AMD Carrizo Pre-release thread

[mikk]

Based on the 3dmark11 scores it could be a 35W cTDP which I think it is. This is what I meant, with AMDs huge cTDP variance this is a mess.

 

[monstercameron]

Based on the 3dmark11 scores it could be a 35W cTDP which I think it is. This is what I meant, with AMDs huge cTDP variance this is a mess.

Why is it a mess? Their test unit is the pavilion 17 which has a large chassis and is probably easier to cool.

 

[mikk]

Why is it a mess? Their test unit is the pavilion 17 which has a large chassis and is probably easier to cool.

This is not the point.

 

[monstercameron]

This is not the point.

Is the point that even Intel uses a ctdp and does the exact same thing yet it is a mess for amd?

Trying not to be passive aggressive but what do you think is the point?

 

[slaven]

I compared results of Cinebench 11.5 from A12-8800p ( https://semiaccurate.com/forums/showpost.php?p=240581&postcount=911 ) and concluded, that a10-8700p was configured to 15 W (even not 15W PPT).
Also, being compared to Kaveri's results, they are closer to FX-7500p (19 Wt) rather than FX-7600p (35W).
Finally, testers says, that " E.g., the GPU can score between 1764 points (15 Watts) to 2359 points (+34% with 35 Watt cTDP) in 3DMark 11 (P). ", and the tested sample scored 1510.

And thank you for graphical results - it was too hard to find them for me )))

 

[Abwx]

In Cinebench ST the CPU is running at 2.8GHz, wich seems to me dubbious.

MT scores suggest that they are peformed at barely 15W while the GPU score imply about 18-20W.

SA s Thenevin got those ST CB11.5 scores when setting the chip TDP values in the bios :

15W = 0.95
20W = 0.98

At 20W the frequency stays at static 3.4GHz in ST workloads.
Meaning in this test the IPC is 9.1% greater than on Steamroller.

So HP laptop chip seems to be configured at 15W PPT, wich means 20W for short periods and 15W the rest of the time; ST tests are not correct and indeed NBC has made a same mistake when they first measured Kaveri s CB scores, seems a carbon copy of said test that they updated later...

 

[Abwx]

Based on the 3dmark11 scores it could be a 35W cTDP which I think it is. This is what I meant, with AMDs huge cTDP variance this is a mess.

You should definitly stop with your forever unsubstancied blank statements that amount to fud in this case, as often..

AMD s power management implementation is very accurate according to the tester who provided the CB scores/watt/frequency, if you want to contradict this opinion bring facts, but i guess that it would be a hard pressure...

 

[mikk]

Is the point that even Intel uses a ctdp and does the exact same thing yet it is a mess for amd?

Intel doesn't do the exact same thing. Did you see a 15W ULV SKU turning into a 35W SKU? Or a 37W SKU turning into a 15W ULV SKU? On Intel this is much more restricted.

Trying not to be passive aggressive but what do you think is the point?

The point is you have a notebook and don't know how is it running/configured, as 15W or 35W. Nobody told 35W is a mess for a big 17" notebook assuming it is running at 35W which we don't know.

 

[Abwx]

Intel doesn't do the exact same thing. Did you see a 15W ULV SKU turning into a 35W SKU? Or a 37W SKU turning into a 15W ULV SKU? On Intel this is much more restricted.



The point is you have a notebook and don't know how is it running/configured, as 15W or 35W. Nobody told 35W is a mess for a big 17" notebook assuming it is running at 35W which we don't know.

.

Up to 46W for a laptop with a "15W" Intel CPU..?..

http://www.notebookcheck.com/Test-HP-Elitebook-820-G2-Subnotebook.144399.0.html

How is it possible, must be the 12"5 screen, , Intel cTDP can only work as marvel to get this chip deserve its i7 denomination..

Another exemple perhaps..? :

http://www.notebookcheck.com/Test-Lenovo-ThinkPad-X250-Ultrabook.136036.0.html

44W, no doubt, and that correlates, those 12"5 screen are power hogs, and of course to help dissipate any statistical mistake we have a confirmation of the above with a third exemple :

http://www.notebookcheck.com/Test-Dell-Latitude-12-E7250-Ultrabook.135450.0.html

And a big lol about your post that show being total non sense in respect of reality, all you are posting are myths.

 

[mikk]

.

And a big lol about your post that show being total non sense in respect of reality, all you are posting are myths.

You have described yourself. We are talking about SoC TDP levels. Notebookcheck doing their measurement on the power supply from the whole notebook. What AMD does with their cTDP on Carrizo reaches a new dimension. Once again, on Intel this is much more restricted. Calling this a myth is a big lol. You will do yourself a favour by doing a reality check. I'm sure it would help you a bit sometimes.

 

[Abwx]

Once again, on Intel this is much more restricted. Calling this a myth is a big lol. You will do yourself a favour by doing a reality check. I'm sure it would help you a bit sometimes.

Reality check say that Intel is using the same chip for TDPs extending at least for 4.5W to 28W officialy, AMD start at 12W and end at 35W.

Anyone with some elementary sense would notice that the former S TDP range is rougly 4 while the latter is barely 3, so much about your new dimension, not only non sense but in complete opposition with numbers, i guess that you are talking complexe ones, specialy of the imaginary parts but for the time you should at least try mastering reals ones...


As for thermal management in laptops Intel has a range that exceed 10 while AMD is below 2, here the peaks allowed for each setting choosen by OEMs :

12W TDP = 20W (PPT xC Min)
15W TDP = 25W (PPT xC Min)
25W TDP = 42W (PPT xC Min)
35W TDP = 42W (PPT xC Min)

PPT is package power tracking

Now provide us the numbers for Intel, although i already know them, but it s for the curiosity..

I compared results of Cinebench 11.5 from A12-8800p ( https://semiaccurate.com/forums/showpost.php?p=240581&postcount=911 ) and concluded, that a10-8700p was configured to 15 W (even not 15W PPT).
Also, being compared to Kaveri's results, they are closer to FX-7500p (19 Wt) rather than FX-7600p (35W).
Finally, testers says, that " E.g., the GPU can score between 1764 points (15 Watts) to 2359 points (+34% with 35 Watt cTDP) in 3DMark 11 (P). ", and the tested sample scored 1510.

And thank you for graphical results - it was too hard to find them for me )))

The three Carrizo SKUs should perform very close CPU wise when all three are restricted at 15W and GPU wise the two lower parts should be on par.

NBC numbers suggest that there s tests that were done while the laptop was very hot, particularly the 3DMark and CB ST test.

For this latter there s no way that the MT score could be 2.36 and that the chip would be unable to get higher than 2.8GHz in ST in both CB 11.5 and CB R15.

 

[slaven]

Average advantage of A10-8700p over FX-7500 is 20%. But as compared to Broadwell i3-5005U, new APU is ~2,5% slower.

 

[slaven]

del me

 

[slaven]

Spoiler

                                                                A10-8700p FX-7500 3-5005U
                                                        TDP     15 W    19 W+   15 W + SB
                                                                Carrizo Kaveri  Broadwell                    Carrizo/Kaveri      Carrizo/Broadwell
Octane V2 - Total Score
                                                        Points  8324    8302    9551                            100,26%         87,15%

Cinebench R15 - CPU Single 64Bit
                                                        Points  69      55      82                              125,45%         84,15%

Cinebench R15 - CPU Multi 64Bit
                                                        Points  194     153     208                             126,80%         93,27%

Mozilla Kraken 1,1 - Total Score
                                                        ms      3488,1  4430    3339,4
                                                                                                                127,00%         95,74%

Geekbench 3 - 32 Bit Multi-Core Score
                                                        Points  4775    4116    4016                            116,01%         118,90%

Geekbench 3 - 32 Bit Single-Core Score
                                                        Points  2006    1769    1912                            113,40%         104,92%

Geekbench 3 - 64 Bit Multi-Core Score
                                                        Points  5105    4193    4113                            121,75%         124,12%

Geekbench 3 - 64 Bit Single-Core Score
                                                        Points  2170    1720    1976                            126,16%         109,82%

3DMark (2013) - Fire Strike Standard Physics 1920x1080
                                                        Points  2685    2365,5  3250                            113,51%         82,62%

3DMark (2013) - Cloud Gate Standard Physics 1280x720
                                                        Points  2007    1721    2289                            116,62%         87,68%

3DMark (2013) - Ice Storm Unlimited Physics 1280x720 offscreen          (Kaveri = estimate)
                                                        Points  22946   20016,8 29332                           114,63%         78,23%

3DMark (2013) - Ice Storm Extreme Physics 1920x1080                     (Kaveri = estimate)
                                                        Points  22721   19820,6 25796                           114,63%         88,08%

3DMark (2013) - Ice Storm Standard Physics 1280x720
                                                        Points  22740   19978   25552                           113,83%         88,99%

Sunspider - 1,0 Total Score
                                                        ms      164,3   177     151,1
                                                                                                                107,73%         91,97%

TrueCrypt - Serpent Mean 100MB
                                                        GB/s    0,188   0,148   0,112                           127,03%         167,86%

TrueCrypt - Twofish Mean 100MB
                                                        GB/s    0,303   0,227   0,192                           133,48%         157,81%

TrueCrypt - AES Mean 100MB
                                                        GB/s    1,5     1,1     1,2                             136,36%         125,00%

X264 HD Benchmark 4,0 - Pass 2
                                                        fps     15,2    13,9    12,96                           109,35%         117,28%

X264 HD Benchmark 4,0 - Pass 1
                                                        fps     74,4    68,5    69,7                            108,61%         106,74%

WinRAR - Result
                                                        KB/s    1328    1478,5  1952                            89,82%          68,03%

3DMark 11 - Performance Physics 1280x720
                                                        points  2181    2236,5  2704                            97,52%          80,66%

3DMark Vantage - P CPU no PhysX 1280x1024
                                                        Points  6586    5569,5  7264                            118,25%         90,67%

3DMark 06 - CPU -
                                                        Points  2843    2770,5  2713                            102,62%         104,79%

Cinebench R11,5 - CPU Single 64Bit
                                                        Points  0,9     0,61    0,96                            147,54%         93,75%

Cinebench R11,5 - CPU Multi 64Bit
                                                        Points  2,4     1,7     2,3                             141,18%         104,35%

Cinebench R10 - Rendering Single CPUs 64Bit
                                                        Points  3264    2166    4183                            150,69%         78,03%

Cinebench R10 - Rendering Multiple CPUs 64Bit
                                                        Points  8949    6453    8966                            138,68%         99,81%

Cinebench R10 - Rendering Single 32Bit
                                                        Points  2343    2060    3116                            113,74%         75,19%

Cinebench R10 - Rendering Multiple CPUs 32Bit
                                                        Points  6443    5482    6837                            117,53%         94,24%

Super Pi mod 1,5 XS 1M -
                                                        Seconds 19,9    26,3    19,3
                                                                                                                132,16%         96,98%

Super Pi mod 1,5 XS 2M -
                                                        Seconds 44,4    60      43,5
                                                                                                                135,14%         97,97%

Super Pi Mod 1.5 XS 32M -
                                                        Seconds 954     1517    962
                                                                                                                159,01%         100,84%

                                                                                                Median:         117,9%          95,0%
                                                                                                Average:        121,8%          99,9%

 

[Abwx]

Average advantage of A10-8700p over FX-7500 is 20%. But as compared to Broadwell i3-5005U, new APU is ~2,5% slower.

If i say that the 5005U is slower, and significantly..

The numbers you re using are extracted from this review :

http://www.notebookcheck.net/Lenovo-ThinkPad-Edge-E550-Notebook-Review.137727.0.html

For sure the chip largely exceed 15W even when throttled :

After short peaks of up to 35.6 watts, the load power consumption drops to the class average of almost 30 watts.

So it can start the benches at 25W TDP and will not finish it at below 20W, i guess that with such protocols it s not difficult for Intel to get their numbers hugely inflated..

At real 15W scores would be largely in favour of the 8700P, not counting the obviously innaccurate ST scores that are contradicted by other sources that made more reliable measurements.

 

[maarten12100]

Based on the 3dmark11 scores it could be a 35W cTDP which I think it is. This is what I meant, with AMDs huge cTDP variance this is a mess.

I think it should be the 15W part though but with dual graphics enabled so probably that same ball park of total consumption. According to AMD's own information the dual gpu config only works when configuring the A10-8700p as 15W chip. (This laptop has a R7 M360 btw)

HP dropped the ball or was this actually meant to be a sub 30W notebook. I think the later so the wait continues. Sure hope CPU numbers will be better for 35W parts. For 15W parts they are not bad.

 

[logiotek]

If you search latest HP AMD APU offerings you will notice that they stick single 8GB DDR3L-1600 DIMMs in their offerings. It's OK for Carrizo-L but it significantly hampers Carrizo. Some of the results don't make sense (especially graphics results) and that can be explained by lack of proper dual-channel RAM configuration.

Despite of this, if you look at the results that actually matter in real world (cryptography and x264) you will notice that Carrizo absolutely destroys the competition and last year's mobile Kaveri platform.

 

[Abwx]

Nice. Here is the gpu http://www.notebookcheck.net/AMD-Radeon-R6-Carrizo-Benchmarks.144290.0.html

@abwx can you compare the marketing slides to these results?

What is most weird are their ST scores in Cinebenches, the 8700P in ST get up to 3.2GHz so its ST scores should be close of the 8800P below :

http://forums.anandtech.com/showpost.php?p=37582356&postcount=1821

This make me think that i saw a CB test at NBC where they used 3 cores in the ST test of a Kaveri by distributing the load, this increase hugely power comsumption while reducing the ST score, see below and notice the cores loadind, the total is 100% of a single core :

http://www.notebookcheck.net/fileadmin/Notebooks/HP/EliteBook_755_G2/multi.png



Dont know exactly their methodology but the post above made me check their FX 7500P scores, and surprise they have done a single review but averaged the numbers with ONE score that they extracted god know from where and wich is 20% lower than the ones they measured in their sample :

http://www.notebookcheck.net/AMD-Kaveri-FX-7500-Notebook-Processor.117332.0.html

Good luck extracting differences with such truncatures disseminated everywhere....

Edit : The laptop they have at hand is this one :

https://geizhals.de/hp-pavilion-17-g054ng-n3x41ea-a1299109.html?hloc=at&hloc=de




....

 

[IntelUser2000]

Despite of this, if you look at the results that actually matter in real world (cryptography and x264) you will notice that Carrizo absolutely destroys the competition and last year's mobile Kaveri platform.

Just one point. Cryptography isn't a good measure of CPU architecture enhancement, because only recently CPUs have started implementing it in hardware and the gains are enormous. It's also certainly not relevant in terms of the "whole picture".

If you imagine the point where the gains are greatly diminished because the algorithm and architecture is pretty much good as it can be is 1.0. Cryptography might now be at 0.2. Until we reach much higher, you'll see Cryptography gains that are enormous every generation.

 

[logiotek]

Just one point. Cryptography isn't a good measure of CPU architecture enhancement, because only recently CPUs have started implementing it in hardware and the gains are enormous. It's also certainly not relevant in terms of the "whole picture".

Although you have a point about cryptography serving as indicator of CPU architecture enhancement, SoC design is moving toward specialized hardware (accelerators) for specialized workloads. Due to this fact and the fact that modern content is becoming more and more data-parallel, CPU core on its own will matter less and less in the future. Only overall SoC performance and ultimately user experience will matter in the end. I think AMD understands that very well and they are moving into the right direction (SoC, HSA, etc.)

We see planar 28nm SoC from AMD is slotting itself solidly in between i3 and i5 Broadwell-Us in a lot of workloads in terms of perf/W. In some workloads it even manages to outdo Intel's offering at similar price point. This gives me a confident feeling about future AMD SoCs on FinFet nodes with Zen core.

To paint a clearer picture we need more data samples, hopefully with optimal RAM configuration (DDR3L-1600+ dual-channel with 2 ranks per channel).

 

[dark zero]

Average advantage of A10-8700p over FX-7500 is 20%. But as compared to Broadwell i3-5005U, new APU is ~2,5% slower.

A dissaster.... the previous generation the i4-4003U was losing against the Beema A8 6410.

 

[Dresdenboy]

Intel doesn't do the exact same thing. Did you see a 15W ULV SKU turning into a 35W SKU? Or a 37W SKU turning into a 15W ULV SKU? On Intel this is much more restricted.

The point is you have a notebook and don't know how is it running/configured, as 15W or 35W. Nobody told 35W is a mess for a big 17" notebook assuming it is running at 35W which we don't know.

Mikk, you got them p.... down!
See from the first google hit:

ULV Ivy Bridge parts will be rated at 17W, similar to the ULV SNB CPUs that are used in Ultrabooks and the new MacBook Air. Intel will also guarantee these chips at a higher frequency with a TDP of 33W. If the Ivy Bridge MacBook Air could dissipate 17W of heat normally but when placed on a docking station with additional cooling capabilities could remove 33W of heat, the CPU would simply run at a much higher frequency when docked.

The same applies in reverse. If you want the chip to behave as if it were a 13W part instead of a 17W part, that will be possible as well.

The extreme edition IVB parts will also support configurable TDP. 55W parts will be able to go up to 65W or go down to 45W.

http://www.anandtech.com/show/4764/ivy-bridge-configurable-tdp-detailed

 

[ShintaiDK]

The problem is people try to compare 35W chips, with or without cTDP down with 15W chips, with or without cTDP down.

One can just claim the 15W 5200U Broadwell-U chips run at 7.5W.

But as seen with a few posters, nice way to use 35W numbers against 15W or sub 15W numbers to distort reality.

 

[Fjodor2001]

The problem is people try to compare 35W chips, with or without cTDP down with 15W chips, with or without cTDP down.

One can just claim the 15W 5200U Broadwell-U chips run at 7.5W.

But as seen with a few posters, nice way to use 35W numbers against 15W or sub 15W numbers to distort reality.

Not sure why you're describing this as an AMD-only-issue. Intel also has CPUs with various selectable TDP levels for the same chip; cTDP, SDP and whatever.

 

[coercitiv]

The problem is people try to compare 35W chips, with or without cTDP down with 15W chips, with or without cTDP down.

We won't compare chips, we'll compare products. Just like we do in Intel's case.