rower30

Distinguished
Dec 16, 2002
264
0
18,790
Here is an idea but I don't have real data to support this at all. So here is a homework assignment.

I posted this in the graphics cards section, and thought you hard core overclockers may have some input.

The 9700pro has the same GPU core / memory clock as the new 9800 (non-pro). 380MHz / 340MHz (680 DDR)

The 9800Pro has an increased GPU core / memory clock. 325MHz / 310MHz (620DDR)

True to the nature of the 9800pro, and the old 9700pro, they didn’t really overclock too well, but were usually pretty tight to spec. Why? Because in each instance, the core / clock is (9800pro) and was (9700pro) set to within an inch of their life to be the king of cards. And that works just fine, too.

That leads me to the R9800 (non-pro). My assumption is to save a few bucks, and to help all the slightly poor gamers, the card is still about $270.00, ATI Bins down the same R350 core and memory (anyone know that it is identical, or is it really different?) to create the R9800. Question, will we have a better price / performance card here after probably better overclocking of the core, GPU or both? The savings is $60.00 or so right now, about what DOOM III or Half-Life 2 will cost you!

9800 verses 9800pro could be an excellent experiment because there is very little downside. The 9700Pro are no FPS slouch, and with update features on the 9800(non-pro) as the 9800pro, it should actually be slightly faster than the 9700pro clock for clock. And, the 9800(non-pro) may overclock even better. The 9800(non-pro) GPU core is most probably a good overclocking candidate because I doubt that the same R350 GPU is ever WAY off of the original 380MHz bin. EXAMPLE; a “bad” 9800pro 360MHz GPU bin marked down to 325MHz with a 9800 (non-pro) BIOS flash. It will move the hardware, and save ATI a few bucks. And, you can get it back to 355MHz with an overclock utility! This is an example, your results will of course vary. But the risk is well worth the effort. Those unused GPU are really eating into the bottom line profit dollars. The memory may actually be a slower part but that’s OK because I’ve found that the GPU clock speed FPS efficiency is more effective than the memory clock speed. But still, even the slower memory is still waiting for a tweak. Help me out here people, anyone know the memory part number for the non-pro 9800 part? Are they slower than the 9800pro part?

With the help of; 2nd Generation cinematic architecture, SMARTSHADER 2.1 ™, SMOOTHVISION 2.1™, and HYPERZ III+™ the 9800 (non-pro), is more efficient, and faster, than the older 9700pro. So, is the table below out of the question? The data, if we all work to fill it in, seems to say that the 9800(non-pro) is a nice card. Even with stock GPU/memory settings on the 9800 (non-pro) we see a 3Dmark/dollar ratio increase of 18.93%. And, we get a better looking 3D world over the 9700pro, which is still a good value at about the same 3Dmark/dollar ratio. This is what I call a good upgrade.

So, do we have any takers on this task? All you gamers out there have a few cards to round up? We’ll need to “normalize” the reference systems to the same computer with the two, or even three, different cards.

3D Mark 2001 SE (b330)
XP 2700+, ASUS A7NX
1024 x 678
..................................................3D Marks.....Deficit..........Price............3Dmark / $
Radeon 9800Pro.............................16225.........Reference.....$334.00(1)....48.6
Radeon 9800 overclocked 5% gain*...15977(?).....neg 1.53%.....$271.00(1)....58.9
Radeon 9800**.............................15672(?).....neg 3.41%......$271.00(1)....57.8
Radeon 9700Pro............................15216.........neg 6.23%......$259.00(1)....58.7

* Figure 5% over the 9700pro, just a guess.
** Figure 3% faster than the 9700pro, just a guess.
(1) Price Watch as of July 14, 2003.

Regards,
Rower30@earthlink.net
 
G

Guest

Guest
someone please answer this guy. that must've took 30mins to type, lol.

<font color=red>Crack Addict=Baaaaad </font color=red>
<font color=green>THG Addict=Goooooood </font color=green>
<A HREF="http://www.tylersite.8k.com/photo.html " target="_new">http://www.tylersite.8k.com/photo.html </A>
 

10GHZ

Distinguished
Jan 21, 2002
963
0
18,980
good analysis, but i dont many plp out there would be bothered to do the same thing, unless they want to post a bench or some sort on their tech site. good effort thou
 

scottchen

Splendid
Jun 3, 2003
5,791
0
25,780
I would believe that ATI probably used a lower voltage on the 9800 non-pro as they did with 9700 non-pro compare to the pro versions. I can't wait to get my Sapphire 9800pro next weekend.

