Tweaktown : Nvidia Jetson (Tegra K1)

[TrulyUncouth]

Is it too late to get popcorn? Nerds everywhere fighting about everything yay

I swear I didn't pick a fight, this guy is just crazy. Who claims something, then when called out on it says they have a secret source? I don't even dislike the SOC/Tablet he is so eager to defend... I just want to know if it is actually able to run a game at the settings he claims, I swear I feel like I'm taking crazy pills.

The more I look at it I am thinking of buying a shield because it is a great piece of hardware... but I think its perfectly normal to expect a source when someone claims something you haven't read in the many articles you've read or seen in the hours of video you've watched.

Oh well, the story here is Shield tablet great, NV fans... not so much

Heading to work so, good night to everyone.

 

[TrulyUncouth]

skyrim is a nonstarter.

I have to head to work, but just wanted to tell you I think I remember one of the Eurogamer articles saying Skyrim was playable, or maybe it was in a link in the article to a forum of people posting their experiences on the T100... I am gonna try my damndest, hopefully there is a collection of mods that can get it working.

 

[ams23]

I swear I didn't pick a fight, this guy is just crazy.

Look who's talking. First you were skeptical that mobile Kepler (ie. the most advanced and powerful ultra mobile GPU on the market today!) could render Half Life 2 and Portal at 1080p with higher detail settings using the full OpenGL code path compared to the lower IQ OpenGL ES2 code path (which is pretty crazy in and of itself). When you asked for a source, it was very clearly explained to you that it was from an online preview or video for Shield tablet. You couldn't accept that as an answer, were too lazy to do the extra reading and research yourself, and started crying and whining about it to demand a direct link. So the only one here who is confused and even petulant is you. Seriously, this is one of the stupidest conversations ever, because anyone with half a brain knows how good mobile Kepler is by now. If you bothered to do some homework first instead of arguing about something you don't know anything about, then there wouldn't be any of this nonsensical back and forth here.

In order to stop feeding the you-know-what, I will post the link here: http://www.youtube.com/watch?v=RfCLsvEpwHs

 

[antihelten]

Look who's talking. First you were skeptical that mobile Kepler (ie. the most advanced and powerful ultra mobile GPU on the market today!) could render Half Life 2 and Portal at 1080p with higher detail settings using the full OpenGL code path compared to the lower IQ OpenGL ES2 code path (which is pretty crazy in and of itself). When you asked for a source, it was very clearly explained to you that it was from an online preview or video for Shield tablet. You couldn't accept that as an answer, were too lazy to do the extra reading and research yourself, and started crying and whining about it to demand a direct link. So the only one here who is confused and even petulant is you. Seriously, this is one of the stupidest conversations ever, because anyone with half a brain knows how good mobile Kepler is by now. If you bothered to do some homework first instead of arguing about something you don't know anything about, then there wouldn't be any of this nonsensical back and forth here.

In order to stop feeding the you-know-what, I will post the link here: http://www.youtube.com/watch?v=RfCLsvEpwHs

link to quote on Half life 2 and portal (I'm assuming this is the part ams23 is referring to):
http://www.youtube.com/watch?v=RfCLsvEpwHs&t=3m46s

transcript:

"but we have actually taken it a step further with the shield tablet. So for one they both run at 1080p, and they run at much higher frame rates, they also use the full OpenGL render path, not OpenGL ES"

 

[SPBHM]

Sorry if it's a repost, but here is Tegra K1 running a few games, including HL2 with "fraps"

http://www.eurogamer.net/articles/digitalfoundry-2014-xiaomi-mipad-review

http://www.youtube.com/watch?v=QC0TH5jUd00

Hl2 looks OK but the framerate is not always higher than 30, I guess Android is not so great, also the CPU performance is questionable compared to what the game was made to use originally!? (fast single core), they copied the game from a Tegra 4 shield, but it clearly have some K1 optimizations going on.

 

[Nothingness]

Hl2 looks OK but the framerate is not always higher than 30, I guess Android is not so great, also the CPU performance is questionable compared to what the game was made to use originally!? (fast single core), they copied the game from a Tegra 4 shield, but it clearly have some K1 optimizations going on.

