PCI Express & CrossFire: Scaling Explored

Page 4 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
What about World in Conflict?
http://www.tweaktown.com/articles/1472/9/intel_p45_vs_x48_crossfire_performance/index.html

OK, after considering both the Crysis and World In Conflict results of that site, I have to conclude that their results are invalid. Perhaps something was wrong with their P45 configuration.
 
[citation][nom]Crashman[/nom]OK, after considering both the Crysis and World In Conflict results of that site, I have to conclude that their results are invalid. Perhaps something was wrong with their P45 configuration.[/citation]
There is nothing quite like a biased conclusion...
 
[citation][nom]Slomo4shO[/nom]There is nothing quite like a biased conclusion...[/citation]

The P45 motherboard they used had a PCIe hub that allows the v2.0 pathways to be split across four slots, so like I said, maybe there was something wrong with their P45 configuration. It also appears that they were using a very old Beta version of ATI Catalyst which was known to be problematic with some HD 4xxx cards in CrossFire, so the thing that was wrong may have been software as well. At any rate, people should question the relevance of any crossfire articles using Catalyst version prior to 8.8
 
[citation][nom]Slomo4shO[/nom]There is nothing quite like a biased conclusion...[/citation]

After further consideration of the context of your comment, I'd like to add that maybe the "problem" is that there was something right with Tom's Hardware's P45 configuration.
 
Hello Guys,

this is my first post on tom's hardware because i just subscribed :)
i have been reading the articles here for some time now and I just wanted to compliment the testers (i believe Thomas?) and the commenters. A lot of useful info!

Anyone has any tips on how to create and customise my appearance here?

greetings,
Me
btw, sorry for any mistakes. English is not my native language (I am Dutch)
 
I was planning on building a new system based around a DFI X38 T2RB because of its overclocking abilities. I was also going to use HD 4850's in Crossfire with it. If I'm going to actually do crossfire would I be better off getting a P45 or X48 board instead? I guess I'm not seeing where the X38 is performing compared to the others. Also I will be running dual 1680x1050 monitors.
 
[citation][nom]Crashman[/nom]1920x1200 is almost right in the middle between 1680x1050 and 2560x1600, so hopefully most people can figure out "about" where that resolution would fall on the charts.Is it time to get rid of 1024x768? I'm in favor of ditching that resolution and picking a different one.[/citation]

I completely agree that the 1024x768 resolution should be dropped. Obviously, very few people with enough hardware interest to be reading Tom's will have anything like that on the desktop. And replace it with 1920x1200 !! This isn't the only review I've read lately where this very popular resolution was ignored. I understand the appeal of including 2560x1600 despite its rarity - but not at the cost of leaving out a much more common gamer setup. I went from 1280x1024 to 1920x1200. Stopping at 1680x1050 didn't make sense, especially since a surprising number of last-gen games don't support it.

Thanks for the review. Love my HD4850!
 
[citation][nom]Cainmos[/nom]I was planning on building a new system based around a DFI X38 T2RB because of its overclocking abilities. I was also going to use HD 4850's in Crossfire with it. If I'm going to actually do crossfire would I be better off getting a P45 or X48 board instead? I guess I'm not seeing where the X38 is performing compared to the others. Also I will be running dual 1680x1050 monitors.[/citation]

X38 and X48 chipsets are functionally identical, so if there's a performance or overclocking problem it's the fault of the motherboard manufacturer, not the chipset.
 
Did I miss something or was bandwidth intensive AF not used for 4/5 of the tests?

X48 is the best for core2 CFing, since X58 uses entirely different CPU and RAM it can't be compared but you can be it would wipe the floor with everything tested in this article. Not really because of the chipset, but because multi-gpu with todays powerful cards has become CPU limited.
 
[citation]I'm in favor of ditching that resolution and picking a different one.[/citation]
I actually use this rez, though I'm sure it's rare. And you might also take into consideration that most 3-400 dollar systems come off the shelf with 17 inch monitors. They're then upgraded by stingy college kids, and played to death for 4 years.
 
[citation][nom]marraco[/nom]Good work!.Altought, I have an Athlon X2 system, and probably gonna update to a I7 920. It would had be better comparing to an cheap i7 as a reference[/citation]
The problem with that is that Core i7 uses DDR3 exclusively, and would skew the results.
 
[citation][nom]Arbie[/nom]I completely agree that the 1024x768 resolution should be dropped. Obviously, very few people with enough hardware interest to be reading Tom's will have anything like that on the desktop. And replace it with 1920x1200 !! This isn't the only review I've read lately where this very popular resolution was ignored. I understand the appeal of including 2560x1600 despite its rarity - but not at the cost of leaving out a much more common gamer setup. I went from 1280x1024 to 1920x1200. Stopping at 1680x1050 didn't make sense, especially since a surprising number of last-gen games don't support it.Thanks for the review. Love my HD4850![/citation]
There are a lot of gamers using 4:3 resolutions, so it might be best to keep it or use 1280x1024. In the Steam hardware survey, most common resolutions are 1024x768 and 1280x1024, with 25% of the share *each*. 1920x1200 is at about 3.5% and the resolution I use, 1680x1050, is at 14%. http://store.steampowered.com/hwsurvey
 
[citation][nom]wicko[/nom]There are a lot of gamers using 4:3 resolutions, so it might be best to keep it or use 1280x1024. In the Steam hardware survey, most common resolutions are 1024x768 and 1280x1024, with 25% of the share *each*. 1920x1200 is at about 3.5% and the resolution I use, 1680x1050, is at 14%. http://store.steampowered.com/hwsurvey[/citation]
Probably all the people playing CSS on their crappy 5-year-old rigs.
 
