New Video Shows Cortex-A15 Decimating Tegra 3; Be Wary of The Kool-Aid

New Video Shows Cortex-A15 Decimating Tegra 3; Be Wary of The Kool-Aid

Texas Instruments has posted a new video at YouTube that claims to show a dual-core OMAP5-based Cortex-A15 at 800MHz beating the snot out of an unspecified "commercially available" 1.3GHz quad-core device. Given that the only quad-core Tegra 3 tablet available is Asus Transformer Prime, it's not exactly hard to put two and two together. The Transformer Prime is extremely well-regarded, but there've already been rumors that the chip doesn't pack enough horsepower; Asus' highest-end TF700T, with its 1900x1200 screen, reportedly relies on a Qualcomm solution due to Tegra 3's limited memory bandwidth.

At first glance, the TI video seems to confirm that.



Both devices use Ice Cream Sandwich and are loading pages from EEMBC's BrowsingBench benchmark. TI claims that the systems are also downloading a video and playing back an MP3. The results show TI's 800MHz dual-core Cortex-A15 completing its page loads in 95s seconds, while the Tegra 3 device takes 201 seconds to finish the same workload. Huge victory for TI and OMAP5, right?

Not so fast. The reason we're so suspicious of these results is the size of the performance gap. Forget the dual-vs-quad argument -- clock for clock, the Cortex-A15 is expected to be 40% faster than the A9. That should put a theoretical 800MHz A15 just behind a 1.3GHz A9 in a best-case scenario. Instead, TI claims their chip is more than 2x as fast. A gap that large strongly implies that other factors are at work, with network configuration the prime suspect.



The Transformer Prime's WiFi reception and transmission speeds aren't all that great, with the first run of products disproportionately affected. TI's test, with its simultaneous video download, plays to that particular weakness. As part of an overall review, it's completely fair to measure and compare the Prime's WiFi performance with other shipping products -- but the OMAP5 TI demonstrated is months away from that point.

Next up, there's BrowsingBench itself. The test is designed to allow reviewers to simulate a wide range of network conditions, from high-latency/low bandwidth connections to "Sitting on the WiFi unit." BrowsingBench has a default mode, which it uses to calculate general performance—but that mode isn't being used here. Instead, all we see are page load times. There's no proof that the two systems are equally configured or even connecting to the same router.

Finally there's the question of when, exactly, it's fair to start publishing comparisons. It's always been a given that Tegra 3's quad-core A9's would face stiff competition from manufacturers like Qualcomm, but Qualcomm's Snapdragon S4 is set to ship in the first half of the year. OMAP5, in contrast, isn't expected until 2013. Even if TI's performance figures are valid, the video compares the performance of a system you can buy today with that of a system you won't be able to buy for at least 9 months. That's dubious, at best.

Nvidia is no stranger to benchmark shenanigans; the company has been caught with its hand in the cookie jar more times than we can remember -- but that doesn't make obfuscation and questionable result publishing a good idea. It's possible that TI's comparison is fully on the level, but comparing against hardware without a firm ship date is never fair -- and there's good reason to think the results weren't measured fairly.
0
+ -

TI OMAP 5432 is scheduled for the second half of 2012, not 2013!

The 40% for Cortex A15 is actually the power efficiency difference to A9, and not a direct indicator of the performance difference.

TI is actually claiming a 50% performance boost at the same clock speed, with up to 8GB of dynamic memory access and hardware virtualization support, and capable of speeds of up to 2GHz per core.

I agree though that the video demo was probably leveraged heavily in TI's favor. In addition to networking, other factors like memory bandwidth, and whether the test was even optimized for more than dual cores could have all factored.

0
+ -

Incorrect on the second point, possibly inaccurate on the first.

http://www.intomobile.com/2012/01/16/texas-instruments-shows-off-omap-5-ces-admits-its-not-going-come-until-2013/

The explanation is the long lead time in device manufacturing. If TI sampled in the later half of 2011 and ships in volume by mid-2012, vendors may not have solutions in the market until late-2012 / early 2013. Check Qualcomm's roadmap, or Samsung's, and you'll see the same pattern. Device shipments often lag chip shipments by 4-6 months, and the manufacturers aren't precise as to whether a given timeline targets one or the other.

As for the 40% performance increase, I direct you to the original graphs: http://eda360insider.wordpress.com/2010/11/10/arm-cortex-a15%E2%80%94does-this-processor-ip-core-need-a-new-category%E2%80%A6superstar-ip/

That's not power efficiency, that's performance.

The Cortex-A15 remains a 32-bit CPU; it can support >4GB through the use of address extensions similar to Microsoft's old PAE. These are ugly and gross and highly undesirable. There's no reason to put >4GB of RAM in a tablet or smartphone anyway and given that RAM consumes power, that's not going to change. Unlike desktops and laptops, where the energy cost of an extra 1GB of RAM was miniscule, it matters in tablets. Expect to see RAM loadouts increase much more slowly and more in line with advances in DRAM manufacturing and/or the adoption of lower-power DDR standards.