HL2 recommended CPU was 3 GHz P4, so that should be OK CPU wise (look at this GB3 results with a grain of salt as usual).

EDIT: BTW the Shield Tablet is now the fastest Android device on Geekbench 3.

 

[TrulyUncouth]

See, this is why people, Smartly, ask for sources. Let me remind you of your original statement...

Shield tablet runs Half Life 2 and Portal...smoothly at 1080p with high details

See a big difference here? Now you have posted a link to a guy simply saying it uses the OpenGL render path... Not a single syllable about high settings, nothing, nada...

You can try to change what you and I said as much as you'd like- fact is its all available in black and white. I never said TK1 couldn't render at higher quality than T4- ever.

So next time you make something up with no sources, then attack someone for asking where you heard it- make sure its not me... because I will not give you a free pass

And I have not been "Confused and even petulant", the only thing I think you could even get a little malice out of is my annoyance at you trying to put the burden of proof on me. And this is a perfect example of why we have a simple logic device like burden of proof. You claimed HL2 and Portal run smoothly on high settings at 1080p. You not only failed to provide proof, but then tried to change your argument and post a video that doesn't back you up at all.

I think you should re-read what I've written. I am a big fan of TK1 and the shield tablet/controller especially- I had/have some questions about certain aspects and feel we should able to discuss them here without anyone getting their panties in a ruffle.

I am going to report your post, simply because it brings a ton of vitriol and derails the thread even further. I apologize for anything offensive I've done as it was not my intention. We both clearly have similar interests in the mobile gaming space and I hope our disagreements here don't color our future interactions.

In closing, game on brother.

Look who's talking. First you were skeptical that mobile Kepler (ie. the most advanced and powerful ultra mobile GPU on the market today!) could render Half Life 2 and Portal at 1080p with higher detail settings using the full OpenGL code path compared to the lower IQ OpenGL ES2 code path (which is pretty crazy in and of itself). When you asked for a source, it was very clearly explained to you that it was from an online preview or video for Shield tablet. You couldn't accept that as an answer, were too lazy to do the extra reading and research yourself, and started crying and whining about it to demand a direct link. So the only one here who is confused and even petulant is you. Seriously, this is one of the stupidest conversations ever, because anyone with half a brain knows how good mobile Kepler is by now. If you bothered to do some homework first instead of arguing about something you don't know anything about, then there wouldn't be any of this nonsensical back and forth here.

In order to stop feeding the you-know-what, I will post the link here: http://www.youtube.com/watch?v=RfCLsvEpwHs

 

[TrulyUncouth]

HL2 recommended CPU was 3 GHz P4, so that should be OK CPU wise (look at this GB3 results with a grain of salt as usual).

EDIT: BTW the Shield Tablet is now the fastest Android device on Geekbench 3.

I really hope someone starts running some benchmarks with the core loading overlay- I'd be interested to see if its just a heavily single-threaded game. I really want to know if this is all a GPU or CPU bound situation. Considering the geekbench results for single thread are not a complete stomp in favor of the A15.

Throw in inefficiencies considering this is ported from x86 and it could simply be a game that doesn't thread well. I guess once Denver comes out we will see if framerates jump on these source games.

Mount and blade looks like it was ported much better. Considering all the AI that game leans heavily on it runs very well on the old Shield and I can only imagine is a dream on the TK1 tablet. If anyone finds a video of it running on the new shield- please post.

Edit: Also, I think its worth mentioning here that the Gamestreaming only works at 720p, unless you plug in an ethernet cable- I haven't seen anything about a USB/ethernet adapter but I would assume generic ones work.

 

[Nothingness]

I really hope someone starts running some benchmarks with the core loading overlay- I'd be interested to see if its just a heavily single-threaded game. I really want to know if this is all a GPU or CPU bound situation. Considering the geekbench results for single thread are not a complete stomp in favor of the A15.

Throw in inefficiencies considering this is ported from x86 and it could simply be a game that doesn't thread well. I guess once Denver comes out we will see if framerates jump on these source games.

Mount and blade looks like it was ported much better. Considering all the AI that game leans heavily on it runs very well on the old Shield and I can only imagine is a dream on the TK1 tablet. If anyone finds a video of it running on the new shield- please post.

