|
Hi PCL,
I am purchasing a laptop to run Kinfu (and other GPU-intensive tasks). I have both Kinect and XTION Pro Live. There are many laptop choices. It sounds like "Kepler" is the best GPU out there, but I'm not sure which laptops have this available (if any). Does PCL/Kinfu run under Kepler? I have identified the following laptops, and they all seem to have good GPUs, but I'm not familiar with the distinctions between one model and the next, nor how big a difference 2 or 3 GB of RAM will make. ASUS G74SX NVIDIA® GeForce® GTX 560M with 3GB GDDR5 VRAM ASUS G75VW NVIDIA® GeForce® GTX 660M/670M with 2GB/3GB GDDR5 VRAM ASUS G55VW NVIDIA® GeForce® GTX 660M with 2GB GDDR5 VRAM IdeaPad Y480 NVIDIA GeForce GT 640M 2GB graphics I've heard stories of some laptops (Sony Viao) requiring proprietary drivers and so the NVidia drivers could not be installed, essentially making the purchase worthless. I hope to avoid this situation. I'll be installing Linux and dual-booting, running GPU code in both Windows and Linux. Any purchase advice will be much appreciated. Thanks, -k. _______________________________________________ [hidden email] / http://pointclouds.org http://pointclouds.org/mailman/listinfo/pcl-users |
|
i will suggest to go for ASUS G75VW + NVIDIA® GeForce® GTX 660M/670M with 2GB/3GB GDDR5 VRAM.
For Linux distro, i will also suggest to use Fedora 17, since the latest Ubuntu 12.04 seems to be buggy. About the Sony Vaio, the information is false, since you should just install the Nvidia GeForce Mobile drivers and the installation completes successfully. By the way, i think you should wait to buy a new notebook, since during June a lot of brand new laptops with Intel Ivy Bridge CPU will be released. |
|
In reply to this post by mankoff
Hi,
I would choose 660M (the fastest Kepler notebook) or a bit slower 670M (Fermi). Wouldn't 560M. Doubt about 640M (640M LE is a slow Fermi, check is you are not buying it). 2GB Memory size is enough. You need to compile PCL with the latest CUDA to get better performance with Kepler. BTW, read this post about improved KinFu for large scale: http://www.pointclouds.org/blog/tocs/all.php Cheers, Anatoly -----Original Message----- From: [hidden email] [mailto:[hidden email]] On Behalf Of Ken Mankoff Sent: Monday, June 04, 2012 11:35 AM To: [hidden email] Subject: [PCL-users] Laptop Purchase for Kinfu Hi PCL, I am purchasing a laptop to run Kinfu (and other GPU-intensive tasks). I have both Kinect and XTION Pro Live. There are many laptop choices. It sounds like "Kepler" is the best GPU out there, but I'm not sure which laptops have this available (if any). Does PCL/Kinfu run under Kepler? I have identified the following laptops, and they all seem to have good GPUs, but I'm not familiar with the distinctions between one model and the next, nor how big a difference 2 or 3 GB of RAM will make. ASUS G74SX NVIDIAR GeForceR GTX 560M with 3GB GDDR5 VRAM ASUS G75VW NVIDIAR GeForceR GTX 660M/670M with 2GB/3GB GDDR5 VRAM ASUS G55VW NVIDIAR GeForceR GTX 660M with 2GB GDDR5 VRAM IdeaPad Y480 NVIDIA GeForce GT 640M 2GB graphics I've heard stories of some laptops (Sony Viao) requiring proprietary drivers and so the NVidia drivers could not be installed, essentially making the purchase worthless. I hope to avoid this situation. I'll be installing Linux and dual-booting, running GPU code in both Windows and Linux. Any purchase advice will be much appreciated. Thanks, -k. _______________________________________________ [hidden email] / http://pointclouds.org http://pointclouds.org/mailman/listinfo/pcl-users _______________________________________________ [hidden email] / http://pointclouds.org http://pointclouds.org/mailman/listinfo/pcl-users |
