Posted by Jinx on Wed Sep 21st at 3:28am 2005
P4 3.47GHz
250GB Maxtor IDE Hard Drive 8MB Cache 7200rpm
GeForce 6800GT 256mb graphics card
1GB Dual-Channel PC3200 (ie 2 matched 512mb sticks)
+
512mb PC3200 regular
=
1.5GB single-channel PC3200
With the extra 512 stick iin, the other pair does not run dual-channel.. but I have a good bit more ram.
So here's the deal. I'm thinking of doing an upgrade, and I'm considering the following options:
1) 1GB stick of PC3200 - will give me 2.5GB of ram, but only single channel
2) 1GB Dual Channel - will give me 2GB of dual-channel ram, but I waste the old single-channel 512 stick
3) SATA Hard Drive - either the Raptor or maybe a 160gb WD SATA. I would keep the 250GB IDE drive for storing data, and use the SATA drive for OS & Programs
All of these options run me roughly about $125.
What I'm wondering is this: what do you think would be the best upgrade, to keep the system running smoothly during mapping and compiling? A big question for me is whether any of you have noticed Dual-Channel being significantly better for compiling or mapping. In theory it should be, but in practice I don't see much difference myself. Certainly the gaming benchmarks show little to no improvement from dual-channel.
So.. which do you think will give the best, noticable performance boost, especially as my map gets bigger and more detailed?
Posted by Nickelplate on Wed Sep 21st at 3:40am 2005
Computer's have been using ram without dual channel for years. I don't see it making much difference at work, I'd go with the 2.5G
like they say in the engine trade, "there's no replacement for displacement", I say "there's no replacement for LOTS of RAM."
Nickelplate
member
2770 posts
327 snarkmarks
Registered: Nov 23rd 2004
Location: US

Occupation: Prince of Pleasure
http://www.dimebowl.com
Posted by Jinx on Wed Sep 21st at 3:53am 2005
Nickel: That's what I'm leaning toward, even though I accidentally voted myself for the dual-channel.
The other thing I'm thinking: If I want EVEN MORE ram in the future, the 1GB stick is a better investment.
Posted by Crono on Wed Sep 21st at 3:54am 2005
I think, unless you're a graphic artist and you KNOW you use an intensive application that's proven to run better with duel channel support, it will always be a better choice to get the "more available ram" option.
As far as I know, Windows doesn't use physical ram as often as it should anyway. (Ever gotten the, "Out of Virtual Memory, resizing Pagefile" popup? I have, several times, in which case I always check my Ram consumption, it's usually around half)
So ... in either case, even IF the ram does go faster ... there's no way to really control which set of ram it gets written to for the duration of the programs life. Which is why I suggest getting a fast ass harddrive. A lot of program will run much faster if the OS can access the page file(s) faster.
As for how much ram it uses and when ... I really don't know why it doesn't use a larger percentage, or maybe it's a bug?
Also, you're a bastard.
Oh yes, also, as for HDD speed, I'd be more concerned with the Cache size. Taking that, if the drive goes 7200RPM still ... it wont really matter if it's over SATA or ATA133/160/whatever connections. However, SATA is used to handle faster drives, in which case, you could eventually look at getting a 15,000 RPM drive, or something as such. But, in this case, the physical moving parts in the drive are the slowing down factor. (16M Cache would be the minimum. It's decently priced right now and blah blah blah. But ram would do a world of good for run time intensive applications ... since, no matter what, you'd be doubling it, or close to it)
Posted by Jinx on Wed Sep 21st at 4:33am 2005
I did rough benchmarks... 1GB dual channel vs 1.5GB single channel. Basically identical times for compiling. Same times if I set my pagefile to 2mb, too. And tbh nothing is using more than a few 100mb of ram at this point. So maybe it's a waste of $$, I'm not sure. I'm thinking of what it will be like when the map gets more complex, though, too I guess.
The Raptor is a 10,000 RPM drive, and its seek times are almost half that of most other drives. Only an 8MB cache, but still.. very nice.
Posted by Windows 98 on Wed Sep 21st at 4:59am 2005

Windows 98
member
757 posts
86 snarkmarks
Registered: Apr 25th 2005
Location: USA