Well actually I think that Wicko is right. Allmost everyone has 4:3 display at this moment, but the change is going on rapidly. Have you ever wondered why games only use one core? Or why most games at this moment are bacically dx9 based?
The hardware change actually very slowly. I see a lot of computers in a week, and only about 5% or less are widescreen monitors at this moment. (And those who has are mainly my friends that are "computer geeks" like most of us in this forum. I am guite sure that allmost half of us in this forum have WS-monitor, but in "real" world the situation is not the same.
But the movement is stellar. We are moving fast towards WS-monitors. When markets sells their old inventory out, the bigger part of newer monitors will be WS-screens maybe not including some really cheap models. And what kind of models are selling best? The cheap ones... So that survey is not as out of day as it seems to be to our eyes.
It will not take many years untill over 50% of motors in use are WS... Well maybe I am a littel bit optimistic in here, but in anyway so big persentage that in games the support for WS will get even better than today. (WS is still better supported feature in games than multible CPU cores...)

 
It's not that I think 4:3 monitors are uncommon, because that couldn't be further from the truth. But playing at 1024x768? I don't see alot of 15" LCDs any more, and most likely the people playing at 1024x768 don't even know what a resolution is so they just use the default, which is usually 800x600 or 1024x768.
 
Hi, I'm the owner of Asus P5W DH Deluxe motherboard. Now it works with Intel E7200 and it rocks. I found one mistake in this article. You wrote that 975x has only 200MHz and 266MHz straps. But the truth is that this chipset has also 333MHz strap. All Asus boards switch to 333MHz strap at speeds over 400MHz. So this means that all boards in this article were running at 266MHz strap. It's true and my own testing proves that. At 400MHz FSB speed and 1000MHz memory speed I had memory bandwidth of 9401MB/s. But at 425MHz and 1063MHz memory bandwidth reduced to 9301MB/s. So 333MHz strap really exists on this board. This proves also that every chipset was overclocked by the same amount. So 975X simply has the best memory performance and it's not caused by the fact that it was overclocked the most.
 
I thought this was a great article in response to upgrading equipment however when I first saw the name of the article I was hoping that it would refer more to the new i7 boards out, according to NOV's article where it is mentioned that it would be interesting to explore the scaling of PCI slots and the use of the N200 chip from NVIDIA. As many know that would add the true 16 lanes to the additional two supported by the X58. For example, the new 2.0 x 16 is double the bandwidth. So that being said, does a dual GPU solution effectively take up 8 and 8 lanes on one slot? So far it doesn't seem like this matters much because current games do not use all that bandwidth. But for future games and graphics uses, its appealing to see if the need for a true 3 or 4 slot x 16 is needed in order to avoid using up all the bandwidth with a dual GPU-single card solution. The new ASUS P6T6 WS Revolution supports true 3 x16 slots and certainly from an SLI standpoint it is appealing ( if ASUS can ever fix their 3-way SLI problem with this board). If any moderators read this post it would wonderful to see some kind of tests measuring the throughput of information from dual GPU-single card solutions and if they're capable of using up all the bandwidth.

 
[citation][nom]ImmortalJman[/nom]I thought this was a great article in response to upgrading equipment however when I first saw the name of the article I was hoping that it would refer more to the new i7 boards out, according to NOV's article where it is mentioned that it would be interesting to explore the scaling of PCI slots and the use of the N200 chip from NVIDIA. As many know that would add the true 16 lanes to the additional two supported by the X58. For example, the new 2.0 x 16 is double the bandwidth. So that being said, does a dual GPU solution effectively take up 8 and 8 lanes on one slot? So far it doesn't seem like this matters much because current games do not use all that bandwidth. But for future games and graphics uses, its appealing to see if the need for a true 3 or 4 slot x 16 is needed in order to avoid using up all the bandwidth with a dual GPU-single card solution. The new ASUS P6T6 WS Revolution supports true 3 x16 slots and certainly from an SLI standpoint it is appealing ( if ASUS can ever fix their 3-way SLI problem with this board). If any moderators read this post it would wonderful to see some kind of tests measuring the throughput of information from dual GPU-single card solutions and if they're capable of using up all the bandwidth.[/citation]

No matter how many PCIe hubs (nforce200) you add, you're still limitted to the same bandwidth through the chipset.
 
Status
Not open for further replies.