|
In reply to this post by blackibiza
On Mon, 4 Jun 2012, blackibiza wrote: > i will suggest to go for ASUS G75VW + NVIDIA® GeForce® GTX > 660M/670M with 2GB/3GB GDDR5 VRAM. For Linux distro, i will also > suggest to use Fedora 17, since the latest Ubuntu 12.04 seems to > be buggy. OK. > About the Sony Vaio, the information is false, since you should > just install the Nvidia GeForce Mobile drivers and the > installation completes successfully. Thanks for the info. I will pass along to my friend with the Viao. > By the way, i think you should wait to buy a new notebook, since > during June a lot of brand new laptops with Intel Ivy Bridge CPU > will be released. I've just waited for Kepler. There will always be an upgrade coming. I need to purchase sooner rather than later... -k. _______________________________________________ [hidden email] / http://pointclouds.org http://pointclouds.org/mailman/listinfo/pcl-users |
|
In reply to this post by Anatoly Baksheev
On Mon, 4 Jun 2012, Anatoly Baksheev wrote: > BTW, read this post about improved KinFu for large scale: > http://www.pointclouds.org/blog/tocs/all.php This is GREAT! Thanks for the link. >> I have identified the following laptops, and they all seem to >> have good GPUs, but I'm not familiar with the distinctions >> between one model and the next, nor how big a difference 2 or 3 >> GB of RAM will make. >> >> ASUS G74SX NVIDIAR GeForceR GTX 560M with 3GB GDDR5 VRAM >> ASUS G75VW NVIDIAR GeForceR GTX 660M/670M with 2GB/3GB GDDR5 VRAM >> ASUS G55VW NVIDIAR GeForceR GTX 660M with 2GB GDDR5 VRAM >> IdeaPad Y480 NVIDIA GeForce GT 640M 2GB graphics > > I would choose 660M (the fastest Kepler notebook) or a bit slower > 670M (Fermi). Wouldn't 560M. Doubt about 640M (640M LE is a slow > Fermi, check is you are not buying it). > > 2GB Memory size is enough. You need to compile PCL with the latest > CUDA to get better performance with Kepler. OK thank you for the recommendation. A few more questions. I'm not wedded to the ASUS, that is just what came up when searching for "Gaming Laptop". I see from the NVidia website that there is a 675M too. I don't see mention on their site that 660M is Kepler and 670M is Fermi. What is the 675M? Should I look for a laptop that contains that GPU? Thanks, -k. _______________________________________________ [hidden email] / http://pointclouds.org http://pointclouds.org/mailman/listinfo/pcl-users |
|
If I may chime into the conversation, I am also about to purchase a new PC, and rather start a new thread with virutally the same questions, I figured I'd just add here.
I was wondering if when you say Ubuntu 12.04 is buggy, is it buggy with PCL's Kinfu, or are you saying it is buggy in general right now? I'm okay with waiting for some basic OS functionality to be patched, but I would like to start using KinFu right away on the new PC.
Also, is the GeForce 680 a good card for running Kinfu on Ubuntu (12.04 and 11.10) and Windows? I'd also like to echo the thanks for all the information so far in this thread.
-Andy S. On Mon, Jun 4, 2012 at 2:48 PM, Ken Mankoff <[hidden email]> wrote:
_______________________________________________ [hidden email] / http://pointclouds.org http://pointclouds.org/mailman/listinfo/pcl-users |
|
Ubuntu 12.04 is a bit buggy for the Operating System itself (i get random crashes with System's applications), while Fedora 17 seems to be more stable.
PCL is fine with both. |
|
In reply to this post by aespielberg
Ubuntu 12.04 is a bit buggy for the Operating System itself (i get random crashes with System's applications), while Fedora 17 seems to be more stable.