Occupation: Student
Posted by Jinx on Wed Sep 21st at 5:07am 2005
I suspect what is going on is this, Win98:
512mb - 448 = 64mb
Probably you have onboard video, and it is 'borrowing' 64mb of system ram. This is common... add a separate video card and that should get you the 64mb back. That's my guess, anyway... assuming you have onboard video.
Posted by Crono on Wed Sep 21st at 5:10am 2005
But, You could have: 256 MB + 128 MB + 64 MB (Probably internal) = 448 MB. [update] Yes, or what Jinx is saying is possible too. Which makes sense as well, just because HP is actually pretty notorious for having onboard video.[/update]
Jinx, OH! I didn't know it was 10,000 RPM. In that case, I know more ram would make a difference in a few months (with coming applications/games and such requiring a minimum of a Gig, Lost Coast is an example.) I'd say, go for the ram, and save up for the HDD. Since, transferring stuff over drives is a pain in the ass anyway.
Yeah, Windows is weird, I don't know why the option is there to change the sizes if it doesn't really listen. I guess it thinks it knows better, like, "Hmm ... I know they didn't mean to set it to that amount". I don't know, for all I know, maybe the larger you make the pagefile/VMM allocation the more it would use it, since it's a larger reservoir. I don't really think a lot of Windows type stuff is really concerned with how fast it runs at times. (If it were it wouldn't drop the page file ... ever ... which it does quite often ... so annoying)
Posted by Madedog on Wed Sep 21st at 5:39am 2005
Then you would want to take a SATA (windows has issues on running on sata, I know from my own experience, trust me) for the DATA, and keep the IDE for OS.
Just my2cents..
Madedog
member
487 posts
119 snarkmarks
Registered: Jan 5th 2005
Location: Estonia

Occupation: Level Designer
217.159.236.34:27050 - CSS Server - Clean | koffer.ee
Posted by Jinx on Wed Sep 21st at 5:52am 2005
I am on an 865 board that is AGP and DDR1, so that limits me somewhat. I'm not upgrading the mobo; I'll just build a new system at that point.
And believe me, a 6800GT is no slouch
Posted by Gaara on Wed Sep 21st at 7:40am 2005
SLI really isn't worth it in performance boost.
My computer is ok, although I have dial up
. I'd downgrade anytime for cable. I'm too far out to get cable or even adsl and satellite isn't worth it.
Gaara
member
219 posts
22 snarkmarks
Registered: Aug 12th 2005
Location: Australia

Occupation: Freelance Gynacologist
Posted by MisterBister on Wed Sep 21st at 2:54pm 2005
If you are mainly focusing on mapping and compiling on the computer and p4 would be the best choise, why?
Because compiletimes seem to only care about one thing: MHZ!
The more, the better.
The compiletimes scale almost exact with the mhz.
I ran a test on a friends 3ghz p4 and it was almost twice as fast as my athlon xp 1,8ghz.
Fiddeling with things like more RAM, higher FSB or memory timings doesnt actually affect the compiletimes, trust me ive done some testing on this one.
But if you are going to play lots of games the Athlon 64 is the way to go.
Those processors are cheaper, less warm, much more energy efficient and faster (and more fun to overclock >=) )
However one thing that requires RAM is Hammer v4. I remember running the program with 512mb RAM and it wasnt funny to scroll down in the texturebrowser for a long while.
It seemes like the textures gets loaded right into the RAM when you are accessing the browser, and if you have been mapping for a while, hammer can take about 500mb of memory.
I noticed a HUGE difference when moving from 512 to 1gigg when using the texturebrowser, but almost no difference in other things, such as gaming performance.
Im currently running with 1 gigg of RAM and lots of people do, getting more under the hood is, in my opinion, useless. On an athlon 64, you can actually notice worse performance when getting more than 1 gigg, however im not sure of this one.
Maybe because it takes more time to find the data when the storage is bigger.
About the graphics card:
The 6800GT is a great choice, however I chose the x800xl. The two cards are almost exactly as good when it comes to performance.
Although, 6800gt is better in doom and x800xl is better in Hl2, but the difference is very small.
If i where you, i would compare the price between the cards before going for 6800gt. But thats your call.
You asked a question regarding getting more RAM or getting a fast harddrive.
If i where you, i would get a fast harddrive, no doubt.
And if i where to chose, a Raptor.
250giggs is enough for storage and 74 giggs is more than enough for installed prorams.
2 giggs of ram is from my point of view completely useless. And if the p4's are doing as the athlon 64's when it comes to get more than 1 gigg of RAM. Then i would definitly only get 1 gigg.
MisterBister
member
277 posts
78 snarkmarks
Registered: Oct 17th 2004
Location: SWEDEN

Occupation: studying
Posted by Nickelplate on Wed Sep 21st at 3:04pm 2005
You've probably got an onboard Video card that is using a 64 MB of your 512 MB. It's called "shared" memory and it sucks.
Nickelplate
member
2770 posts
327 snarkmarks
Registered: Nov 23rd 2004
Location: US

Occupation: Prince of Pleasure
http://www.dimebowl.com
Posted by Jinx on Wed Sep 21st at 4:24pm 2005
I'm still mixed. Back in the day, when I was compiling on my Pentium II, upgrading from 192 to 384mb of ram made a HUGE difference on the bigger maps. At the moment, I can't really tell any difference between 1GB and 1.5GB... I'm just thinking that down the road the ram might make a difference as I'm mapping more. Then again, maybe I should just wait and get the ram when I think I need it? The prices on DDR are only coming down.
The Raptor would certainly speed up load times on anything I put on it, and I've been assured by everyone that I -would- notice a difference if I had one. I can only afford a 36GB one, though, and that might be cutting it a little close in size.
/me frets
Posted by Crono on Wed Sep 21st at 7:25pm 2005
You also have to remember, as I'm sure you're aware, that the program currently "taking your attention" is the one that gets precedence in physical ram (or is suppose to, past the OS anyway) ... of course there are settings to change this, such as if a program comes in with a certain level of authority and runs in the background (AUPDATE, for example. It'll use all your resources, no matter what you're doing, until it's done)
As for "noticing" a difference because of VMM speed. Well ... yeah ... but, you'd also notice a speed up if you had one small drive (around 40GB) which had the OS and some programs and another drive with all your other programs and data.
It's weird, because, it also depends on if the program actually loads everything it needs into ram. Most do, but some don't.
Bister, while, a single cored system will probably compile at the same speed as a duel core system (Since, the very process of compiling is a very linear process, and if something isn't really made to be pipelined, it usually doesn't run faster in that situation) ... but, all compiling anything is is one calculation for number crunching after another. While, the processor is the place this happens, where do you think it all gets stored? Every time the compiler finishes with an important calculation, I'm pretty sure it'll do a memory write, since all that information can't be saved in the L1 or L2 cache, it's too big. That means, that a FSB does matter. IT always matters. It's just that, to a person's perception it is negligible. After all, a HDD access is "negligible", and it takes tens of thousands of cycles.
Now, of course, as we all know, the CPU waits when there is a memory access or memory write. It waits for confirmation from the Chipset. So, you can't say that any of the Bus speeds on the system Bus "don't have an effect". Because they do. In fact, the system bus is WHY systems are slow.
I wish some company would suck up the billions of dollars it'd take and build a system with little to no Bus. Physical memory, D-Ram I might add, would be like the L3 cache. But, for that it'd physically have to be built on. Combine the chipset and processor into a triple core chip or something like that. Just get rid of physical part HDD and use Flash Memory (L4 cache) ... hmm ... sounds to me like things wouldn't need to be loaded as often. Hell, you could have a separate core of the processor made to deal with system utilities (other cards) and a separate set of ram for the addresses. That'd be sweet.
Posted by wil5on on Wed Sep 21st at 11:49pm 2005
wil5on
member
1733 posts
323 snarkmarks
Registered: Dec 12th 2003
Location: Adelaide

Occupation: Mapper
- My yr11 Economics teacher
Posted by Jinx on Thu Sep 22nd at 2:01am 2005
bwahahahahaha
doubt it will effect much now, but it's going to be nice as I map more I suspect.
[EDIT] No change in compile times, though my map is still fairly simple. Hammer, however, does run smoother. Before, there was a slight lag when I would pan around in the textured camera mode... now it's smooth as silk.
Also, the ram has blue heater spreaders and looks pretty hehehe
Posted by Underdog on Thu Sep 22nd at 3:43am 2005
One must ask, "where is the plateau?" At what point does hardware upgrades make no noticeable difference?
I do not know so this is a serious inquiry.
Lets assume that you could create a functional,playable map so large that you could actually measure a decrease in compiling times with your current processor. RAM increases storage capability, but if you have more than your processor can deal with would you really compile faster?
The point I am asking is similar to the old video card problem where you had more card than your processor could deal with and your frame rates didn't increase when you upgraded.
I think the best question would be, "Does anyone know exactly how much information is created during a maps compile?" Would any map fill 2.5 gigs?
Before I vote, I would like these questions addressed.
Thanks for your time.
Underdog
member
1018 posts
102 snarkmarks
Registered: Dec 12th 2004
Location: United States

Occupation: Sales-Construction
Posted by Crono on Thu Sep 22nd at 4:55am 2005
For example, VMM allows at least 4 GB of ram to always be allocated. It'd be nice to have that in physical ram, but is doubtful.
Well, as for map sizes ... how big is the BSP? a few megs, right? Maybe 20 something if you include all the materials. But those get put into video card memory anyway.
The memory requirements really come from programs specifications, more or less, HL2 is what requires the memory.
To be honest, it's hard to answer your question, because it's not really applicable. It's like saying, "How do we go about a 100% effecient vehicle?"
I could elaborate more when I'm not distracted.
Wilson, I know, but companies blow money away over less things more often. However, it'd be just as upgradable as our current machines. You'd still use certain amounts of RAM, different speed processors, HDDs, everything, my point was, they need to eliminate the bus. But by "built on" I meant, on the outer bus, but close enough so it doesn't take too long. Like the same amount of time it'd take to go inbetween processors (if you have an MP system). Just removing the system bus. But there'd have to be a system BUS still, to talk to expansion cards.
Posted by wil5on on Thu Sep 22nd at 5:09am 2005
wil5on
member
1733 posts
323 snarkmarks
Registered: Dec 12th 2003
Location: Adelaide

Occupation: Mapper
- My yr11 Economics teacher
Snarkpit v6.1.0 created this page in 0.0135 seconds.

