Dolphin CPU hierarchy [UNOFFICIAL]

[admin89]

I forgot to mention about Phenom II

Quote:Desktop :
Category 3: Fast
These cpus are fast enough to run most games at fullspeed with HLE audio, especially when overclocked. When overclocked they can perform as well as category 2 cpus.
-Intel Core 2 Extreme
-Intel Core 2 Quad (very high end models only)
-Intel Core 2 Duo (high end models)
-AMD FX series
-AMD Phenom II
-AMD A10 series
-AMD A8 series
-AMD A6 series

No way Phenom II could be faster than Trinily . Llano is 5% faster than Phenom II

Quote:Category 3: Moderate/fast
Inbetween moderate and fast. Faster than category 4 but slower than category 2.
-Intel Core 2 Extreme Mobile
-Intel Pentium
-AMD Phenom II
Category 4: Moderate
These cpus run most games at fullspeed with HLE audio if they are heavily overclocked. But at stock speeds they may have have trouble.
-Intel Core 2 Duo Mobile
-Intel Core 2 Quad Mobile
-AMD A10 series
-AMD A8 series
-AMD A6 series
-AMD A4 series
-AMD Turion II

Phenom II is faster than Core 2 duo ?

[NaturalViolence]

The phenom II mobile chips clock up to 3 GHz (stock speeds). They have about the same or slightly higher IPC than piledriver cores. The high end models are extremely rare in laptops but I have to count them none the less since they do exist.

[admin89]

If you're talking about clock rate , A10 4600M (turbo core 3.2GHz) beats them all . It even chop off the low-end i3/i5 , for example : i5 430M , i3 380M )
And the desktop A10 5800k (3.8 -> 4.2) beats all Phenom II , Core 2 Duo , Core 2 Quad

[NaturalViolence]

I'll say it again. The top of the line phenom II mobiles are comparable to trinity or faster in single threaded tests.

[admin89]

I thought Trinity did well in single threaded benchmark (i haven't checked any benchmark yet) . It seems i was wrong on so many levels
Just look at the Cinebench R10 - Single Threaded Benchmark , I'm totally disappointed

[tutu]

How does a A6 5400K compare to a A10 5700 in terms of dolphin performance?

The A6 is unlocked and could overclock higher.

But the A10 has more shaders and is quad core. The A6 has 1 FPU unit where the A10 has 2 FPU units.

[admin89]

A6 5400k (3.6 -> 3.8GHz)
A10 5700 (3.4 -> 4.0GHz)
In term of clock rate , there is not much difference
Dolphin uses 2 cores + 1 core (if "DSP LLE on thread" is enabled and you must have LLE backend : "dsp_coep.bin" & "dsp_rom.bin")
A10 is a quad core , A6 is dual core . Leaving 2 idle cores for background processing will speed up dolphin

-> At Stock speed : A10 is faster for Dolphin
->OCed A6 @ 4.4GHz vs A10 : It's hard to say which one is faster than the other , A10 is probably faster but the difference is not noticable for Dolphin
Price at introduction : 67$ vs 122$

A8 5600k (3.6 -> 3.9GHz ,Quad Core, $101) is a better value than both A6 and A10

[NaturalViolence]

Quote:I thought Trinity did well in single threaded benchmark (i haven't checked any benchmark yet) . It seems i was wrong on so many levels
Just look at the Cinebench R10 - Single Threaded Benchmark , I'm totally disappointed

But those are desktop cpus that you're comparing. This is what you should be looking at: http://www.anandtech.com/show/4444/amd-llano-notebook-review-a-series-fusion-apu-a8-3500m/9

For reference:
P920 is a quad core athlon II with 512KB L2 cache per core and no L3 clocked at 1.6GHz, marketing as phenom II mobile
N660 is a dual core athlon II with 1MB of L2 cache per core and no L3 clocked at 3.0GHz, marketed as phenom II mobile
A8-3500M is a quad core llano based cpu with 1MB of L2 cache per core and no L3 clocked at 1.5GHz with turbocore up to 2.4GHz
P520 is a dual core athlon II with 1MB of L2 cache per core and no L3 clocked at 2.3GHz, marketing as turion II

All 4 of these cpus are based on the K10 core microarchitecture.

As you would expect the older N660 kicks the A8-3500Ms ass in single threaded tests due to its ridiculously high clock rate (for a laptop cpu).
Llano obviously beats the P920 in single threaded tests but not by as much as you would think.
The P520 beats it in single threaded tests by a small to moderate margin depending on the test.
The N660, P920, and A8-3500M all perform similarly in multithreaded tests, trading blows with each other. The P520 obviously lags behind in this area.

However I consider this unfair as they are comparing the top of the line llano from 2011 against a midrange phenom II X4 mobile from 2010. The top of the line phenom II X4 mobiles would kill the llano in both single and multithreaded tests, maintaining higher clock rates with similar IPC. Considering that they're basically a P920 with a 25-37.5% higher clock rate. Although these cpus were rare in laptops they did exist in high end AMD laptops. You can only imagine what these cpus could have done if they had switched them to the husky cores, which have turbocore.

