[AT] NVIDIA Denver performance

[Arachnotronic]

The benches are in...

Courtesy of AT's review: http://www.anandtech.com/show/8670/google-nexus-9-preliminary-findings

Looks like Geekbench may have overstated Denver's performance. It's a solid core, but it doesn't seem to be anything to write home about.

 

[NTMBK]

Wow, that's some inconsistent performance. Hopefully updated versions of the code morphing software will improve things.

 

[Roland00Address]

It looks to be better than anything else on android (apple soc is superior though). Yes it loses some benches, part of that is probably due to its code morphing thingy, and part is probably due to the fact it is still a dual core and it is competing against quad cores so if the bench can take advantage of 4 cores it is harder for Denver to keep up (I am guessing the physic benchmark in 3d mark is such a thing). Also it may have problems with web and java?

That said for a 28nm silicon it is pretty good, I only count 4 times it loses against the galaxy note 4 (sunspider, basemark II web, basemark II system but it is very close, 3d mark physics). Galaxy Note 4 is a Qualcomm 805 2.5 ghz part. We are not going to see anything better from Qualcomm till 20nm with cortex a57.

 

[ams23]

Denver's CPU performance is very good when it can take good advantage of the dynamic code optimization process. It performs better than enhanced Cyclone and R3 Cortex A15 in Geekbench 3 single-threaded score, Kraken, and likely Octane too (according to someone who has tested the retail version at XDA developers forum). Sunspider is an anomaly of course, but some reviewers did note in previews that web pages subjectively load just as fast on Nexus 9 as they do on iPad Air 2.

Note that the 3dmark Physics test takes good advantage of all four cores in the quad-core Cortex CPU, so the dual-core Denver CPU would obviously be at a disadvantage in comparison, but even so it did outscore the tri-core enhanced Cyclone CPU in that test.

 

[Sweepr]

Seems a bit inconsistent and based on the mixed reviews I'd rather buy the cheaper NVIDIA Shield Tablet or an iPad Air 2 right now.

 

[TuxDave]

Seems a bit inconsistent and based on the mixed reviews I'd rather buy the cheaper NVIDIA Shield Tablet or an iPad Air 2 right now.

I generally not too surprised at the performance numbers. An architecture like this needs quite a bit of performance tuning. It makes a great research vehicle and I'm sure Google has assigned enough programmers to make sure the OS runs well on it. The problem is how many programmers the Denver compiler team can convince to fine tune their software.

 

[tviceman]

Lets wait until shipping firmware is out. If there is a chance that it can drastically affect numbers, it's premature to come to a conclusion.

 

[zebrax2]

The performance look great while the battery life OTOH doesn't look that good specially since it is running Lollipop which supposedly have a few enhancements that should help battery life.

 

[Roland00Address]

Any battery life tests is probably not on the official "release" firmware instead of beta firmware. The official update came out and was 800mb yesterday. This is the same software that is available on release date from Google.

 

[ElFenix]

The front page needs to hire copy editors.

 

[greybaby]

Thats too bad, I was really hoping it would be something significant just for competition sake. I mean it still looks somewhat competative, just not that impressive.

 

[ams23]

This is what Ars Technica said about Nexus 9 web browser performance:

"The Nexus 9 performed well when focusing on a single task, not a surprise given its excellent single-core performance (see our browser scores for evidence). Our iPad Air 2 review included an informal test of Web browser performance by loading and perusing an endless Facebook page, scrolling for days' worth of feed content, and we did the same thing again with HTC's tablet. The Nexus 9 handily beat the iPad Air 2 in that regard; it simply never crashed, nor did it lose track of any content it loaded during scrolling. (This stands in stark contrast to its surprisingly lousy Sunspider test score, by the way.) "

 

[monstercameron]

any word on clockrates?

 

[Enigmoid]

The two threads on this topic need to be merged.

 

[TrulyUncouth]

This is what Ars Technica said about Nexus 9 web browser performance:

"The Nexus 9 performed well when focusing on a single task, not a surprise given its excellent single-core performance (see our browser scores for evidence). Our iPad Air 2 review included an informal test of Web browser performance by loading and perusing an endless Facebook page, scrolling for days' worth of feed content, and we did the same thing again with HTC's tablet. The Nexus 9 handily beat the iPad Air 2 in that regard; it simply never crashed, nor did it lose track of any content it loaded during scrolling. (This stands in stark contrast to its surprisingly lousy Sunspider test score, by the way.) "

Don't most people run multiple tabs? I always just assume everyone browses like I do, but guess I could be wrong.

I think the fact that its performance looks inconsistent on mainstream benchmarks that we can assume were optimized in advance makes me think we will see worse performance the more obscure the app is.

The reports of laggy interface and other issues in android might be down to the new version, but doesn't really build much confidence in google/NV. TK1 with A15 cores seems to be the better bet right now, think this might move some nvidia shields in the long run.

 

[NTMBK]

Interesting note from The Verge:

The 9 also gets warm with just light use and downright hot when it’s pushed while playing a game. The iPad isn’t immune to heating up during gameplay, but the Nexus 9’s heat feels less like a discomfort and more like something is terribly wrong.

Sounds like the K1 is seriously pushing the limits of what you can fit in a fanless tablet. It would be interesting to see how it performs in a microconsole with a fan- the better cooling may let it stretch its legs and perform even better.

 

[MisterLilBig]

Looks like Geekbench may have overstated Denver's performance.

For some reason everyone kept ignoring that those scores where on a 2.5GHz Denver, not a 2.3GHz Denver like on the Nexus 9.

I wonder if the newest firmware will actually have a strong effect on it.


And, then we gotta see how all 64-bit capable devices do with Android 5.

Sounds like the K1 is seriously pushing the limits of what you can fit in a fanless tablet.

Sounds like they should have down clocked it even further!


This is starting to sound like another very bad product. They should have put this on a new SHIELD console.

 

[meloz]

How can a CPU be so good at Kraken and yet be utterly mediocre in Sunspider? This is hard to believe, there must be some error in benchmarking procedures.

Is this something to do with single-threaded versus multi-threaded performance? Anyone willing to speculate what is going on?

 

[monstercameron]

How can a CPU be so good at Kraken and yet be utterly mediocre in Sunspider? This is hard to believe, there must be some error in benchmarking procedures.

Is this something to do with single-threaded versus multi-threaded performance? Anyone willing to speculate what is going on?

code morphing? firmware/software issues? throttling? cpu design not suitable? could be alot of things.
It doesnt use alot of power but it gets very hot. maybe they should have downclocked it to win battery benchmarks vs performance benchmarks?

 

[MisterLilBig]

It doesnt use alot of power but it gets very hot. maybe they should have downclocked it to win battery benchmarks vs performance benchmarks?

Which is why the HTC Volantis is at 2.5Ghz and the Nexus 9 is at 2.3GHz?

If the heat is as bad as it sounds, its a bad product.

Uncomfortable Tablet = Bad Tablet = Bad Product.

Does not matter what company makes it or what SoC it has, if I can't use the Tablet the way and for the reasons a Tablet is used, that's just awful!

 

[Erenhardt]

How can a CPU be so good at Kraken and yet be utterly mediocre in Sunspider? This is hard to believe, there must be some error in benchmarking procedures.

Is this something to do with single-threaded versus multi-threaded performance? Anyone willing to speculate what is going on?

Heat capacity + light workload turbo vs thermal throttling durning heavy multi core load. Benchmark duration? But that's just my speculation

 

[sm625]

That sunspider score is totally whacked. I'm assuming they got somebody working on it, it seems like an obvious place to look for a rather simple bug. (Simple to the guys who designed this cpu anyway.)

 

[SlowSpyder]

This is what Ars Technica said about Nexus 9 web browser performance:

"The Nexus 9 performed well when focusing on a single task, not a surprise given its excellent single-core performance (see our browser scores for evidence). Our iPad Air 2 review included an informal test of Web browser performance by loading and perusing an endless Facebook page, scrolling for days' worth of feed content, and we did the same thing again with HTC's tablet. The Nexus 9 handily beat the iPad Air 2 in that regard; it simply never crashed, nor did it lose track of any content it loaded during scrolling. (This stands in stark contrast to its surprisingly lousy Sunspider test score, by the way.) "

I'm not sure what that means for real life use, though? How many people care about tablet performance for seeing a year of someone's FB posts? If that can translate into performance in other ways, that's great. But it seems like an odd test to me.

 

[Arachnotronic]

I'm not sure what that means for real life use, though? How many people care about tablet performance for seeing a year of someone's FB posts? If that can translate into performance in other ways, that's great. But it seems like an odd test to me.

Also an oddly subjective one that is not very easily verifiable. For the record, web browsing is a pure pleasure on all of the iOS devices that I use. I can't say the same for the choppy scrolling on all of the Android devices that I've used, but that's just my experience...

 

[krumme]

Clearly subjective remarks is useless, also because this is fed by giant PR machines. We need the numbers.

But imo considering this is new tech the performance judged by numbers is good and very promising. More competition is always welcome, and a stronger presence from NV in this market is just great for us consumers. Things was comming to a slowdown and this will help to keep up the speed.

2015 is looking good on mobile/tablet side. QQ needs something more than a57 to stay at top. 14nm Atom. 20 & 16nm A57. Denver successor on 20/16nm.