I hate Computers! I could only overclock my Barton 2500+ to 2645mhz. My Radeon 9700pro core/mem speed won't go above 410/720. I really hate computers!
 
Well, since it's actualy turned into a discussion, here's my 2 frames worth;

As long as the R9700non-pro is out there, there is no better card/bang for your buck. IF this comparison ever did reach fruition like a few other bang/buck tests out there (see ExtremeTech I've included)); I'm sure you'd see the R9700NP being above the cards listed above. So as long as it is on the shelves, it's a winner. It has good overhead, and it's relatiely cheap for the 'poor' gamer.
The thing would be to insure that more than just one test is used.
EXTREMETECH has a <A HREF="http://www.extremetech.com/print_article/0,3998,a=41363,00.asp" target="_new">price/performance review</A> (which I'e posted before) which uses their 'game-guage' whihc is made up of a bundle of benchmarks. It's still a little vague as a guage, but it does include a wider variety of tests. And of course they've include 3dmk03 also.

Now their test shows the R9600Pro to be the winner, but I would say that's because they didn't include the R9700non-pro. The only other issue is that YES unfortunately it didn't include overclocking but I don't think the R9800's will knock my R9600P off the list if that's included, if anything it would elevate my card even further. I think the theory is somewhat flawed as there is to much speculation that the R9800 may have more overhead than it does in reality (at least from the reviews I've seen).

The main thing is that the OLDER cads usually make a better value due to their 'not so fresh' feel. Of course don't expect it to win you any benchmarks as the newer cards will always win, however for actual gaming the older cards like the R9700NP will be a great deal for 'price-conscious' gamers IMO.

<A HREF="http://www.extremetech.com/print_article/0,3998,a=41363,00.asp" target="_new">http://www.extremetech.com/print_article/0,3998,a=41363,00.asp</A>

BTW, rower, why not test your own theory? That's usually the way it works around here, there is no 'homework' except for what you make for yourself.

- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! <A HREF="http://www.redgreen.com" target="_new"><font color=green>RED</font color=green> <font color=red>GREEN</font color=red></A> GA to SK :evil:
 

lhgpoobaa

Illustrious
Dec 31, 2007
14,462
1
40,780
From what ive seen the cores overclock well enough, the real limiting factor is the ram.

Manufacturers may or maynot skimp on ram quality of the non pro's. If they do u are stuck. If you are lucky it might be the same as the pro.

Personally im not fussed.
I got a R9700pro a little while ago and found that overclocking only adds a few percentage points at best.
stock is 325core/310mem...
I can run mine at up to 336/336 without anything looking wrong and extreemly stable.
I could push the core higher but i like the synchronous operation :)

310 to 336 is a nice boost, but nothing too amazing. (using stock 2.86ns sambsung chips)

<b>Regards,
Mr no integrity coward.</b>
 
Yeah, heck I'm running the 2.86ns memory on my R9600Pro at MAX 363, and at EASY 349 (since that's where CasualCat got his set for standard) without additional cooling.

So yeah, 336 seems low.


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! <A HREF="http://www.redgreen.com" target="_new"><font color=green>RED</font color=green> <font color=red>GREEN</font color=red></A> GA to SK :evil:
 

lhgpoobaa

Illustrious
Dec 31, 2007
14,462
1
40,780
well thats what it does.
339 gives very minor graphical glitches after a few hours
340 gives them after a few minutes. wierd colours and pixellation

<b>Regards,
Mr no integrity coward.</b>
 

scottchen

Splendid
Jun 3, 2003
5,791
0
25,780
I just got my Sapphire 9800pro, modded a little, got to 475 core, and 750(375x2) memory.

I hate Computers! I could only overclock my Barton 2500+ to 2645mhz. My Radeon 9700pro core/mem speed won't go above 410/720. I really hate computers!
 
Actually I was able to get to 371 with my mem. but experienced slight 'blackflies' glitches in 3dmk03 in the Wings tests (nowhere else). However no a singler lock-up. Anomaly free was 361mhz (363 was after 2 previous benchmarks at 361). I'm just surprised because isn't the 2.86ns actually RATED to 350mhz? So anything under that would seem like a bad sample to me.
Anywhoo, I HAVE noticed they are all different it seems (not enouraging).


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! <A HREF="http://www.redgreen.com" target="_new"><font color=green>RED</font color=green> <font color=red>GREEN</font color=red></A> GA to SK :evil:
 

rower30

Distinguished
Dec 16, 2002
264
0
18,790
You guy are supposed to be hobbiests, not a bunch of conversationalistic clouds wandering over the landscape spreading rain. So yes, it is my job to suggest interesting things to avoid making glib comments. Comments are cheap, and typically worthless. That, and the lot of you may be able to pull things off simply based on the dumb luck of having equipment around. Why sit around an listen to Tom or Anand's? I'm not stuck out front in a rocker just yet. You guys can sit around and yack all you want I've got a hobby to run.

TheGreatGrapeApe listen up, be careful what you ask for, you might get it. My testing is complete.

Asus P4PE 845 chipset
P4 2.4GHs @ 2.61GHz
256RAM

PowerColor 9600np 13500 2001SE 3Dmarks +\- 100.($290.00)
Sapphire 9600pro 15500 2001SE 3Dmarks +\- 100. ($374.00)

9600np 46.5 3D marks per buck.
9600pro 41.4 3D marks per buck.

Sapphire verses PowerColor? The Sapphire didn't overclock well at all. 400/350 is all. Zilch change in 3D marks (200-300 3Dmarks at best.)I didn't get a chance to test the PowerColor 9600np before I sent it back. But overclocking is mostly a hollow victoy that is fun to do but largely irrelavent in blind tests. A motion picture study was done showing most don't notice FPS issues until they go below 20FPS. The motion picture industry uses 25FPS on purpose to add the "flash" to the screen people are accustom too. If they go to 30+ FPS, people complained that the picture was worse!

So we see a 14.8% increase in 3Dmarks for over a 29% increase in price on this system. Unless 3D marks are at least 5000 different, you got the same problem in reality with either card what ever they may be. It won't matter if you have a 1.8GHz P4 or a 3GHz P4 or whatever. Too slow will always be too slow and fast enough will alway be fast enough. Will a narrow range of you will notice a 9800np seeming too slow compared to a 9800pro? 15% increase on 15FPS gets you to about 17FPS. That's my two frames worth, and it's a fact.

The 9800np is the best wife/husband, the 9800pro is the best girl/boyfriend friend. What do you want?
 
Well first of all where did you buy those cards (what currency?)? My BBAR9600Pro cost $299CDN on release day (they didn't charge me for shipping despite the site saying they would) and sells in stores for less now. And sapphires go for cheaper. My OC was 533/363 and that was STABLE (533/361 was run for about 2 days of benchmarking). Like I said in a post in the Graphics forum, it appears the SAPPHIRE boards are terrible overclcokers and alot have been found to have the 3.0ns mem, not the 2.86 (even a few had the 3.3ns). So I'd say they are the wrong choice. IMO it's a good thing you did do the Powercolor as the SUCK IMO, almost as much as the Gigacubes.

Now I agree overclocking isn't guaranteed, heck I've said that everywhere, even here, cause results vary depending on model (most BBAs sseem to reach 500/340; but some here have experienced lower results [like `460/340 is the lowest BBA I remember])

NOW as to your FPS theory it's not that simple, likely it's a nice study, but it's not THAT simple. There are ALOT of things that affect what is perceptible (colour, contrast, motion, direction, background, # of objects, distance, focus). The main thing that will change you perception of smooth is the motion within what you are watching (I guess that's the point/focus of it all isn't it?), and the WORST (biggest stres-test) is angular motion on a moving background. Angular motion (similar to jump-straiffing) will require MUCH higher FPS to look seemless on screen. The reason is that the brain has different 'slices' geared to be triggered by different receptor info. There is different parts of the brain responsible for different angle of lines (like on for 15 degree line one for 30 degree line, 45 degree, horizontal, vertical, etc.) And the motion, and motion against a background moving in a different direction also trigger receptors all at once. All this makes for a WIDE range of frame rates that appear seemless to the viewer. Less than 16fps can also look seemless despite being commonly held as the bare minimum for seemless motion, this would work for a game like Splinter Cell, a slow creeper with only occasional quick motion (depends on HOW slow you're moving); however something like UT2K3 may need wel above 60fps to look seemless. Up to about 90fps there can be issues, most of the population (like 97+ percentile) is fold above 90fps, and the higher you go the quicker it narrows down, 'til you reach around 100fps (I think the top IIRC was 97fps where it was much LESS than 1 percent of the population). What you need to realize is that MOVIES are MUCH different than TV (where I worked as a cameraman to pay for school [economics/psychology {hence the physiology info from Sensation and Perception which I was VERY interested in, and aced :cool: }), Movies usually have a larger frame than your field of view, are distant, and are in darkened rooms, where people are focused on the screen. The last one also makes a difference as it maintains a point of focus. Try watching your screen while moving your head, often you will actually see frame changes because of that introduced motion.

Anyways, just trying explain WHY the common held 'theories' about FPS don't apply to every example, it's alot more complex than a 'standard' fixed number.


