PCI Express & CrossFire: Scaling Explored

[malveaux]

What about AMD chipsets?

 

[Lurker87]

[citation][nom]Crashman[/nom]You're right! The problem is trying to test a whole bunch of different resolutions. 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 really think it'd make sense to drop 1024x768. A person who games (or even is active) on 1024x768, and reads articles like this is so rare, in my opinion. The 1680x1050 is becoming so affordable at all sizes now, that really it should be replacing the 1024x768 as the minimum. Having 1680x1050, 1920x1200, and 2560x1600 as the standard resolutions for testing just makes much more sense. I mean, you are testing games like Crysis and Supreme Commander, and the like. For a person who is putting the cash into being able to run such games, and are wanting to know how best to get good FPS out of them is going to most likely want to have the eye candy to go with it, wouldn't you think?

It just seems a bit out of place to put in 1024x768 tests for such demanding games. I mean, really, how many people are running a 4870, let alone crossfiring it, on their 1024x768?

 

[radguy]

1024x768 and the low settings half of the article doesn't make sense here. Your testing bandwith issues for crossfire in games and then putting all the stress on the cpu where bandwidth doesn't become a factor. I mean your results show this pretty clearly. I think it would have been more worthwhile to add the 1920x1200 res and drop that half of your testing.
The other half of the article was some really great info that I have been waiting a long time to see. Appretiate it very much.

 

[hannibal]

Well it also tells that crossfire is not usefull untill you go to bigger resolutions. It can be usefull information to those people who has an older computer with "crossfire" motherboard.

 

[Pei-chen]

Great article; this is one more reason for me to keep my P5B Deluxe WiFi-AP Edition. I always believed SLI/Crossfire to be a waste of money especially on an old motherboard.

BTW, could you tell me if your "CPU undervolt at idle" function works with overclocked FSB? Mine wouldn't lower the voltage if everything wasn't set to auto; it would just lower the multiplier. Thanks.

 

[bounty]

1024x768 tells you max fps, it shows a situation when bandwidth is not a factor, so you have something to compare bandwidth limiting resolutions to.

P.S. a majority of the people on this site probably game between 1024x768 and 1280 x 1024. Most don't sit around playing Crysis all day (if they did it would have sold more)... and if they did, they could probably only run it at those settings. Not everyone can afford fast stuff, doesn't mean they don't read about it.

http://www.steampowered.com/status/survey.html

 

[jthorn]

Anti acid wrote: This article shows that even in the best conditions, x48 vs p45 is at most 5% difference. Price-wise, this confirms my observations that the lower priced P45 boards are much better performance/value than the x48 premium counterparts.

This conclusion strikes me the most and I agree with it. However, I remember an unflattering P45 article 2 months ago which seem to miss this point. For > 1680 X 1050, two PCIE 8X slots on a lowcost P45 MB seems to provide great crossfire capability (20 percent more FPS) if your budget can justify a second GPU.

 

[Crashman]

[citation][nom]jthorn[/nom]Anti acid wrote: This article shows that even in the best conditions, x48 vs p45 is at most 5% difference. Price-wise, this confirms my observations that the lower priced P45 boards are much better performance/value than the x48 premium counterparts.This conclusion strikes me the most and I agree with it. However, I remember an unflattering P45 article 2 months ago which seem to miss this point. For > 1680 X 1050, two PCIE 8X slots on a lowcost P45 MB seems to provide great crossfire capability (20 percent more FPS) if your budget can justify a second GPU.[/citation]

X48 would be a better choice for CrossFireX with two 4870X2's, but that's way beyond the scope of an article that only tried to answer the question "will a second card give me the performance boost I want"

 

[orangedrink]

[citation][nom]malveaux[/nom]What about AMD chipsets?[/citation]

I would love to see this too.

 

[avatar_raq]

P45 Is definitely the champion....Kind of disappointed about my P35 P5kSE though....I give 2 thumbs up for choosing ASUS mobos for this article, for me they are the best.
Thanks for the info...keep up the good work..and thank you crashman for watching the comments at the forum and for everything.

