Debunking the "PhII sucks at Crossfire/Trifire" myth

[keithlm]

We've all seen the reviewers who have done some comparisons based on "bottlenecks" and come to the conclusion that the i7 is much better than the PhII with multiple video cards. They didn't actually prove it, but they did come to that conclusion.

Let's do our own little comparison using the Futuremark ORB and 3dMark Vantage GPU scores. I've tried to search for stock speeds on the CPU and GPU but it is not always obvious. Regardless it doesn't matter anyway... because the GPU scores appear to be more dependent on the GPU and how many of them exist versus the CPU frequency anyway. (Overclocking the GPU can get you a few hundred more points also; but again the ballpark figures are going to be about the same.)

SO other than my own results below, the other results are guesses based on what appears to be the "average" score for the GPU type and # of GPU. But even if you go to the orb and pick the highest and/or lowest it won't really affect my final results.

================================================
Lets looks at 4850 cards:

(Remember these are the GPU scores.)
With a Phenom II at 3.0Ghz and 4850 video cards at stock I myself have the following results:
1x4850 = 6652
2x4850 = 11771 Gain of 77% over single
3x4850 = 15233 Gain of 29% over dual

-----------
Searching for i7 at stock on the Futuremark orb I can only find a 1x4850 and 2x4850:

1x4850 = 7136
2x4850 = 11577 Gain of 62% over single
(No 3x4850 results available for the i7 at this time.)

NOTE: The single card scores is higher for the i7 than for the Phenom II. But going to dual cards... the Phenom actually gains MORE than the i7 chip and actually pulls ahead of it.


=============================================
Just for fun lets look at 4870 cards.

(Remember these are the GPU scores.)
Phenom II 940 at 3.0Ghz
1x4870 = 8547
2x4870 = 15317 Gain of 79.2% over single
3x4870 = 21059 Gain of 37.49% over dual

i7 at around 2.8Ghz
1x4870 = 9044
2x4870 = 15005 Gain of 65.9% over single
3x4870 = 19812 Gain of 32.0% over dual

===========================================

RESULTS: Guess what? The Phenom II 940 does just fine at Crossfire and Trifire. In fact it does better than the i7.

Not only does it scale better... but it ends up with higher end scores at the top than the i7; even though the i7 chips start with higher single card GPU scores.

Based on this data... it appears that claiming that the i7 is better than the PhII with multiple video cards is not entirely accurate.

You might be able to find a few scores that change the "who gets the highest at the top" but it won't really change the percentages by much. And thus even if someone went to the trouble to do that... the only thing they could show is that both chips are more equal and not that the i7 is better. But with the size of the percentage gains, I'm going to guess it would be hard to do that anyway.

[demonkevy666]

perhaps it's due to the L1 cache.
L1 cache size maybe ? 64 kbytes phenom II vs 32kbytes I7
more raw date can flow in bigger packets.
SMT isn't going to help when the core can only take so much data.
adding more doesn't satisfy video cards all time.
core speed can change things

[roofsniper]

interesting. have you tried it in any games?

[keithlm]

interesting. have you tried it in any games?

Mostly just LOTRO... I've been using Linux too much lately.

Need to try the new AOD version though.

EDIT: No big change to the new AOD version. Doesn't really do anything different that I can see; it works but then so did the last version.

[cdawall]

it doesn't surprise me i bet the AMD chipset just runs the cards better. probably all implementation of the PCI-e lanes and TBH intel boards suck at that

[keithlm]

it doesn't surprise me i bet the AMD chipset just runs the cards better. probably all implementation of the PCI-e lanes and TBH intel boards suck at that

True... but these results completely contradict a common theory that is accepted by a lot of people in many forums.

In many forums people will say things similar to the following: "The PhII is better in many games at playable settings if you are going to use a single card... but everyone knows the i7 is always better if you plan to use multiple cards."

But in reality the data contradicts that theory.

[cdawall]

True... but these results completely contradict a common theory that is accepted by a lot of people in many forums.

In many forums people will say things similar to the following: "The PhII is better in many games at playable settings if you are going to use a single card... but everyone knows the i7 is always better if you plan to use multiple cards."

But in reality the data contradicts that theory.

i have always seen intel+single card and AMD with multiple my A64 scaled better with a multi card setup than some intels i tested...dont quote that 100% though i compared with two way diff mobo's

[Titan7171]

when looking at those scores and looking at my 4850x2 (single card) results......it makes me happy I picked the card I got, My gpu score is 13,921 with cpu at 4.1ghz with it at 3.8(24/7 settings I get 13,753 not to bad for this card at all.

[demonkevy666]

i have always seen intel+single card and AMD with multiple my A64 scaled better with a multi card setup than some intels i tested...dont quote that 100% though i compared with two way diff mobo's

the gigabyte Dq6 would be the only board that is AMD an Intel setup that I know.

[keithlm]

when looking at those scores and looking at my 4850x2 (single card) results......it makes me happy I picked the card I got, My gpu score is 13,921 with cpu at 4.1ghz with it at 3.8(24/7 settings I get 13,753 not to bad for this card at all.

I need to remove the single 4850 and test just my 4850x2 and see what happens....

Right now when I just disable it... I can't do full screen crossfire because it REALLY messes up. (Wavy lines etc.)

