Cortex A15 Benchmarks

[Arachnotronic]

Hey all,

So I was looking at Anand's review of the ARM-based Chromebook and saw that is absolutely decimated the Atom N570. However, I noted a couple of things:

- The Atom N570 gets ~1034ms in SunSpider on the Chromebook, but the Atom Z2760 in the Intel based tablet that Anand tested against the Surface got ~714ms. What could explain this discrepancy if the cores really are similar? (and both were running the Chrome browser)

- Power consumption for the A15 was quite high compared to what is typically needed in a tablet/smartphone; does this mean that tablet uses of the A15 will need to really kill the clock speeds and/or performance?

 

[grkM3]

its all about what running in the backround and how agressive the clocking steps are used on the a15,like witch govenors are used and it could not be clocking full speed threw the whole bench.

I think I saw a 600ms sunspider bench on the a15 somewhare also but that benchmark is so all over the place,like os and browser tweaks can make north of 500ms differences.

I scored 1600-1700 with my gs3 running ics on stock browser but can score 1100 on my old nexus with outdated hardware and better browser and os.

the gs3 should destroy the old nexus but it losses big time on that bench

 

[jpiniero]

Pretty sure the Atom in the surface review is also using IE 10.

As for the power draw, the Nexus 10 has the Exynos Dual as well so it's doable in a tablet. Clock speed is probably not 1.7 though.

 

[Ferzerp]

sunspider is not a cpu benchmark.....

 

[grkM3]

sunspider is not a cpu benchmark.....

We have a winner

Its a browser benchmark

 

[Ferzerp]

I mean, you can use it as one with the exact same javascript engine, but if they are different (as they are here), it is a meaningless comparison with regard to a cpu. It is almost like taking gpu limited benchmarks to compare cpus, but then using different gpus on top of that.

 

[Khato]

- Power consumption for the A15 was quite high compared to what is typically needed in a tablet/smartphone; does this mean that tablet uses of the A15 will need to really kill the clock speeds and/or performance?

Nice to at least see some useful power measurements for an A15 based design to go along with the performance isn't it? Such basically indicates that it's performance per watt hasn't really improved, it just has a reasonable amount of performance headroom unlike previous ARM designs.

The ball's now in Intel's court, but sadly it sounds like we're going to have to wait awhile to get their real response.

 

[USER8000]

The old Chromebook used a battery which is twice as large as the new version.

 

[superccs]

Chromebooks can become Roku's right?

 

[dguy6789]

The single most interesting thing in that entire review to me is the A15 power consumption. Simply put there is no way they can put that processor as it is in the Chromebook into a modern smartphone without completely destroying battery life. The way I'm reading it if they cut clock speed enough to make the A15's power consumption similar to a Snapdragon S4 at load the performance would be quite a bit lower than the S4. Is there going to be a die shrink or a new revision of the A15 before it's expected to hit phones?

Is too early to tell but S4 vs A15 is looking like it's going to be similar to the Radeon HD 5870 vs Geforce GTX 480 to the letter.

 

[podspi]

The single most interesting thing in that entire review to me is the A15 power consumption. Simply put there is no way they can put that processor as it is in the Chromebook into a modern smartphone without completely destroying battery life. The way I'm reading it if they cut clock speed enough to make the A15's power consumption similar to a Snapdragon S4 at load the performance would be quite a bit lower than the S4. Is there going to be a die shrink or a new revision of the A15 before it's expected to hit phones?

Is too early to tell but S4 vs A15 is looking like it's going to be similar to the Radeon HD 5870 vs Geforce GTX 480 to the letter.

Couldn't you say the same thing about the Atom?

Architecture != Product. This is an A15(architecture)-based product. Presumably the A15-based smartphone products will have lower performance and power consumption, just like the Atom counterparts.

 

[Nemesis 1]

Nice to at least see some useful power measurements for an A15 based design to go along with the performance isn't it? Such basically indicates that it's performance per watt hasn't really improved, it just has a reasonable amount of performance headroom unlike previous ARM designs.

The ball's now in Intel's court, but sadly it sounds like we're going to have to wait awhile to get their real response.

Sadly thats not true . Intels real target right now is phones. Medfield does well and the new dual cores are soon to be released, so now intels setting pretty good in the phone space . This is the space intel was after .. Sure the 22nm OoO will be a good chip that late in 2013.

 

[augiem]

Hey all,

So I was looking at Anand's review of the ARM-based Chromebook and saw that is absolutely decimated the Atom N570. However, I noted a couple of things:

It's too bad there wasn't a good GPU benchmark to compare the Mali T604 to the PowerVR on the Clovertrail (single-core SGX545 if I'm not mistaken). I've been looking forward to seeing how the T604 stacks up to the iPad's quad-core SGX543, but I'd also love to see how it does against the HD4000 in the i5 Windows 8 tablets. We need a mobile GPU throwdown!