0
+ -

"As for the 40% performance increase, I direct you to the original graphs: http://eda360insider.wordpress.com/2010/11/10/arm-cortex-a15%E2%80%94does-this-processor-ip-core-need-a-new-category%E2%80%A6superstar-ip/

That's not power efficiency, that's performance."

Sorry, perhaps I should explain it better. The 40% is only measuring integer performance, which isn't the whole picture on performance.

While integer performance is a strong indicator of general performance it's not the whole picture of how well a given architecture performs compared to another. Things like In Order Processing and Out Of Order Processing, spreading work load over multiple cores, etc are not factored.

While also the FPU performance is not factored. Processors that have similar performance on integer tasks can have very different performance on floating point operations. So the 40% doesn't tell us how well it would handle things like spreadsheets, some graphics programs, and games.

Mind that floating point is more important to calculation intensive tasks like graphics, which is more heavily used in a graphical OS like Android. Along with factors like hardware acceleration in the browser.

Your link for example shows a graph indicating up to 7x improvement for FPU for the A15 vs A8. Also the integer performance doesn't factor for all the memory performance, which your link also states is twice as good on top of the 40% it indicates for integer performance. Along with other memory performance/management enhancements and other features of the A15 architecture that enhances it from the A9.

Then there is the rest of the system like the GPU, system drive, and software optimization also factor into the benchmarks.

So like I said, TI is actually claiming 50% better performance and we can't go by just the 40% integer performance to say they can't provide that much of a performance boost. Though they could have easily lopsided the advantage in the video demo in a number of ways.

As for the delay, yes it can take more months than predicted but that's mainly for the full line up. Like for comparison, Intel is delaying the release of Ivy Bridge but they're still coming out with the quad core versions on time. So a delay doesn't mean they won't come out with anything in the intermediate time period.

0
+ -

Cool....too bad i still dont have the money for a smart phone or tablet.

0
+ -

I don't know I have an odd perspective on this. I joined T-Mobile last year and I first had a G2X which ran a Nvidia dual core and was one of the first in the US to do so. That phone had issues top to bottom and I returned it in 2 daysat which time T-Mo gave me another one in the box. I took it home same issues I checked the serials etc and it was from a totally different batch than the original, but had the same problems. They asked me to bring it in and they switched the Sim thinking it was a addressing issue with the SIM. The phone promptly killed it so they put another SIM in and it worked so I left the store but had the same issues a lot of dead connection time. I played with them though dl'd a bunch of apps when they were operational. They ran hot and it eventually in another 2 days killed it's SIM. The performance was decent but not really spectacular especially compare.

Anyway they tried to get me to take another one etc and even gave me another one in box but I did not end up leaving the store with it. I decided to wait 3 weeks for the Sensation and turned my old phone back on and waited. I have made the Sensation since the first day and first hour of official release. It uses a Snapdragon and the performance is better across the board. It also runs much cooler and has been a stable unit throughout my 9 months of usage now.

I know it is not the same processor types I just think Nvidia is full of hot air a lot and uses the over emphasis as a selling point. Saying I was the best at anything never seemed to make it so in my life and I doubt it does in either TI's or Nvidia's in the end. I am just saying talk is cheap as are virtual benchmarks and there showing on video's. I have learned that from computing hardware in general over many years of use personally.

-1
+ -

rapid1, you used so many words while managing to say absolutely nothing. Do you work in politics at all? Public relations?

0
+ -

Jim Jones gave Flavor-Aid to his followers at Jamestown in Guyana, not Kool Aid, it's a common misquote.

0
+ -

I know it is not the same processor types I just think Nvidia is full of hot air a lot and uses the over emphasis as a selling point. Saying I was the best at anything never seemed to make it so in my life and I doubt it does in either TI's or Nvidia's in the end. I am just saying talk is cheap as are virtual benchmarks and there showing on video's. I have learned that from computing hardware in general over many years of use personally.

0
+ -

Hey jpapenfuss good job on your second post ever, as far as it goes my first job is in the IT sector in communications my second job is servicing your mother! She pays me pretty well too!

0
+ -

I read rapid1's response, it was pretty clear what he was implying jpanpenfuss...

0
+ -

Rapid, Jpapenfuss may have been a bit rude, but your comments are over the line. Please refrain from such.

0
+ -

Free Speech is hard to deal with at times is it not! Oh wait we are still in america right as I don't think I said anything vulgar unless your old enough to understand it, of course at which point it would not matter then would it!

0
+ -

rapid, please tone it down. We like to keep things civil here and attacking other members with insults isn't going to be tolerated. Thanks

Login or Register to Comment
Post a Comment
Username:   Password: