Anandtech benches Krait and Adreno 225

[dagamer34]

I think we've finally approached the point in mobile SoCs that video cards hit a while ago, so let's bring it out now:

"I don't care about benchmarks, show me a GAME".

 

[smartpatrol]

http://www.anandtech.com/show/5563/qualcomms-snapdragon-s4-krait-vs-nvidias-tegra-3

Part 2 of Anandtech's preview shows S4 utterly destroying Tegra 3 in single-threaded benchmarks and even some multi-threaded ones. Even the "underpowered" Adreno 225 dominates the Tegra 3 and Exynos in the supposedly more useful Basemark GPU benchmarks.

 

[ChronoReverse]

As noted in another thread (http://forums.anandtech.com/showthread.php?t=2228163), GLBenchmark seems to be a really terrible benchmark so I'm glad tests are being run using other benchmarks.

Still, I wish Unreal released benchmarking tools for their engine so that we'd have an useful real-life benchmark.

 

[poofyhairguy]

Part 2 of Anandtech's preview shows S4 utterly destroying Tegra 3 in single-threaded benchmarks and even some multi-threaded ones.

Browsermark has the Tegra 3 ranked higher, which supports the results I have seen.

Even the "underpowered" Adreno 225 dominates the Tegra 3 and Exynos in the supposedly more useful Basemark GPU benchmarks.

I don't know about "dominates." Of the two offscreen tests the Tegra 3 wins the more difficult one, and the Adreno wins the less difficult one.

 

[Bateluer]

Browsermark has the Tegra 3 ranked higher, which supports the results I have seen.



I don't know about "dominates." Of the two offscreen tests the Tegra 3 wins the more difficult one, and the Adreno wins the less difficult one.

Whats clear though, once the 3xx series launches, Tegra 3 is a has been. Even with the 225, they don't have a definitive lead. :/ I expected better from Nvidia.

 

[poofyhairguy]

Whats clear though, once the 3xx series launches, Tegra 3 is a has been.

That is true. Just like Tegra 2 was low-end last year, Tegra 3 is low-end this year.

 

[ChronoReverse]

Right now I can't take the GLBenchmark offscreen tests seriously since there are some severe discrepancies.

e.g., for the SGS2 with Exynos
34.6 fps @ 800x480 (onscreen)
42.5 fps @ 1280x720 (offscreen)

This isn't something explainable with vsync so the offscreen test really isn't usable.



As an aside, this might be interesting, I've contacted Antutu (and btw, they responded politely and quickly so I tip my hat to them) and they've explained that their CPU FP test doesn't use VFP nor NEON for compatibility reasons (presumably with the older A8's). Not using NEON is justifiable but to not even use VFP is pretty unrealistic I'm afraid =(

 

[ITHURTSWHENIP]

The only thing i can truly make out of all these benchmarks is that they are broken and we need better tools.

 

[poofyhairguy]

The only thing i can truly make out of all these benchmarks is that they are broken and we need better tools.

Amen. Though I like this CPU test, I wish Anand would use it:

https://market.android.com/details?id=eu.chainfire.cfbench&hl=en

 

[ChronoReverse]

The other problem is too many of the tests are running on the stock browser (like Browsermark) instead of a 3rd party browser that doesn't get specific vendor optimizations.

Normally, I'm all for these optimizations as they provide real benefits but if Asus improved their stock browser, this means Tegra3 on other devices won't have the same benefits. It also doesn't tell us what would happen if Asus optimized for another CPU (could be similar, greater or less performance boosts).

It makes more sense to select a particular version of a third party browser and test using that so that the only differentiating point is the CPU.

This is why Sunspider, V8 and Browsermark are all less than ideal benchmarks.

 

[poofyhairguy]

The other problem is too many of the tests are running on the stock browser (like Browsermark) instead of a 3rd party browser that doesn't get specific vendor optimizations.

Normally, I'm all for these optimizations as they provide real benefits but if Asus improved their stock browser, this means Tegra3 on other devices won't have the same benefits. It also doesn't tell us what would happen if Asus optimized for another CPU (could be similar, greater or less performance boosts).

It makes more sense to select a particular version of a third party browser and test using that so that the only differentiating point is the CPU.

I completely agree with that. Luckily when more things are ICS Chrome can be a baseline.

 

[ITHURTSWHENIP]

Whats clear though, once the 3xx series launches, Tegra 3 is a has been. Even with the 225, they don't have a definitive lead. :/ I expected better from Nvidia.

Tegra 3 is available today though. Krait with Adreno 225 wont be out for a few months yet and 3 series is not expected until late this year

Wich makes it a competitor to Tegra 4

 

[ChronoReverse]

What are the useful 3D benchmarks BTW?

There's Taiji (assuming if you can run it at 1024x768) but all I see otherwise is GLBenchmark of which I already pointed out problems with if you're using the offscreen render. Are there any other useful benches?

@poofyhairguy

Is your Transformer Prime running ICS? I'm curious as to what the V8 and Kraken scores would be.

 

[poofyhairguy]

What are the useful 3D benchmarks BTW?

I don't think there is one. The problem is all the different resolutions of devices.

@poofyhairguy

Is your Transformer Prime running ICS? I'm curious as to what the V8 and Kraken scores would be.

Yep, it is. I will run those tests in Chrome later today to let you know.

 

[stormkroe]

I wonder how hard it would be to code a timedemo into Shadowgun? I've played that on my Evo3d and haven't had a single bit of slowdown so maybe it's not demanding enough, though a custom recording with insane amounts of characters/explosions, and a 'frame cap disabled' setting?

 

[sontin]

That is true. Just like Tegra 2 was low-end last year, Tegra 3 is low-end this year.

Tegra 3 will not be low-end. At the end of the year i expect that DualCore A9 SoCs will be low-end. Everything from Tegra 2 to Snapdragon 3 and OMAP4. Tegra 3 will be something in the middle. Better than the DualCore Designs but slower than Dual/QuadCore A15.

 

[BenSkywalker]

GLBench seems to give an enormous advantage to the SGX543MP2. But, Basemark also shows the A5 coming out on top of the competition according to Anandtech's tests. See http://www.anandtech.com/show/5163/a...gra-3-review/3

There was a bug in that version of Basemark, Anand discussed it in the Krait article.

http://www.anandtech.com/show/5563/qualcomms-snapdragon-s4-krait-vs-nvidias-tegra-3/2

Tegra 3 is about 50%-100% faster then the A5 in Basemark depending on the bench. The 543MP2 is a dated GPU at this point, what people in this thread are calling slow, low end and obsolete easily bests it.

Whats clear though, once the 3xx series launches, Tegra 3 is a has been.

You seem to fail to consider that Tegra6 will utterly crush the 3xx series though. This is a game I have never understood on this forum. I can buy a Tegra 3 powered device right now, today. People talk like it is a major accomplishment that sometime within the next year something is going to come along that beats it, as if that is some sort of seismic event in the technology sector. It also clearly makes the part I can buy *today* somehow weaker. What is weak right now overall is pretty much everything *not* called Tegra3 if I look at numbers for shipping products.