It looks like HL2 was ported using a DX -> GL runtime translator:

$ nm -DCv libtogl.so
...
000000000001d330 T IDirect3D9::~IDirect3D9()
000000000001d330 T IDirect3D9::~IDirect3D9()
000000000001d364 T IDirect3D9::~IDirect3D9()
000000000001d4b0 T D3DXMATRIX::operator float*()
000000000001d4c4 T D3DXMATRIX::operator()(int, int)
000000000001d4e0 T D3DXMATRIX::operator()(int, int) const
000000000001d4fc T D3DXPLANE::operator[](int)
000000000001d514 T D3DXPLANE::operator==(D3DXPLANE const&)
000000000001d58c T D3DXPLANE::operator!=(D3DXPLANE const&)
000000000001d5a4 T D3DXPLANE::operator float*()
000000000001d5b8 T D3DXPLANE::operator float const*() const
000000000001d5cc T D3DXVECTOR2::operator float*()
000000000001d5e0 T D3DXVECTOR2::operator float const*() const
000000000001d5f4 T D3DXVECTOR3::D3DXVECTOR3(float, float, float)
000000000001d5f4 T D3DXVECTOR3::D3DXVECTOR3(float, float, float)
000000000001d614 T D3DXVECTOR3::operator float*()
000000000001d628 T D3DXVECTOR3::operator float const*() const
000000000001d63c T D3DXVECTOR4::D3DXVECTOR4(float, float, float, float)
000000000001d63c T D3DXVECTOR4::D3DXVECTOR4(float, float, float, float)
000000000001d670 T IDirect3DResource9::SetPriority(unsigned int)
000000000001d688 T IDirect3DBaseTexture9::GetType()
000000000001d6a0 T IDirect3DBaseTexture9::GetLevelCount()
000000000001d6c0 T IDirect3DBaseTexture9::GetLevelDesc(unsigned int, _D3DSURFACE_DESC*)
000000000001d748 T IDirect3DDevice9::CreateTexture(unsigned int, unsigned int, unsigned int, unsigned int, _D3DFORMAT, _D3DPOOL, IDirect3DTexture9**, void**, char*)
000000000001d984 T IDirect3DTexture9::LockRect(unsigned int, _D3DLOCKED_RECT*, _RECT const*, unsigned int)
000000000001d99c T IDirect3DTexture9::UnlockRect(unsigned int)
000000000001d9b4 T IDirect3DTexture9::GetSurfaceLevel(unsigned int, IDirect3DSurface9**)
000000000001daa0 T IDirect3DDevice9::CreateCubeTexture(unsigned int, unsigned int, unsigned int, _D3DFORMAT, _D3DPOOL, IDirect3DCubeTexture9**, void**, char*)
...

Note that layer should be good if we are to believe Valve results for Linux Left 4 Dead 2, but that was true for Linux x86 and might not be that good for Android ARM.

 

[jdubs03]

@nothingness

good find on the geekbench. that result for both single and multi-thread makes sense, and kind of proves for me that denver will be over 2000 for single-thread (haswell-y perf). also based on this result its ~5.5% faster than S805 (not even in devices yet), so that should be the king for the next year.

edit: on second thought, after seeing the core i3-y (11.5W) data from the surface pro 3, i'm not as optimistic for TK1 once cherry trail comes out, at least gpu wise.

 

[ViRGE]

After spending the last half of an hour going through this thread there are a number of posts by a handful of posters that fall well below the kind of conduct we expect in a technical forum. Infractions and vacations have been handed out accordingly, in a reminder of what kind of conduct we expect in these forums.

As for the rest of you: this is the CPUs forum, so we would ask that the subject stick to the CPU. If you wish to discuss the platform or software, that discussion needs to happen in Mobile Devices.

-Thanks
ViRGE

 

[jdubs03]

here is some more benchmarks compiled; higher scores than i have seen anywhere.

http://www.notebookcheck.net/NVIDIA-Shield-Tablet-with-Tegra-K1-and-Controller-announced.121871.0.html

good to see some data points trickling out.

 

[jpiniero]

2-3 hours battery life gaming?

 

[sontin]

2-3 hours battery life gaming?

That comes directly from nVidia. 2-3h while playing games like Trine 2. Mostly 5h with less demanding games and up to 10h with games like Angry Birds.

 

[sontin]

Battery life is great for a SoC which consumes 30W+

 

[NTMBK]

Battery life is great for a SoC which consumes 30W+

And when you try to use that powerful GPU, what happens?

Throttling doesn't look too bad though:

Looks like that magnesium head spreading thingy is actually doing the job quite well.

 

[tential]

Still looks quite good for gaming. I'm excited for it for some reason I think it will be very fun to use. Makes me wish I had an Nvidia card to use it with. If I had this device I'd beat a LOT more single player games than I currently have.

 

[Erenhardt]

Good handheld for winter. They should definitely wait with the release!

 

[jdubs03]

Good handheld for winter. They should definitely wait with the release!

Makes me wonder how the Nexus 9 will be, as long as it has Denver in it; the S805 would just be a downgrade aside from integrated LTE. I'd expect 4GB of RAM with that too.

 

[Khato]

And when you try to use that powerful GPU, what happens?

Ayup. And having both the web browsing and gaming battery life figures gives us a pretty good idea of how much power the SoC will draw under that load. Namely, 10.828 hours on a 19.75 Wh battery equates to 1.824W for the entire tablet under 'active idle' conditions while 2.246 hours on that same battery equates to 8.793W for a delta which can be placed entirely upon SoC + memory of 6.97W. So yes, K1 is definitely burning through over 6W on GFXBench 3.0 - sadly I haven't seen figures regarding how active the CPU is in that benchmark so no idea on the CPU/GPU balance in that figure/how much higher it could theoretically go if all four CPU cores were fully loaded.

Either way pretty much as expected - excellent performance with high power usage to match.

 

[sontin]

The T-Rex Battery test runs unlimited. So the Shield Tables archives 2,2h with 60FPS.
The Shield Portable number is wrong.

Galaxy Tab gets 23FPS and archives a time of 3,367hrs
Or: Shield Tablet is 3x faster while having 61% of the runtime.

 

[Homeles]

It's out in time for holiday season... definitely a good thing for Nvidia. CPU performance isn't likely to hold the crown for long, though.

 

[dawheat]

The T-Rex Battery test runs unlimited. So the Shield Tables archives 2,2h with 60FPS.
The Shield Portable number is wrong.

Galaxy Tab gets 23FPS and archives a time of 3,367hrs
Or: Shield Tablet is 3x faster while having 61% of the runtime.

Pretty impressive - competitive in CPU and dominant in GPU compared to other Android SOCs. Battery life in non-gaming tasks also seem quite competitive (surfing, video).

Seriously - what happened with Qualcomm this year. They went from hands down the best Android SOCs in 2013, especially when the S800 first came out in the summer of 2013. Since then, it feels like they've really stagnated. S801 and S805 are incrementally faster, but they've left the door open for competitors to come back.

 

[jdubs03]

It's out in time for holiday season... definitely a good thing for Nvidia. CPU performance isn't likely to hold the crown for long, though.

Quad-core CPU and GPU performance will be the highest for Android until Cherry Trail comes out in Q4 this year, so a solid ~4-5 months at least. We'll see what happens with the S810 in late Q1-Q2, my guess is it won't dethrone CT. Erista will probably take the throne again after CT, with Maxwell cores.

Denver should easily retain its lead in single-thread performance over everything ARM based, and maybe even match Core 5Y70.

 

[Homeles]

Quad-core CPU and GPU performance will be the highest for Android until Cherry Trail comes out in Q4 this year, so a solid ~4-5 months at least. We'll see what happens with the S810 in late Q1-Q2, my guess is it won't dethrone CT. Erista will probably take the throne again after CT, with Maxwell cores.

Denver should easily retain its lead in single-thread performance over everything ARM based, and maybe even match Core 5Y70.

A8 is due in ~2 months. Not a direct competitor with Tegra, of course, but it should beat it the TK1 with ease. Cherry Trail might be shown off at IDF in September as well, although it wouldn't be available for another while as you mentioned.

It'll be interesting to see what Qualcomm does, as they're quite behind right now. Samsung's custom Exynos 6 isn't far off either, so Qualcomm's not going to have it easy in the coming year.