PCL is fine with both. |
|
In reply to this post by Anatoly Baksheev
Hi,
The 680M, 670M, 675M, and 660M are all too expensive for me. Will this code run (and at approx what FPS?) on a 650M? -k. On Mon, 4 Jun 2012, Anatoly Baksheev wrote: > Hi, > > I would choose 660M (the fastest Kepler notebook) or a bit slower 670M > (Fermi). Wouldn't 560M. Doubt about 640M (640M LE is a slow Fermi, check > is you are not buying it). > > 2GB Memory size is enough. You need to compile PCL with the latest CUDA to > get better performance with Kepler. > > BTW, read this post about improved KinFu for large scale: > http://www.pointclouds.org/blog/tocs/all.php > > > Cheers, > Anatoly > > -----Original Message----- > From: [hidden email] > [mailto:[hidden email]] On Behalf Of Ken Mankoff > Sent: Monday, June 04, 2012 11:35 AM > To: [hidden email] > Subject: [PCL-users] Laptop Purchase for Kinfu > > Hi PCL, > > I am purchasing a laptop to run Kinfu (and other GPU-intensive tasks). I > have both Kinect and XTION Pro Live. There are many laptop choices. It > sounds like "Kepler" is the best GPU out there, but I'm not sure which > laptops have this available (if any). Does PCL/Kinfu run under Kepler? > > I have identified the following laptops, and they all seem to have good > GPUs, but I'm not familiar with the distinctions between one model and the > next, nor how big a difference 2 or 3 GB of RAM will make. > > ASUS G74SX NVIDIAR GeForceR GTX 560M with 3GB GDDR5 VRAM > ASUS G75VW NVIDIAR GeForceR GTX 660M/670M with 2GB/3GB GDDR5 VRAM > ASUS G55VW NVIDIAR GeForceR GTX 660M with 2GB GDDR5 VRAM > IdeaPad Y480 NVIDIA GeForce GT 640M 2GB graphics > > I've heard stories of some laptops (Sony Viao) requiring proprietary > drivers and so the NVidia drivers could not be installed, essentially > making the purchase worthless. I hope to avoid this situation. > > I'll be installing Linux and dual-booting, running GPU code in both > Windows and Linux. > > Any purchase advice will be much appreciated. > > Thanks, > > -k. > _______________________________________________ > [hidden email] / http://pointclouds.org > http://pointclouds.org/mailman/listinfo/pcl-users > _______________________________________________ [hidden email] / http://pointclouds.org http://pointclouds.org/mailman/listinfo/pcl-users |
|
yes
-----Original Message----- From: Ken Mankoff [mailto:[hidden email]] Sent: Friday, June 08, 2012 8:20 PM To: Anatoly Baksheev Cc: Point Cloud Library (PCL) users Subject: RE: [PCL-users] Laptop Purchase for Kinfu Hi, The 680M, 670M, 675M, and 660M are all too expensive for me. Will this code run (and at approx what FPS?) on a 650M? -k. On Mon, 4 Jun 2012, Anatoly Baksheev wrote: > Hi, > > I would choose 660M (the fastest Kepler notebook) or a bit slower 670M > (Fermi). Wouldn't 560M. Doubt about 640M (640M LE is a slow Fermi, > check is you are not buying it). > > 2GB Memory size is enough. You need to compile PCL with the latest > CUDA to get better performance with Kepler. > > BTW, read this post about improved KinFu for large scale: > http://www.pointclouds.org/blog/tocs/all.php > > > Cheers, > Anatoly > > -----Original Message----- > From: [hidden email] > [mailto:[hidden email]] On Behalf Of Ken Mankoff > Sent: Monday, June 04, 2012 11:35 AM > To: [hidden email] > Subject: [PCL-users] Laptop Purchase for Kinfu > > Hi PCL, > > I am purchasing a laptop to run Kinfu (and other GPU-intensive tasks). > I have both Kinect and XTION Pro Live. There are many laptop choices. > It sounds like "Kepler" is the best GPU out there, but I'm not sure > which laptops have this available (if any). Does PCL/Kinfu run under > > I have identified the following laptops, and they all seem to have > good GPUs, but I'm not familiar with the distinctions between one > model and the next, nor how big a difference 2 or 3 GB of RAM will make. > > ASUS G74SX NVIDIAR GeForceR GTX 560M with 3GB GDDR5 VRAM > ASUS G75VW NVIDIAR GeForceR GTX 660M/670M with 2GB/3GB GDDR5 VRAM > ASUS G55VW NVIDIAR GeForceR GTX 660M with 2GB GDDR5 VRAM > IdeaPad Y480 NVIDIA GeForce GT 640M 2GB graphics > > I've heard stories of some laptops (Sony Viao) requiring proprietary > drivers and so the NVidia drivers could not be installed, essentially > making the purchase worthless. I hope to avoid this situation. > > I'll be installing Linux and dual-booting, running GPU code in both > Windows and Linux. > > Any purchase advice will be much appreciated. > > Thanks, > > -k. > _______________________________________________ > [hidden email] / http://pointclouds.org > http://pointclouds.org/mailman/listinfo/pcl-users > [hidden email] / http://pointclouds.org http://pointclouds.org/mailman/listinfo/pcl-users |
|
This post has NOT been accepted by the mailing list yet.
I am running kinfu on an ASUS G74ST with the 560M and getting about 8 fps. I have to have the portability so performance had to suffer. Too bad Iam not quite smart enough to build kinfu with color or larger voxel grid.
|
|
In reply to this post by Anatoly Baksheev
I realize that this thread is a wee bit dated, but I'm faced with the same question - and have a few questions/clarifications that I would love to have some light shed on.
1. I notice that you mention: "660M (the fastest Kepler notebook) or a bit slower 670M (Fermi)" Is this stating that 660M is among the faster of the Kepler notebooks, and 670M among the slower Fermi, or that the 660M is likely to outperform the 670M overall? - and does this remain true 2. The laptop in question mentioned also is sold in a version with 3D display capabilities - does this have the possibility of causing headaches insofar as running any of the libraries is concerned. (i.e. if one creates a reconstruction but with any and all 3D display disabled, is there a possibility that trouble may crop up) 3. Given that 3 months or so have passed since the start of this discussion, are there some alternatives with similar range processor/GPU (or better possibly) that anyone might recommend? Any insight into any of the preceding would be very much appreciated. |
|
I am running a modified version of kinfu on a 4GB 680m and am getting roughly 20 fps. This is on a windows 7 laptop driver version 306.23 and cuda sdk 4.2.
My modifications have been to increase enum { VOLUME_X = 512, VOLUME_Y = 512, VOLUME_Z = 512 }; to enum { VOLUME_X = 768, VOLUME_Y = 768, VOLUME_Z = 768}; in internal.h to take advantage of the 4GB and by tweeking kinfu_.volume().setTsdfTruncDist (0.0025f/*meters*/); kinfu_.setIcpCorespFilteringParams (0.005f/*meters*/, sin ( pcl::deg2rad(20.f) )); kinfu_.setDepthTruncationForICP(0.7f/*meters*/); kinfu_.setCameraMovementThreshold(0.02f); in kinfu_app.cpp and running the code with a volume_size of 0.7 Vanilla kinfu (without the tweeks) with default volume_size gives me slightly less at around the 18-22 fps. As a side note, I noticed that the reported frame rate reported by kinfu is not quite right. It correctly reports how long it took kinfu to register the current frame to the previous frame, but does not include the other parts of kinfu_app such as spinning the meshes to screen etc. So even though it may be reporting 30 fps or 33ms the other "stuff" takes up another 20-30ms per cycle so you are only getting 15-20 actual frames of Kinect data per second. I noticed this when trying to plot the camera movement relative to time and noticing some large discrepancies. Anyway, hope this helps! Phil |
|
In reply to this post by bruce
And FYI I was never able to get this to work in Linux. The Nvidia 6XXM series (M for Mobile) have some "Optimus" technology which isn't fully supported. I can get CUDA working, and OpenGL working (using Bumblebee) but not both at the same time, so I could not run Kinfu. I'm currently trying to get that to work in Windows 7 where there is full driver support... -k. On Sun, 23 Sep 2012, bruce wrote: > I realize that this thread is a wee bit dated, but I'm faced with the same > question - and have a few questions/clarifications that I would love to have > some light shed on. > > 1. I notice that you mention: > > "660M (the fastest Kepler notebook) or a bit slower 670M (Fermi)" > > Is this stating that 660M is among the faster of the Kepler notebooks, and > 670M among the slower Fermi, or that the 660M is likely to outperform the > 670M overall? - and does this remain true > > 2. The laptop in question mentioned also is sold in a version with 3D > display capabilities - does this have the possibility of causing headaches > insofar as running any of the libraries is concerned. (i.e. if one creates a > reconstruction but with any and all 3D display disabled, is there a > possibility that trouble may crop up) > > 3. Given that 3 months or so have passed since the start of this discussion, > are there some alternatives with similar range processor/GPU (or better > possibly) that anyone might recommend? > > Any insight into any of the preceding would be very much appreciated. > > > > -- > View this message in context: http://www.pcl-users.org/Laptop-Purchase-for-Kinfu-tp4018905p4022491.html > Sent from the Point Cloud Library (PCL) Users mailing list archive at Nabble.com. > _______________________________________________ > [hidden email] / http://pointclouds.org > http://pointclouds.org/mailman/listinfo/pcl-users > _______________________________________________ [hidden email] / http://pointclouds.org http://pointclouds.org/mailman/listinfo/pcl-users |
|
A tip in windows with Optimus enabled GPU's is to manually force windows to always use the NVidia card and in the NVidia control panel set "Power Management Mode" as Prefer maximum performance.