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! <A HREF="http://www.redgreen.com" target="_new"><font color=green>RED</font color=green> <font color=red>GREEN</font color=red></A> GA to SK :evil:
 
Good point. It would be interesting to find out, as it does seem out of place.

- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! <A HREF="http://www.redgreen.com" target="_new"><font color=green>RED</font color=green> <font color=red>GREEN</font color=red></A> GA to SK :evil:
 

rower30

Distinguished
Dec 16, 2002
264
0
18,790
Cards are shown in US dollars unless otherwise stated which is typically the case. Go to price watch.com

I don't compare overclocked cards simply because it's so unstable and statistically invalid. A none overclocked cards represents the 99.9 percentile you can expect from ANY card of the same brand. No so with overclocked cards. The 9800np is still the best performance for the dollar using this metric.

As far as what cards overclock well, I haven't seen a statistical comparison of ANY brand of card. I expect if you rounded up at least 15 samples of the SAME card to get an 80% probablility number, you would be displeased to see the results congregate around a rather bland overclock number on core or memory. Manufacturers aren't dumb and don't give away easily aquired performance unless you pay for it. The 99.9 percentile performance is already established out of the box.

But since this overclocking stuff is so purely emotional as far as "bands" are concerned, I simply don't care. Show me statistical overclocking data on 15 samples of the same card on the same machine with the same benchmarks across several brands otherwise, go fish. You have no real basis for comment against anybody's product. You're simply hiding behind what is clinically known as, "appeals to the ignorant". EXAMPLE; "I saw an invisible pink elephant in my bedroom" Now, prove I'm wrong.

Problem is, I can take your overclocked card and easily show you are wrong with a 15 lot sample. But, you simply hide behind the fact that no one will test 15 cards. I see a lot of this nonesense. I posted a request to do a simple test and all I heard was, "you do it". I did, and everyone moved to a new metric, statistically invalid overclocking benchmarks. We seem to purposefully want to avoid valid metrics testing two different cards in the same computer WITHOUT overclocking and test bed variations. Sorry, I did and the 9800np is simply a better value.
I also tested power supplies in the same system to prove a point. When you test the manufacturer's product in a "stock" state and good statistical quality assurance was used in its development (typical now days) you can test fewer items to regress to the mean performance of ALL items of that configuration. Change states, and at least 15 sample sets (30 is ideal for 99.8% probability) need to be culled for an 80% probability that the next one you buy will reach an asymtotical mean across the universe (all cards).

I shot a sparrow out of the air with an old BB gun. I guess I had a VERY good gun and / or I was a VERY good shot? Well, lets try it fifteen times and see what I was...lucky!
It sounds like many of you have no background in statistical product eveluation or you wouldn't make such baseless comments. You all look cute to one another, but can look pretty silly to someone who does know the ropes. The sad thing about it is you also don't learn the truth, and isn't that what this is suppoosed to be about?

My point with FPS agrees with your observation(s). Most games are fine 90% of the time until large amount of texture is changing simultaneously. You indirectly point this out in the different games. Full motion video (NOT cartoons / games) is often compresses to STOP background positional refresh. Trees, bushes ETC are frozen in the background . Video games are hardly "full motion". Your perspective changes as YOU move, but NOT the screen. And, you DON'T change your angle of view to the monitor. It's always head on. This, and full motion video, IS NOT real life! There is no 3D space on a flst screen. It just ain't so or ever will be. It's 2D. You can't put a pencil mark anywhere on a movie screen or monitor that isn't described by X and Y coordinents. I know we all like to think we are some how are dealing with the most powerful thing imaginable in our PC's but...we aren't. Video games are pretty lame. Fun, but still hardly anything that is really all that awesome. How can that be? You've already pointed out just how fast your eyes superimpose and integrate motion. No video card will ever come close to closing that capability. As games try to look more realistic, they also look worse. Your mind has a good basis of comparison to reality and it says, " this isn't even close". Keep games closer to good quality cartoons, and we avoid that subconscious comparison.
How can full motion video look seamless at 30FPS while you say 15%, at best, background refreshed cartoons not look good at the same 30FPS? Sorry this does not compute. If you fall below 30FPS, and you do center your attention on a refreshed area of the screen, than yes, you will see video stutter. 100FPS for cartoons? Nonesense.

Cartoons and full motion video are so vastly different that gain, this argument is baseless. Also, comparing 2D to 3D motion is silly, too. Common sense tells you that full motion 100% refresh video will need the best available "refresh" compared to animated cartoons. 30FPS sustained does it. 25FPS does it with the added "flash" for the movie like experience.