laptop gfx card for opencl : GTX 480M or ATI HD5870 ? (dual card or not ?)

[ker2x]

Friendly greetings !
I'm willing to buy a (big (expensive)) laptop for OpenCL programming, from : http://www.clevo.com.tw/

I'm used to be NVidia only, but the ATI 5870 seems to be very, very, very powerfull (No cuda, optix, physX, ... but i never took time to use it thoses NVidia only technology anyway).

It's much cheaper too. i can have (crossfire) 2xHD4870 for cheaper than a single 480M. (yes, clevo build some insane laptop with 2 gfx card).

I don't know ATI cards (and ATI Stream) well enough to choose between both solution. Can you help please ?

Thank you

[hobold]

If the primary purpose is GPGPU programming, then IMHO a laptop is not a good choice. For one, you cannot upgrade GPUs in a laptop. And secondly, mobile GPUs are much more power constrained than their desktop counterparts. You are allowed to pay for a top of the line GPU, but you can't really run it at full speed. Up to 250 Watts of power are converted into heat ... in the confined space of a laptop ... not for long.

Nvidia and AMD have been following opposite hardware design philosophies. Nvidia spends quite a bit of effort to utilize the computational units as efficiently as possible, while AMD just packs more brute force. Theoretically, the 5870 has roughly double the peak computational performance of a 480. In practice it's a wash: comparable net performance across a wide range of graphics and GPGPU benchmarks, comparable number of transistors, and only slight power consumption advantages for AMD.

You can't really go wrong with either. Especially if you are not locked into Nvidia's exclusive CUDA infrastructure.

[lycium]

Gotta agree with hobold (greetz fellow Ompf'er!), a laptop is fundamentally unsuited to high performance computing applications.

For GPU programming I'd say the Fermi architecture has a decisive edge over the 5xxx / 6xxx chip series from AMD, however be aware that double precision throughput is crippled in GeForce cards (cf the Tesla range) if that matters to you.

[cbuchner1]

Friendly greetings !
I'm willing to buy a (big (expensive)) laptop for OpenCL programming, from : http://www.clevo.com.tw/

Using a Clevo with GTX 295M graphics chip here at work equipped with a Core i7 for high performance mobile radio simulation.
The GPU has 128 shader cores and is essentially a G92M chip (roughtly comparable to nVidia 8800GT). After months of testing and frustration we've finally got this platform working.

On Linux there were several problems (each one being a potential showstopper).

Only recent nVidia drivers starting at 256.19 would ever clock the GPU to its nominal speed. Previous versions were having an ACPI related problem and always used a very low GPU clock rate (400MHz). This problem lingered for over a year until nVidia finally fixed it! Luckily we bought the hardware only about 1 month before the fix was made available.

The BIOS on this graphics module has trouble detecting externally connected monitors on the analog VGA output. It always
switches off the internal LCD panel and misdetects the external monitor that is connected. We had to apply some nasty overrides in the video driver to get a correct image.

Too old Linux kernels won't detect the power management features on the Core i7 properly, so Turbo Boost would not function. Just run a recent distribution (the patch was submitted to the Linux kernel around January 2010) or apply a kernel patch manually.

Oh and any Linux support with the vendor (Clevo or our local distributor MySN) is entirely non existent, it seems.

I cannot comment on Windows, as we've never tried to install Windows on this machine.

Meanwhile you can get GTX 470M/480M from Clevo, but not in the smallest form factor PCs they sell (the 15 inch platform we have here, based on the Clevo W860CU barebones platform). The GTX 460M with its 192 shaders seems to be available for the 15 inch models now.

Instead I'd recommend getting a low cost PC equipped with a 32 or 48 shader GPU for development on the road, and a good desktop PC with plenty of shaders for production runs, unless you absolutely require the high performance moblity. And you get two PCs for the price of one Clevo

I hope this helps choosing right.

[hobold]

(greetz fellow Ompf'er!)

Yes, that's indeed me over there, too. I have nothing of interest to contribute there. But maybe my raytracer number N + 1 (currently in slow production) might progress far enough that it will be worth talking about. Someday.

Back to topic. I should probably add that I am currently in the AMD/ATI camp, and that I expect Nvidia to change beyond recognition over the next two years. Separate graphics cards will become a rarity when both Intel and AMD integrate on die GPUs with their upcoming next CPU generation. Unfortunately, OpenCL is not yet ready to take over if CUDA were to be dragged down by an agonizing Nvidia. Unpredictable times ahead, and rumours of Nvidia's death could end up being wildly exaggerated. I am happy that I don't have to make hard decisions about basing a commercial product on GPGPU codes.