 

[ViRGE]

Couldn't you say the same thing about the Atom?

Architecture != Product. This is an A15(architecture)-based product. Presumably the A15-based smartphone products will have lower performance and power consumption, just like the Atom counterparts.

Eh, yes and no. Atom's poor power consumption out of the gate was due to being built against the 45nm process and requiring too many additional chips. By the time we're talking about Clovertrail we're talking about S0ix sleep states, tight (and proper) SoC integration, 32nm low power process, etc. That's a lot of changes compared to Atom as it launched.

A15 on the other hand won't be going through that kind of a change. Partners can at best play with implementation-specific details (lower clocking, less leaky transistors), and of course adjust clockspeeds and voltages. At a minimum, load power consumption can definitely be reduced by bringing down clockspeeds and voltages. The big question is whether it will be enough to make A15 viable, and even then whether the idle power consumption of A15 will be good enough, and whether the resulting performance of A15 will be any better than things like Krait and Swift given the fact that they're all on the same process. So there are still a lot of unknowns at this point.

 

[augiem]

Finally some Nexus 10 (Mali T604) benchmarks!

http://www.anandtech.com/show/6426/ipad-4-gpu-performance-analyzed-powervr-sgx-554mp4-under-the-hood

Mali T604 performance is utterly abysmal. I'm shocked, yet I'm not. Par for the course. And it's got to push more pixels than the iPad with a squirrel under the hood. (Yes, I know I'm exaggerating.)

All Android and Windows phone and tablet makers are absolutely negligent for not only continuing to let Apple's massive graphics lead go on for a 6th year, but for allowing the gap become even wider than its ever been.

Nobody but Imagination knows squat about making a GPU. It's really frustrating as hardware enthusiast to see this year after year. I'm sad.

 

[sontin]

Imagination only provides the design. It's Apple which is producing these big SoCs. A6X is 123mm^2 big - 45% bigger than Tegra 3.

Every other company could do this with the market share.

 

[augiem]

Imagination only provides the design. It's Apple which is producing these big SoCs. A6X is 123mm^2 big - 45% bigger than Tegra 3.

Every other company could do this with the market share.

It's been done before. Sony PS Vita uses SGX543MP4, the same GPU in the iPad 3. The CPU part is different, but that's fine. Other SoC makers need to get PowerVR's in their SoCs. Not that hard. Intel is doing it, but even they are doing a half-baked version with a single-core PowerVR slated for the Clover Trail (from the info I can find). Apparently nobody but Apple can see the importance of having a cutting-edge GPU in their devices.

 

[sontin]

TI latest SoC - OMAP 4470 - is using PowerVR tech and yet it's only as fast as Tegra 3:

http://www.anandtech.com/show/6158/the-archos-101-xs-review/3

There is no magic. As long as companies like nVidia are not producing bigger SoC they will never catch Apple.

 

[augiem]

TI latest SoC - OMAP 4470 - is using PowerVR tech and yet it's only as fast as Tegra 3:

Single core. Of course its slower. PowerVR's core count goes up to 16.

 

[sontin]

Single core, older generation GPU. 540 is from 2007. Of course its slower.

You said they "need to get PowerVR's in their SoCs". TI is doing it and you don't like it? D:

 

[augiem]

You said they "need to get PowerVR's in their SoCs". TI is doing it and you don't like it? D:

Oh come on. That's just playing with words. They need to use the best PowerVR offers. NVidia and the others don't offer anything that compares to PowerVR's high end AFAIK. If they do, nobody has chosen to use it.

I specifically said Intel is doing a half-baked version using a single-core PowerVR in the Clover Trail. That should have told you that I know not ALL Power VR chips beat ALL non-Power VR chips, and that just having a PowerVR in your SoC doesn't automatically it faster than everything else.

As a consequence of using a multi-core GPU, the SoC will be bigger, yes. I understand that. I'm not saying you're wrong. But nobody out there even has plans to use the higher end PowerVR offerings that I'm aware of. Nvidia's Logan or Stark are just about the only hope I have of any real GPU competition, and those are 1-1.5 years away from release, and even longer to show up in a device probably.

TI latest SoC - OMAP 4470 - is using PowerVR tech and yet it's only as fast as Tegra 3:

Except the Tegra 3 in the Transformer Pad Infinity is a 12-core GPU running at 520 MHz. (http://en.wikipedia.org/wiki/Nvidia_Tegra) and the PowerVR 544 in the OMAP 4470 is a single-core (AFAIK) running at 384MHz (http://www.anandtech.com/show/4413/...pecs-powervr-sgx544-18-ghz-dual-core-cortexa9, not 100% sure of the MHz in the Archos implementation, but the CPU part runs slower at 1.5Ghz vs 1.8 in the article). So there is a bit of "magic" going on, meaning the PowerVR is running quite a bit faster per-clock and per-core than Tegra 3.