A while back I couldn't understand why Kinfu would run for a couple of seconds at a steady 22fps then drop to a stead 15-16fps for a while before jumping back up. After changing the settings mentioned above the code remained steady at the higher level, which was nice. Phil |
|
The Optimus technology isn't on all 6xxM cards, I've used a machine with a 680M that didn't have Optimus.
Also Phil if you want performance drop back to 512x512x512 (or even 384x384x384). At 768x768x768 you have the GPU doing over three times as much work per frame (8 times as much as 384) on integration (significant) and raycasting (not so slow) steps. Tom On Mon, Sep 24, 2012 at 4:28 PM, philnoonan <[hidden email]> wrote: A tip in windows with Optimus enabled GPU's is to manually force windows to _______________________________________________ [hidden email] / http://pointclouds.org http://pointclouds.org/mailman/listinfo/pcl-users |
|
Tom were you able to get Kinect Fusion working on that laptop in Ubuntu? I currently have it working (offline and at 384 res) on my old Toshiba laptop with a GeForce 525M, but need to get it running on line. I have a reasonable amount of money to spend (~$3000US). So far my best guess for something that would work well is one of these.
However I am pretty paranoid about getting something that won't work in Ubuntu... is Optimus all I have to look out for? Or do I have to be careful with the specific brand and even the model? Ideally I would get one with one of the best mobile graphics cards (680MX, 680M or 675MX) however I would be prepared to go for slightly less power if I knew someone had already got it to work on Ubuntu.
|
|
This is one of the machines I've ran it on before;
http://www.govconnection.com/IPA/Shop/Product/Detail.htm?sku=14251530&cac=Result A fairly hefty spec and the 675M in that machine doesn't have Optimus, works fine in Ubuntu. Tom On Mon, Nov 26, 2012 at 4:04 AM, amploid <[hidden email]> wrote: Tom were you able to get Kinect Fusion working on that laptop in Ubuntu? I _______________________________________________ [hidden email] / http://pointclouds.org http://pointclouds.org/mailman/listinfo/pcl-users |
|
Thanks Tom, that's the best laptop I am aware of so far for which someone has been able to get Kinfu running in Linux.
However, since I have (some) money to burn, I am interested in the following machine: Eurocom Scorpius Particularly because of the following line: "With no Optimus support the EUROCOM Scorpius is all about extreme graphics performance with no compromise!" I guess in the next few days I have to figure out how much of a risk I am willing to take on it, unless I can find someone else who has tried this machine or something else as good. <quote author="Thomas Whelan"> |
|
Hi,
I read the thread and it is very helpful, thanks. This is going to be a bit repetitive but I want to ask a similar question. I want to run Kinfu on Ubuntu (probably 12.04). I have to buy a Dell since my department only supports it. I found one with the following one: Inspiron 7720 with NVIDIA GeForce GT 650M GDDR5 2GB Would this one be ok to use, I certainly don't want to buy something totally useless. Have anyone tried to use a Dell before? Thanks again for all the information in this thread, Burcu |
| Powered by Nabble | Edit this page |
