Hi - I want to buy a laptop to do scientific numerical simulations (N-body codes, kinetic growth, molecular dynamics, quantum monte carlo) in C or C++. I am trying to find a laptop that is exceptionally fast, that will let me do powerful simulations with tons of floating point calculations, that isn't too bogged down with stuff that would be mostly useless for me - "gaming intensive" stuff, I suppose. I don't plan on ever playing any games on my laptop, so... I think there's a lot of stuff I won't need.
I have been booking at Sager laptops and I'm not sure which one will suit my needs. I have $1800 but I don't want to spend all of it if I don't need to, of course. I mainly want to make sure that I get a laptop with an NVIDIA GPU that will let me use CUDA so I can do some intense computing.
So I am looking at a bunch of Sagers, IE the NP7682, NP7350, NP2096, NP8662, NP5793 etc. My question is why exactly should I get one over the other, if they can all be upgraded to a T9800 core? I can sort of rationalize the NP7350 over the NP7682 because the 7350 is smaller...
Then there's things that I just absolutely do NOT need like the NP2096's "fingerprint reader." Which Sager will give me the most bang for my specific needs - a CUDA enabled GPU, no special gaming or graphics bells or whistles, (I'm not going to play Crysis!), just raw speed that will let me take advantage of CUDA?
Sorry if my questions are haphazard, I'm writing this during a ten minute break! Give me some Clevo/Sager guidance here, I don't want to spend $1800 when I could have gotten all I want with $1400 or $1200 or less! I'm probably going to ask a bunch of really annoying questions but I'd rather be safe than sorry.
Right now I'm leaning toward the NP7682 with T9800 and 4GB SRAM, but wondering why I should perhaps spend more for the NP2096 or NP8662? Are there are reasons that will matter for me, where my primary interests are processing speed and taking advantage of CUDA on the GPU?
-
You obviously need a high end CPU, Quad core would be ideal for simulations but they are far exceed your budget. 7682 with T9800 seems like a good choice but the GPU is integrated, which means no CUDA for you. What about 2096 with T9800 and 9600M GT?
-
Hmm, okay. I'm not too worried about Quad core vs. Dual core, I am sort of under the impression that getting a faster GHz dual core is better than a slower quad core...
Well, I would like to use CUDA if at all possible. So if 7682 won't let me use CUDA I won't get the 7682. Will any of the Sager laptops under $1000 to start with let me use CUDA? ?? -
1) High end CPU, no GPU
2) Dedicated GPU, average CPU. -
CUDA technology is used by Nvidia cards. Get an medium level Nvidia card like the 9600m GT.
I recommend you got a fast core 2 duo processor, the time to wait for the quad core ability is not worth your time and may not be necessary to go over budget.
The Sager 2096 would honestly be your best bet if you want to go with a Sager.
That or you could find a M860TU that has the 9600m GT. -
OK, so let me try to get this annoyingly clear - the 2096 will let me use CUDA? I look at CUDA's enabled product list (at http://www.nvidia.com/object/cuda_learn_products.html) and the GeForce 9600M GT (which Sager's site says the 2096 has) is listed as a CUDA enabled product. Also on the CUDA enabled product list is the GeForce G105M GPU, which the NP7682 has. So I'm assuming that I'm missing something, if the 2096 is CUDA enabled and the 7682 is not CUDA enabled (considering NVIDIA says the GPUs of both are CUDA enabled).
Dedicated GPU, average CPU. But what is an "average" CPU? If I get an NP2096 with T9800, will I have a dedicated GPU? Will I have an average CPU? Just based on price range, I could go for Sager 5797 with Core 2 extreme 3.06 GHz and upwards of a $3500 price tag. Is this more in line with an "above average CPU" and a dedicated GPU? (I think I'd be fine with a dedicated GPU and an average CPU, but again, what is average?)
I am not sure if I will go for Sager. They just have a friendly site to look through products and customize. I am also considering Clevo and the M860TU or M865TU. For Clevos, can I "get" a 2.9+ GHz processor as I can from Sager? ("Get" in the sense of easily get, for someone with very little computer hardware savvy.) Would a Clevo be a better option than a Sager?
I live in Medford, OR, and I'm not sure if I'll be able to find any of these things locally very easily although my friend said there may be a Clevo dealer in Medford, but their selection is poor. Right now I am just searching for the optimal balance between price and processing. -
1) 9600M GT - dedicated GPU; G105M - integrated GPU
2) G105M - low end level; 9600M GT - average level; 9800M, GT260M/GTX280M - high end level
3) Sager buy laptops from Clevo; you can't buy from Clevo, unless you are ordering lot of 1000 or so.
As said your best bet would be 2096 with T9800 and 9600M GT. End of story. -
OK. I think I understand a bit better now. I am looking at the 2096 closely. Now I begin the hunt for the best price and service.
-
But just so you know I think sager reputation come from their clevo models and the NP2096 is a compal which I don't think has the same build quality as clevo although I have never seen a compal so you'd have to ask a compal owner
-
Yes, that is another worrying thing. Can anyone speak to this? Does anyone have experience with a 2096? How good is it? The T9800 and 9600M GT seems like a good combination for a good price, but I understand that the 2096 is not based on a Clevo.
Just going off of Sager's website, I see that for a few hundred $ more I can get a NP5793 which is Clevo, but this is with a T9300 instead of a T9800, plus it's 17" (I'd prefer 15.4" or less). I'm not sure if this is an upgrade or not?
Should I look beyond Sager? Could I get a Clevo-based model, with a T9800 or comparable speed processor, dedicated GPU etc., not from Sager, for comparable prices? -
-
-
OK, so I'd like to use CUDA but I'd also like a fast processor. I suppose if I fully exploit CUDA then the processor won't matter so much, but I'd like to be able to write non-CUDA programs and have them run as fast as possibly as well... Definitely balancing a whole bunch of needs, wants, impulses, confusions, etc...
Upping the processor definitely adds $. I suppose that from where I'm coming from - hoping to use CUDA - that an extraordinary GPU is what matters most.
Or should I shoot for a good processor too?
Any word on the 2096 from anyone? -
I am also considering a similarly priced laptop I found on xoticpc, the Force 3551 which is built on "MSI-1651". It has a nVidia GeForce 9600M GT 512MB PCI-Express DDR3 DX10 which I'm assuming is better than the NP2096's 512MB PCI-Express nVidia GeForce 9600M GT DDR2 DX10 (DDR3 > DDR2). Will this have an effect on CUDA power?
Very impressive simulations have been written on a GeForce 8800 GTX (16384 particle N-body simulation) using CUDA code. I'm assuming both the 9600M GT DDR3 and the 9600M GT DDR2 are both better than the 8800 GTX, or am I mistaken?
But which will be distinctly better for CUDA, DDR2 or DDR3?
Also, I could just fork out ~$2000 for the NP8662 with T9800. Obviously this would be the "best" laptop in this area... For what I'm going for - a laptop that will last me, that will be powerful and let me use CUDA effectively - would it be worth it to just pay the extra $400 or so?
Thanks. -
Just for you information, 8800M GTX is better than both 9600M GT. It is the same card as 9800M GT (8800 GTX=9800GT).
The NP8662 would have the best graphics card (GTX 260M or FX 2700M), and would perform better. Depends if you need that much power. Seems like, for your uses, the higher the better, so I guess the NP8662 would be a good choice, if your willing to pay the extra $$. -
Maybe Asus's G50 VT? they have 9800m GS, as well as 2.66Ghz dual core CPU.
-
Here's what I'm comparing so far:
NP8662, T9800 (2.93 Ghz), GTX260M, 4GB SDRAM: $1934 (Sager) or $1704 w/ T9550 (2.66 Ghz) (Sager)
NP2096, T9800 (2.93 Ghz), 9600M GT, 4GB SDRAM: $1359 (Sager) or $1164 w/ T9550 (2.66 Ghz) (Sager)
Asus G50 VT, T9550 (2.66 Ghz), 9800M GS, 4GB SDRAM: $1589.83 (XoticPC)
Force 3551, T9550 (2.66 Ghz), 9600M GT, 4GB SDRAM: $1400.68 (XoticPC)
I'm doing CUDA, but I also want a fast processor for non-CUDA optimized programs. Which of these should I look more closely at? Which simply does not compare? All are with Windows Vista. I'm sure the prices could be fuzzed a bit, plus or minus $100 or more. Who knows. The NP2096 definitely looks better than the Force 3551: same GPU, same SDRAM, but cheaper price for a 2.93 Ghz processor?
Comments? Thank you for putting up with me. I'm being annoying. This is how I get before spending $1000-2000. =) -
Signifier, are you going to use the 4 pin firewire port at all? If you are, then that cancels the NP2096, as it lacks such a port.
From that list, if you can afford it, I'd say go with the NP8662 for $1934. It'll be worth the money and your work and productivity will pay it off.
If its too high, go with the Asus G50. -
My recommandation would be to go to 8662 with a lower end dual core cpu.
1. you can upgrade the cpu much easier by yourself later on if needed, but with the gpu you will be stucked.
2. CUDA research might require more VRAM and a faster gpu (9600M GT is 32 cores, 9800M GTS is 64 and gtx 260m is 112) -
The Np8662 also has quads in option. You might wanna see if your uses will take advantage of multi-threading. If so, Q9000 is same price as P9600
-
I am not planning on using the firewire port. The NP8662 has the best GPU by far. Looking at prices:
NP8662, P8600 (2.40 Ghz), GTX260M, 4GB SDRAM: $1574 (Sager) or $1512 (XoticPC).
Would it truly be cheaper to buy the NP8662 with the P8600 and upgrade the processor / RAM later? Quite a bit cheaper? Or would it be wiser for someone with little computer hardware experience to just buy the whole thing as a package deal - T9800, etc. for $1861.43 (XoticPC)?
This is a difference of $350 as it stands. -
Keep in mind that even if T9800 is the fastest , the other processors might still be only 10-15% slower. Does a 10-15 % increase in perfromance justify a double price for the processor?' Besides I was able to overclock a T9300 from 2.4 up to 3 Ghz with a simple program (setfsb).
Somebody suggested a mobile quad core, that might tbe better if you plan to do numerical calculations. The quad core has a lower speed, but you will be able to run more threads and I think it could still be overclocked (not sure).
Ram you can add later, just be sure to buy it in only one piece (such as 1x2GB and not 2x1 GB). If you buy now 1x2GB ram you could add later at resonable price 1x4GB in total having about 6 GB of ram on your system. If you plan to overclock buy a 800 MHz instead of 667, it will be useful. If no overclock then nopoint for faster ram becasue 667 is the maximum for that chip.
PL -
I know nothing about overclocking (is it easy? How is it done? What is it?) and I'm not sure how I'd take advantage of a quad core. I guess for now I'll just go with the included processor in the 8662 and think about updating it later - perhaps, if it turns out that I'll be able to take advantage of the quad core, with a quad core processor. (I'm not sure how to write programs for a quad core, or if I even need to think like that... I'm just starting with adjusting to CUDA).
Thank you pasoleatis for the tip about the 800 mhz RAM.
It seems I can get a pretty good 8662 laptop for $1600 or so. Since I'm going now for most powerful graphics card over all else, it looks like I'm going for the 8662 unless someone can recommend a different setup at this point... or add any more info! Thanks! -
Cuda isn't all that it claims to be go for the best quad available.
-
NGH, care to comment more. I know nothing about quad processors or how easy it would be to take advantage of them.
A NVIDIA GeForce 8800 GTX GPU with CUDA was able to do a 16,384 N-body simulation with all force pairs consider (268435456 forces calculated each loop) and calculated more than 10 billion interactions per second. Not on a quad core processor. This is an impressive showing of CUDA's power.
But if I could easily take advantage of a quad core, just writing simple C programs to do floating point calculations, and achieve even further increase with this... from what I'm looking at, I have either a P8600 dual core (2.40 Ghz, $0 extra) or a Q9000 (2.0 Ghz, $170 extra) quad core as my options. -
PL -
Kill CUDA as part of a development sphere. I want to use CUDA on my own to do intense calculations.
That article doesn't convince me of anything, really. However, I would like to hear your thoughts on why to spend $$ on a Quad. Most games don't take advantage of Quads, but I am not gaming. Is it relatively easy to take advantage of a quad and get tremendous results? Or should I stick with the much chepaer, 2.4+ Ghz dual core processors?
Anyone feel free to comment.
Thanks! -
Anyway, dual core or quad core it will never be at the level of a server with thousands of processors or running CUDA. If you are going to use it a lot for computations you might want to have a quad, but it is up to you to evaluate how much will you need it or not. Quads generate more heat also. The mobile processors have the ability to downclock quite low when they are not used but still it is almost two times more heat compared to the dual core.
PL -
I think that what I want more than anything is to have a decently fast processor and to be able to take advantage of CUDA. I want the programs I write that may not exploit CUDA to be fast, and the programs I write that fully exploit CUDA to be lightning fast.
XoticPC tells me that the best quad that can go into the 8662 is the Q9000, which is $170 extra and I really don't know if I'll get much out of it. I'm not planning on running that many programs at once, and honestly my C code will look like (when reduced to the very core):
for (int x = 0; x < N; ++x)
{
// floating point calculations
}
where N is a gigantic number. If a quad core will let me do this simple stuff much faster, then I am interested.
Otherwise... heating is an issue, price, and the central question of will I use it? I guess I really don't know enough right now to say if I will use it. I am openly listening still for more experiences with quads though! -
The quad core will let you run more stuff in the same time. Like 2 copies of your C program + some tools to ananlize the data. In theory you can also parallelize the code (using c + openmp). If you rcode is parallalizable then you are going to see a big decrease in the time your program finishes (orginial time divided by number of cores you use). Keep in mind that not all problems can be divided among processors. You can also make it parallel with dual core. A safe choice now for you would be to take the dual core and overclock it.
The program used is called setfsb and works from windows. You can use it without retstarting the computer so you can overclock when you need it and then restore the original speed in order to minimize the heat realesed. If overclock you have to ask for a faster memory (800 MHz). I was able to overclock a T9300 from 2.4 GHz to 3.0 GHz and run some tests, and it was stable.
Here is a long thread about overclock.
http://forum.notebookreview.com/showthread.php?t=249470
Ask here if you possible future laptop would support the overclock.
I really can not tell you which one ot buy, because it is different for everyone. If you are unsure what to buy there is a high probability that you would go along with a dual core. If you are tudent you can try to ask you colleagues how do they manage with there computers.
By the way how do you manage with your present computer?
Pl -
Thanks for all your help pasoleatis. I will read about overclocking and I think that for now the best thing would be to get a dual core. My "colleagues" all use Toshiba, Dell, Sony, etc. I walk into a library and see thirty Dells... Even the one student I know who is sort of doing high intensity work (running Gaussian 03 quantum mechanics calculations) uses a Dell XPS which he got for $1800+.
I manage with my present computer, a 14.1" Toshiba w/ 1.4 Ghz that I bought a long time ago, well enough. But my simulations take lots and lots of time. And a massive N-body simulation or a PDE solution on a fine grid is prohibitively long. My computer is good for the internet, writing documents, etc. It falters as a scientific machine.
I think the NP8662 w/ 2.40 or more Ghz and CUDA will let me do a lot, lot more. -
PL
Comparing laptops under $1000 with CUDA enabled GPU
Discussion in 'Sager and Clevo' started by Signifier, Apr 20, 2009.