Maybe if I just removed the crossfire bridge. I'll try that first.

[slpdLoad]

Wouldn't this make sense because AMD has a hand in the CPU, chipset, and GPU? :shrug:

[roofsniper]

Wouldn't this make sense because AMD has a hand in the CPU, chipset, and GPU? :shrug:

maybe but still as the op is saying chipset or not people are saying i7 is better for dual cards. which is what he is trying to prove right here. people always say the reason why amd is better at something is because of the chipset. well that doesn't make sense because when was the last time i saw an i7 965 running with a 790fx?

[keithlm]

maybe but still as the op is saying chipset or not people are saying i7 is better for dual cards. which is what he is trying to prove right here. people always say the reason why amd is better at something is because of the chipset. well that doesn't make sense because when was the last time i saw an i7 965 running with a 790fx?

Yep. I got tired of many posters on various forums just throwing the "PhII is good for gaming with a single card but i7 is much better if you plan on multiple video cards" and nobody questioning that theory.

When enough people say it and others accept it without question then pretty soon many people online will just accept it as being an indisputable fact without looking any further.

Then later if you ever attempt to tell them they are not accurate, many people will flame you for not just accepting the general consensus even though it was created by inaccurate data.

[demonkevy666]

Yep. I got tired of many posters on various forums just throwing the "PhII is good for gaming with a single card but i7 is much better if you plan on multiple video cards" and nobody questioning that theory.

When enough people say it and others accept it without question then pretty soon many people online will just accept it as being an indisputable fact without looking any further.

Then later if you ever attempt to tell them they are not accurate, many people will flame you for not just accepting the general consensus even though it was created by inaccurate data.

I though that was because they bench low resolution more often...?

[Oldguy932]

I actually said that even the Phenom I's were much better at xfire scaling than their equivalent speed in a core 2 quad. I showed it with my xfire scores on hwbot with 3DMark03, 05, and 06 with an x1650xt. While the cards themselves weren't very good at all, my Phenom I at 3ghz was beating core 2's that were I think at 3.6ghz. I haven't done much on hwbot lately so I don't really have an idea on any of it. I might just upgrade my brothers comp to get those cards back and rebench for scores on my Phenom II.

While this is slightly off topic by not being xfire, I wonder how i7 and Phenom II compare with single to multiple card with nvidia. Then we could really see whether its pcie implementation or just that fact that amd makes the cpu, chipset, and graphics.

[merkk1]

Well i think my rig is like the poster child for crossfire . I'am running 4 3850 i have score 22,137 in 3dmark-06 i pull down a 140 FPS score in canyon flight test (test of Crossfire/SLI) i love the way my rig works . One of teh best thing is i don't even need to push the CPU all that hard to break into the 20,000 point land . It only take a 3.6 clock on the cpu to break the into the 20,000 land and i'am just running some cheap 256 mg 3850 the lowest molde they make . I still can not belive how well it plays games the card are a dead on macth for running at my max gaming rez.1440x900 (19"") LCD

[vinister]

Just wanted to chime in here.

I went from 9850BE at 2.5Ghz (would not allow OC), to a 720BE X3 at 3.6Ghz. That's the only change.

Check out this gain:

http://lh4.ggpht.com/_uo9kciAVieo/SajUcqUNoGI/AAAAAAAADjs/itBgb0AN_-Q/s800/3dmark-X3-3600.jpg (http://picasaweb.google.com/lh/photo/PQkyC_gAeBsSxXLR0nz4Uw?feat=embedwebsite)

I'm going to be picking up the MSI board with 4 slots when it becomes available, and I will be looking for a 4th 3870 ;)

[demonkevy666]

keithlm what resolution where you running I think higher up the phenom

[keithlm]

keithlm what resolution where you running I think higher up the phenom

Personally... I run 1680x1050.

But 3dMark Vantage doesn't care... it uses set resolutions. The "P" resolution shown is 1280x1024.

I can also run benches using the "H" resolution which uses the 1680x1050 resolution. (Or the lower "E"ntry which is 1024x768.)

(But nobody really ever uploads the "H" or "E" results to the Futurmark Orb...)

[TheHolyLancer]

I honestly its things like:

http://www.pcgameshardware.com/aid,692942/Catalyst-98-reviewed-HD-4870-X2-up-to-47-percent-faster-failing-in-Anno-1404/Practice/

and


http://images.anandtech.com/graphs/phenomiicf_012809142247/18099.png

that got people on the track about this whole thing....


I have to ask about 3d marks and how the other parts of the system were done, but yes, these res are very much in the lower end (esp the one in the first link where it is 1280*1024)


BTW the reason for GY digging is because over at toms, people are quoting this article, and I wanted some more feed back here before saying this thing is looking at it from 3d mark only.

[informal]

You picked out a chart for a game that is well known to run better on intel hardware than AMD's... Yeah,it's Far Cry 2. In almost any other game there is no such big of a difference as there is in this one.

[FlawleZ]

Could do more testing with a few games? I'm looking at grabbing a second HD4850 myself but I'd like a bit more confidence that the boost in performance would be worth it on my system in games and not just 3DMark scores.

[Smartidiot89]

Holy mother of bump!:eek:

Did it take you more then 6 months to find this piece of information and then post it on th AMD section? :p: