did I pick a scab open or something? All I said was, "maybe". AMD's market cap is tiny, Intel or Apple (or a number of other players) could pick up a controlling share with the money they have between the cushions of their respective couches. It's just an idea superstition floated. If you really dislike it that much, go ask him.

And, to repeat, if Intel did it, it would show up on their books somewhere, so . . . I don't know about that.

Back to Carrizo: Stilt, can you post the graphics/physics components of 3DMark Firestrike from your DDR3-2133 run? Your total score is only ~100 points lower from my highest results with a 7700k @ 4.7 GHz with a 1028 mhz iGPU.

 

[mrmt]

Did I pick a scab open or something? All I said was, "maybe". AMD's market cap is tiny, Intel or Apple (or a number of other players) could pick up a controlling share with the money they have between the cushions of their respective couches. It's just an idea superstition floated. If you really dislike it that much, go ask him.

That would show >>on AMD's books<<.

 

[The Stilt]

Back to Carrizo: Stilt, can you post the graphics/physics components of 3DMark Firestrike from your DDR3-2133 run? Your total score is only ~100 points lower from my highest results with a 7700k @ 4.7 GHz with a 1028 mhz iGPU.

Aren´t they visible when you click the link?

3DMark Score: 1708

Graphics Score: 1903

Physics Score: 4316

Combined Score: 639

Graphics Test 1: 8.99 fps

Graphics Test 2: 7.67 fps

Physics Test: 13.7 fps

Combined Test: 2.98 fps

 

[Sweepr]

Firestrike scores (from various processors) I gleaned from notebookcheck for comparison:

Iris Pro 6200 scores 1742. If we believe Fire Strike then 512 SPs Carrizo @ 15-20W with dual-channel DDR3-1600 would be close to 30% faster than HD 520 in games, but in reality NBC shows Intel is faster in most games @ 1366x768 Medium/Normal quality.

 

[DrMrLordX]

Aren´t they visible when you click the link?

I totally missed the link for some reason. Woops!

Aaaanyway, my physics score is 5519 in one of my old runs (total: 1750). The graphics score is the same. On newer drivers I've gotten as high as ~1800 total.

 

[frozentundra123456]

Did I pick a scab open or something? All I said was, "maybe". AMD's market cap is tiny, Intel or Apple (or a number of other players) could pick up a controlling share with the money they have between the cushions of their respective couches. It's just an idea superstition floated. If you really dislike it that much, go ask him.

And, to repeat, if Intel did it, it would show up on their books somewhere, so . . . I don't know about that.

Back to Carrizo: Stilt, can you post the graphics/physics components of 3DMark Firestrike from your DDR3-2133 run? Your total score is only ~100 points lower from my highest results with a 7700k @ 4.7 GHz with a 1028 mhz iGPU.

Maybe people are just tired of unproven accusations, conspiracy theories, and constant rehashing of things that went on years ago. Naw, that couldnt be it.

 

[cbn]

Results with 384SP, 35/42W cTDP and DDR-1600:

3DMark Sky Diver: 4897
3DMark Fire Strike: 1325

That is a good score.

In fact, it is higher than the Firestrike score Anandtech got for the A8-7600 with DDR3 2133 RAM:

http://www.anandtech.com/show/7677/amd-kaveri-review-a8-7600-a10-7850k/14

P.S. A10-8700P has the same iGPU specs (but a bit weaker CPU at 1.8/3,2 vs 2.1/3.4). So it would have been interesting to see how a 35W version of that with dual channel 1600 (single rank or dual rank) compared to A8-7600 both at 65W and 45W in games. (With A8-7600 @ 45W throttling down to 2.4 Ghz with heavy iGPU use A10-8700P @ 35W/42W with dual channel 1600 probably would have done well)

Too bad none of OEMs were able to make the laptops.

 

[Flash831]

Could it be possible AMD wanted the OEM's to pair Carrizo with single rank memory and low end components?

That way it would be easier to take the chip on a respin (Bristol Ridge) and it would seem like a solid upgrade from Carrizo... (As long as Bristol Ridge gets a fair chance by the OEM's. 35w, dual rank memory, etc). :sneaky:

After all, AMD bought far less wafers in 2015 compared to 2014. They need to get the max ROI, and perhaps this is a way to extend the chips lifetime.

 

[cbn]

If we believe Fire Strike then 512 SPs Carrizo @ 15-20W with dual-channel DDR3-1600 would be close to 30% faster than HD 520 in games, but in reality NBC shows Intel is faster in most games @ 1366x768 Medium/Normal quality.

Firestrike doesn't appear to put much load on the CPU so I would expect there is more juice going to the iGPU in that benchmark than we would normally see in actual gameplay.

In fact, look at Stilt's GTA V gameplay comparison where the FX-8800P iGPU @15/25W is only clocking about 350Mhz (average) on the iGPU with the CPU @ ~1.5 Ghz (average).

(Averaging only 1.5 GHz for four excavator cores and 350 Mhz for the 512sp is not a lot of CPU or GPU performance.)

http://forums.anandtech.com/showpost.php?p=37646027&postcount=2351

15W is optimal if either both are loaded but results in the cpu clock suffering mostly.
At 35W you can run the cpu portion at 2.1GHz rather than ~1.1GHz. And run the GPU at 600-800MHz rather than 500MHz. So yes in terms of efficiency it's less good than at 15W but it would make the difference between playing games smoothly or being hit by a cpu bottleneck.

If it can meat or beat a 50W Intel + Nvidia setup with 35W it's still very efficient. (Let's say 15-20W for the Intel CPU + ~32W for the 840M)

I think they could match that Intel + Nv config going by the results Stilt posted for GTA V

15-25W:

35-42W:

 

[DrMrLordX]

Maybe people are just tired of unproven accusations, conspiracy theories, and constant rehashing of things that went on years ago. Naw, that couldnt be it.

Some folks are also tired of the proven accusations, about which little to nothing was done. Probably why superstition mentioned it in the first place. Though you could always go ask him, if you really wanted to know . . .

 

[ElFenix]

That would show >>on AMD's books<<.

intel taking an ownership stake in AMD would show in intel's balance sheet as an asset, so it would be on intel's books. intel would show up in AMD's stock register, but i'm not sure a lot of lay people consider a listing of shareholders as 'the books'