For a hobbyist, though, both OpenCL(AMD) and CUDA(Nvidia) are equally good or bad. Lycium is probably right that Nvidia has an edge for floating point number crunching. On the other hand, Fermi is a bit handicapped with regards to atomic operations (i.e. synchronization primitives). I find it very hard to predict which hardware is going to do better for a specific algorithm. GeForces and Radeons are somewhat complementary in their specific strengths and weaknesses. It's all the more surprising that neither is consistently ahead of the other.

[ker2x]

Thank you for your comments

Why a laptop :
- Yes, i want some kind of mobility (home<->work, nothing more).
- it's hard to upgrade my desktop (with a 8800GTX)
- i noticed that i mostly develop on my tiny Eeepc 12" (with a ion2 gpu)

I choosen the 18" laptop with Dual HD5870 (poweeer).
The autonomy will be very bad (150W just for the 2 GPU) but it's not a problem.
It will be large and heavy, but when i really want to be mobile i use my Eeepc

Thank you again

[ker2x]

i finally ordered my laptop today.
it will be a 15" with 460M 1.5GB.

I will use it at work for some stuff that require high cpu/memory.
That's why i took a smaller card, and a bigger cpu + 8GB of ram.

thank you for your advices, the choice was very hard 

[ker2x]

received !

As a side note, the nvidia sdk for linux is certified for some version of distro. it's certified for ubuntu 10.04, but it work without any problem on 10.10  

[ker2x]

using the latest stable driver and sdk (as far as i know) on ubuntu 10.10 with a GTX460M

Code:

./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

There is 1 device supporting CUDA

Device 0: "GeForce GTX 460M"
  CUDA Driver Version:                           3.20
  CUDA Runtime Version:                          3.20
  CUDA Capability Major/Minor version number:    2.1
  Total amount of global memory:                 1609760768 bytes
  Multiprocessors x Cores/MP = Cores:            4 (MP) x 48 (Cores/MP) = 192 (Cores)
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 32768
  Warp size:                                     32
  Maximum number of threads per block:           1024
  Maximum sizes of each dimension of a block:    1024 x 1024 x 64
  Maximum sizes of each dimension of a grid:     65535 x 65535 x 1
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Clock rate:                                    1.35 GHz
  Concurrent copy and execution:                 Yes
  Run time limit on kernels:                     Yes
  Integrated:                                    No
  Support host page-locked memory mapping:       Yes
  Compute mode:                                  Default (multiple host threads can use this device simultaneously)
  Concurrent kernel execution:                   Yes
  Device has ECC support enabled:                No
  Device is using TCC driver mode:               No

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 3.20, CUDA Runtime Version = 3.20, NumDevs = 1, Device = GeForce GTX 460M

Code:

./bandwidthTest Starting...

Running on...

 Device 0: GeForce GTX 460M
 Quick Mode

 Host to Device Bandwidth, 1 Device(s), Paged memory
   Transfer Size (Bytes)	Bandwidth(MB/s)
   33554432			4573.6

 Device to Host Bandwidth, 1 Device(s), Paged memory
   Transfer Size (Bytes)	Bandwidth(MB/s)
   33554432			3706.7

 Device to Device Bandwidth, 1 Device(s)
   Transfer Size (Bytes)	Bandwidth(MB/s)
   33554432			30042.8

Code:

oclDeviceQuery.exe Starting...

OpenCL SW Info:

 CL_PLATFORM_NAME: 	NVIDIA CUDA
 CL_PLATFORM_VERSION: 	OpenCL 1.0 CUDA 3.2.1
 OpenCL SDK Revision: 	7027912


OpenCL Device Info:

 1 devices found supporting OpenCL:

 ---------------------------------
 Device GeForce GTX 460M
 ---------------------------------
  CL_DEVICE_NAME: 			GeForce GTX 460M
  CL_DEVICE_VENDOR: 			NVIDIA Corporation
  CL_DRIVER_VERSION: 			260.19.21
  CL_DEVICE_VERSION: 			OpenCL 1.0 CUDA
  CL_DEVICE_TYPE:			CL_DEVICE_TYPE_GPU
  CL_DEVICE_MAX_COMPUTE_UNITS:		4
  CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS:	3
  CL_DEVICE_MAX_WORK_ITEM_SIZES:	1024 / 1024 / 64 
  CL_DEVICE_MAX_WORK_GROUP_SIZE:	1024
  CL_DEVICE_MAX_CLOCK_FREQUENCY:	1350 MHz
  CL_DEVICE_ADDRESS_BITS:		32
  CL_DEVICE_MAX_MEM_ALLOC_SIZE:		383 MByte
  CL_DEVICE_GLOBAL_MEM_SIZE:		1535 MByte
  CL_DEVICE_ERROR_CORRECTION_SUPPORT:	no
  CL_DEVICE_LOCAL_MEM_TYPE:		local
  CL_DEVICE_LOCAL_MEM_SIZE:		48 KByte
  CL_DEVICE_MAX_CONSTANT_BUFFER_SIZE:	64 KByte
  CL_DEVICE_QUEUE_PROPERTIES:		CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE
  CL_DEVICE_QUEUE_PROPERTIES:		CL_QUEUE_PROFILING_ENABLE
  CL_DEVICE_IMAGE_SUPPORT:		1
  CL_DEVICE_MAX_READ_IMAGE_ARGS:	128
  CL_DEVICE_MAX_WRITE_IMAGE_ARGS:	8
  CL_DEVICE_SINGLE_FP_CONFIG:		denorms INF-quietNaNs round-to-nearest round-to-zero round-to-inf fma 

  CL_DEVICE_IMAGE <dim>			2D_MAX_WIDTH	 4096
					2D_MAX_HEIGHT	 32768
					3D_MAX_WIDTH	 2048
					3D_MAX_HEIGHT	 2048
					3D_MAX_DEPTH	 2048

  CL_DEVICE_EXTENSIONS:			cl_khr_byte_addressable_store
					cl_khr_icd
					cl_khr_gl_sharing
					cl_nv_compiler_options
					cl_nv_device_attribute_query
					cl_nv_pragma_unroll
					cl_khr_global_int32_base_atomics
					cl_khr_global_int32_extended_atomics
					cl_khr_local_int32_base_atomics
					cl_khr_local_int32_extended_atomics
					cl_khr_fp64


  CL_DEVICE_COMPUTE_CAPABILITY_NV:	2.1
  NUMBER OF MULTIPROCESSORS:		4
  NUMBER OF CUDA CORES:			192
  CL_DEVICE_REGISTERS_PER_BLOCK_NV:	32768
  CL_DEVICE_WARP_SIZE_NV:		32
  CL_DEVICE_GPU_OVERLAP_NV:		CL_TRUE
  CL_DEVICE_KERNEL_EXEC_TIMEOUT_NV:	CL_TRUE
  CL_DEVICE_INTEGRATED_MEMORY_NV:	CL_FALSE
  CL_DEVICE_PREFERRED_VECTOR_WIDTH_<t>	CHAR 1, SHORT 1, INT 1, LONG 1, FLOAT 1, DOUBLE 1

oclDeviceQuery, Platform Name = NVIDIA CUDA, Platform Version = OpenCL 1.0 CUDA 3.2.1, SDK Revision = 7027912, NumDevs = 1, Device = GeForce GTX 460M

System Info: 

 Local Time/Date =  13:30:06, 12/11/2010
 CPU Name: Intel(R) Core(TM) i7 CPU Q 740 @ 1.73GHz 
 # of CPU processors: 8
 Linux version 2.6.35-23-generic (buildd@allspice) (gcc version 4.4.5 (Ubuntu/Linaro 4.4.4-14ubuntu5) ) #41-Ubuntu SMP Wed Nov 24 11:55:36 UTC 2010

Code:

./oclBandwidthTest Starting...

Running on...

GeForce GTX 460M

Quick Mode

Host to Device Bandwidth, 1 Device(s), Paged memory, direct access
   Transfer Size (Bytes)	Bandwidth(MB/s)
   33554432			4710.8

Device to Host Bandwidth, 1 Device(s), Paged memory, direct access
   Transfer Size (Bytes)	Bandwidth(MB/s)
   33554432			4400.1

Device to Device Bandwidth, 1 Device(s)
   Transfer Size (Bytes)	Bandwidth(MB/s)
   33554432			29643.8

[ker2x]

My 8800GTX died this weekend. i bought a 470 an hour ago (before going @work), i'll test this evening (unless i have to  buy a new PSU too  )