What we see here is two disturbing facts. Llano traded away some cpu efficiency (cpu performance per watt) in exchange for adding an IGP on die, faster memory controllers (to feed the IGP), and simpler chipset/interconnect architecture to reduce motherboard cost and heat. This was all in spite of the fact that it received a die shrink to 32nm which was apparently not enough to make up for this. I would also like to point out that AMD still hasn't produced a 35w laptop cpu that can compete with the top of the line core 2 duo mobile T9900 in single threaded performance and are only just now catching up to it in multithreaded performance.

Now where does trinity fall? It's hard to judge due to the lack of benchmarks on older laptop cpus but it seems that the top of the line trinity cpus are the first AMD cpus to outperform phenom II mobile and core 2 duo mobile in multithreaded performance. However in single threaded performance it still loses by a small to moderate margin to both of them. Since dolphin does gain a small performance boost from having a 3rd/4th core available performance on a top of the line phenom II mobile, core 2 duo mobile, and trinity mobile should all be almost identical. With the obvious major difference being that trinity can maintain the same TDP while having fast memory controllers and an IGP on die, therefore making it the superior cpu out of the three. Since historically Intel has priced their high end desktop and laptop cpus outrageously high most Intel laptops use low end or midrange Intel cpus. It's much more common to see AMD laptops with high end AMD chips than Intel laptops with high end Intel chips because AMDs chips tend to be a lot more affordable (since they're crappier).

None the less while AMD has finally outpaced core 2 on the mobile front trinity still struggles to compete with Intels old midrange nehalem chips. Against dual core nehalems it loses big time in single threaded performance but wins in multithreaded tests. In hybrid tests (multithreaded but not heavily multithreaded) it trades blows with dual core nehalems usually taking a slight lead. Against a decent quad core nehalem chip it doesn't stand a chance in single or multithreaded tests. And it's nowhere near the cpu performance of mobile sandy bridge and ivy bridge cpus.

Trinity is a step up from previous AMD laptop cpus but it's still a LONG way behind Intel in laptop cpu performance. On the desktop AMD is usually 1-2 years behind Intel (now closer to 3 thanks to bulldozer/piledriver). But on laptops AMD has consistently been 3 years behind Intel due to the fact that historically their architectures just aren't as energy efficient and therefore suffer from poor scaling to the low end platforms (closer to 4 years now). They had a really hard time catching up to core 2 duo mobile for a few years and basically lost the laptop market in the process. Only now are they even starting to be taken seriously again as a laptop contender since they are offering IGPs that aren't complete shit on cheap laptops.

The Phenom II mobile cpus from 2010-2011 are basically just athlon IIs with lower clock rates/voltages. The dual core models go up to 3GHz, triple core up to 2.3GHz, and quad core up to 2.2GHz while maintaining the standard laptop TDP of 35w. The high end core 2 duo mobiles go all the way up to 3GHz as well and can beat phenom II mobile since they have higher performance per clock.

Quote:How does a A6 5400K compare to a A10 5700 in terms of dolphin performance?

The A6 is unlocked and could overclock higher.

But the A10 has more shaders and is quad core. The A6 has 1 FPU unit where the A10 has 2 FPU units.

Depends how high you OC. The A6 is going to take a fairly steep performance hit from the fact that it's dual core. The piledriver architecture shares resources between cores in what they call "modules". 2 cores per module. With dual core cpus when you run dual threaded software both threads will be sharing certain resources which will reduce performance, especially for applications with very unpredictable memory access (like dolphin). The exact amount of performance reduction is difficult to determine but I can guarantee that the 5700 will win at stock speeds and that you will need to OC the 5400K to reach comparable speeds. Exactly how much of an OC is needed I can't say. The 5700 is a safer bet.

You haven't told us if you're going to be using the on-die GPU yet. If you're not I would second admin89s suggestion of an A8-5600K.

[admin89]

Thanks for your explanation NV .

myself Wrote:A10 4600M (turbo core 3.2GHz)

I found sth interesting
http://forum.notebookreview.com/hp-pavilion-notebooks/670473-trinity-has-dropped-2.html

Quote:my trinity is here.
will post benchmarks and crap later.
All tests done with High Performance Power Options setting.
Only stock tests, since AMD locked down the cpu. No overclocking allowed

Actually , turbo core on mobile APU never work , Starcream has to overclock his CPU with Kstat10 (1.6->2.4GHz)
Without Kstat10 or PScheck overclocking , A10 4600M clock rate never reach 3.2GHz , it will stay at the stock speed 2.3GHz forever

[NaturalViolence]

It does work. But the users doing the tests apparently have no idea how turbocore works and think that their software is magically going to calculate the average clock rate.

The only way to know for sure if turbocore is actually working is through benchmarks. To my knowledge their is no known way right now to use third party software to accurately measure the average clock rate over a period of time.