 

[kschoche]

Seriously? Owning a 30" screen doesnt mean that you are going to always have the latest and greatest computer, hell, the people that could afford these types of things are hit hard by the economy too....
A 30" 2560x1600 screen is a one time purchase that will last through several computers, or will last a single computer through several upgrades.

 

[avatar_raq]

I would like to see a comparison between the performance of (one 4870x2) and (two 4870 in crossfire) on different chipsets.

 

[avatar_raq]

1280x1024 is one resolution that many people are still using, especially for demanding games like crysis and world of conflict...Not every one can afford a $1500+ system, yet many are interested to read and undertand about current high-end stuff..
Again I agree with the guys at Tom's for including low resolutions in their benchmarks.

 

[bydesign]

I have a P35 Asus Blitz Formula which is more like a P45 in crossfire. Dual 8x lanes so my results would be similar to a P45 right? Or is more because of the PCI-E 1.1 vs 2.0. My guess is it's more about x16,x4.

 

[slomo4sho]

[citation][nom]Crashman[/nom]It's all about answering the question "Will a second card do the job".

Lots of guys have midrange or better ATI graphics cards, and the question of "upgrade or replace" is constantly being asked.[/citation]

The best way to answer that question would have been to compare 4850 in crossfire configuration and then match the performance with a single 4870. Or in the case of the 4870s in crossfire, it would be ideal to compare the crossfire configuration with a 4870x2.

The question really boils down to wither you would see at least 20% gains by upgrading a 4850 to a 4870. From your previous articles, its safe to conclude that a 4870 will provide at least 20% gains over the 4850 at 1680x1050 settings. This is especially true for anyone with a P35 or P965 chipset motherboard.

 

[avatar_raq]

[citation][nom]ByDesign[/nom]I have a P35 Asus Blitz Formula which is more like a P45 in crossfire. Dual 8x lanes so my results would be similar to a P45 right? Or is more because of the PCI-E 1.1 vs 2.0. My guess is it's more about x16,x4.[/citation]

I think you are talking about something ASUS calls "crosslinx" claiming it can enable crossfire in 8x 8x mode..Well I'm not sure it can reach for the same results of P45..Perhaps it's somewhere in between other P35 mobos and P45...Can't you check the gaming performance gain after adding a second GFX?? (barrowing it from a friend for example!)..Then compare it to the results above and tell us!!!

 

[avatar_raq]

[citation][nom]Slomo4shO[/nom]The best way to answer that question would have been to compare 4850 in crossfire configuration and then match the performance with a single 4870. Or in the case of the 4870s in crossfire, it would be ideal to compare the crossfire configuration with a 4870x2. The question really boils down to wither you would see at least 20% gains by upgrading a 4850 to a 4870. From your previous articles, its safe to conclude that a 4870 will provide at least 20% gains over the 4850 at 1680x1050 settings. This is especially true for anyone with a P35 or P965 chipset motherboard.[/citation]
I totally agree. Putting that in mind, one would prefer to keep his money and wait for a suitable single card from the next generation, since it will yeild more performance and avoid the potential problems and power requirements of multi-card setups.

 

[Primus462]

[citation][nom]orangedrink[/nom]I would love to see this too.[/citation]
Me too. Maybe they can put something similar together shortly after Deneb is released.

 

[yobigd20]

I have a gaming system that has an A8N-SLI Deluxe Motherboard with dual 7800GTX in SLI, socket 939 4400+ cpu, 2gb DDR ram, and 520W power supply. I know the MB uses nForce4 chipset, but would there be any hardware incompatibility if I added one 4850X2 since the dual chips in crossfire are on the same board? Do you think this would still be a good gaming system if I upgraded to 4gb DDR ram, vista x64 (for dx10), and the sapphire 4850X2?

 

[Niktavalos]

[citation][nom]yobigd20[/nom]I have a gaming system that has an A8N-SLI Deluxe Motherboard with dual 7800GTX in SLI, socket 939 4400+ cpu, 2gb DDR ram, and 520W power supply. I know the MB uses nForce4 chipset, but would there be any hardware incompatibility if I added one 4850X2 since the dual chips in crossfire are on the same board? Do you think this would still be a good gaming system if I upgraded to 4gb DDR ram, vista x64 (for dx10), and the sapphire 4850X2?[/citation]

Simply put, no. That processor is old as hell, you'd get better performance from a $50 AMD X2. If your gaming performance is hurting, a single 4850 is the highest I'd recommend, and even then you'd be severely processor-limited. You need a new processor/motherboard before you can truly see the performance gains from such a powerful video card.

 

[yobigd20]

[citation][nom]Niktavalos[/nom]Simply put, no. That processor is old as hell, you'd get better performance from a $50 AMD X2. If your gaming performance is hurting, a single 4850 is the highest I'd recommend, and even then you'd be severely processor-limited. You need a new processor/motherboard before you can truly see the performance gains from such a powerful video card.[/citation]

Lol. I know the processor is old. So is the chipset. How many chipset generations am I behind now? freakin p35s,p45s, now x58s...and sockets have changed..am2, now intel's 775 and of course the 1366...my box is old. Though I keep it in great shape. I know the cpu is the bottleneck. I just wonder how much of a bottleneck or would the fps gain still be sufficient enough to warrant a $400 last-investment.

 

[radguy]

[citation][nom]bounty[/nom]1024x768 tells you max fps, it shows a situation when bandwidth is not a factor, so you have something to compare bandwidth limiting resolutions to. P.S. a majority of the people on this site probably game between 1024x768 and 1280 x 1024. Most don't sit around playing Crysis all day (if they did it would have sold more)... and if they did, they could probably only run it at those settings. Not everyone can afford fast stuff, doesn't mean they don't read about it.http://www.steampowered.com/status/survey.html[/citation]

1024x768 shows you when gpu power is not a factor not the bandwidth. Showing when bandwith is not a factor can be done by increasing the bandwidth.
-------------
To some other comments and expressing some thoughts
The scope of this article is only to show differences pci express bandwith limiting and there boards vs crossfire gpus. The best cards to use are the fastest cards which could show bandwith limitations. 4870 x2 would be a better option to show this however there are too many problems with quad crossfire to get quality results accross the board.
I wouldn't say from these results that crossfire is only good for high resolutions. It shows that crossfiring two 4870 only shows a real benifit in most these games for a high resolutions because your still cpu bound. Crysis sees a 80% performance increase at 1680 vs 23% overal and is cpu bound at 1024 thats a pretty good increase in performance for the dollar.
The option of crossfire also is a debate for future upgradeability. Picking up a gpu and then another one when its cheaper and you need an a perfomance increase. This becomes hard to recommend given the recent history of graphics cards. The 3850 was the first card that really allowed for DX10 gaming at under 200 bucks. Picking up a second one today for gaming at 1280 or 1400 or 1650 would make a lot of sense except that ati came out with a huge winner in the 4800 series. This alone threw out many sli/crossfire plans that people had. The next series of gpus probably won't bring that type of performance jump for the dollar at least. So today picking up a 4850 or 4870 with plans to crossfire in the future will probably be a better idea than those who picked to do it in the 3800 series. Games with the graphics of crysis will be more common then and the benifit even at lower resolutions will show.
What is nice to know is that if you own a cheaper crossfire p45 board that the 8x pcie2.0 won't hold back the extra card by more than 5% up to a 4870. This shows that crossfire isn't just for those with highend budgets which is one reason I really appreciated the article.

 

[croc]

@ crashman and Thomas...

Wouldn't it have made more sense to organize this article by bus-type and lanes? ie, pci 1, 1.1 & 2, with 2x16, 1x16 and 1x8, 2x8 etc?

 

[Proximon]

I think you totally nailed this. You overclocked the CPU to avoid as much bottleneck as possible. I would only add that, in other comparisons elsewhere, x38 tended to fall right in the middle of x48 and P45.

I have wanted this for a few months, since drivers and BIOSes matured, but figured it wasn't going to happen once i7 hit.

 

[randomizer]

It's great to see that, in general, PCIe 1.1 still provides adequate bandwidth for new cards. It will be interesting to see how long that continues. What was the change going from 1.0x to 1.1?

I think 1024x768 can be dropped from benchmarks with high-end cards, but not midrange and low-end. 1280x1024 still has plenty of users with 19" LCD panels.