is this true? you need an i7 for crossfire.. sorta

[mmj_uk]

It's a complicated situation but the jist of it is:

1> Current games don't utilize quad core CPU's efficiently.
2> People play games at high resolutions inducing massive GPU bottlenecks.
3> i7 is much faster in games at low resolutions with no GPU bottleneck.

So although the Phenom gets a slapping from i7 in just about every situation... in high resolution gaming the Phenom is able to keep pace even with Crossfire, that could change in the future but right now it doesn't seem to be any hinderance.

 

[Rroff]

Thats not "reducing impact" though its giving the GPU "busy-work", nice looking "busy-work"

Thanks for the explanation though makes more sense when put like that.

I didn't quite explain it right but you get the gist... the GPU is taking up the slack by doing AA increasing the time per frame so that teh gap between work coming from the CPU is less GPU idle time.

 

[adamknox]

good point there, when overclocked with NB at stock it sux, as soon as the NB is overclocked it runs a lot better, in crysis i made about 40% higher fps from upping the NB, maybe something to do with the mobo but im sure its not.
so many review sites dont know about overclocking the phenoms NB

an overclocked phenom is fine for 5770's, running fine here

tbh im a 100% noob at overclocking -_- lol never even knew u shoudl touch the NB!! I just changed mine up to 2200 .. had to clear bios >< wouldnt start up lol fail

 

[GunsAkimbo]

It's a complicated situation but the jist of it is:

1> Current games don't utilize quad core CPU's efficiently.

Hopefully they will soon:

From intel's threading building blocks site:
Epic Games Unreal Engine and Intel® TBB 2.2: August 4, 2009: Epic Games, Inc. today announced that Intel® Threading Building Blocks (Intel® TBB) 2.2 has been incorporated into Unreal Engine 3. As part of this collaboration, Intel has joined Epic’s premier Integrated Partners Program (IPP).

Threading building blocks is pretty damn cool at using and abusing those spare cores.
I ran some (albeit simple) code on my PC (Core2Duo iMac, dual boot) the other day. First in visual studio with the MS C++ compiler, then again with the optimised Intel C++ compiler and it was twice as fast... then once again with the threading building blocks code.. and it was 3 times faster than the original run! and that's just with a C2D not a quad core!

in high resolution gaming the Phenom is able to keep pace even with Crossfire, that could change in the future but right now it doesn't seem to be any hindrance.

I doubt it; you can't offload the processing of those extra pixels to the CPU... and if you could it would be HORRIBLY slow. The only way it would pay off is if you;re doing some processing on the GPU (like AI, procedural generation, physics) that can be off loaded to a spare CPU core to free up GPU resources. (as I understand it)

 

[Marine-RX179]

If the bottleneck is due to the CPU being unable to feed the GPU data fast enough - turning up the AA means the GPU has something to do while waiting on the CPU - so you get nicer filtering instead of just losing a lot of fps anyhow.

Yea that's the way I understand it too. I think I was in same arguement with drunkenmaster before, and he's dead set on the idea that games are ALWAYS GPU limited.

But test has shown that even a single 5870 would be faster with a faster CPU when running the same game at the same settings:

HD5870, 4GB Corsair 800Mhz DDR2, 4GB Corsair 1,600Mhz DDR3
Crysis (DirectX 10, 64-Bit, High) at 1,680x1,050

[email protected] (stock speed)=12-35fps
Q6600 [email protected]=24-45fps
[email protected] (stock speed)=19-36fps
E8400 [email protected]=27-48fps
Phenom II X4 965 BE (C3 stepping) @3.4Ghz (stock speed)=23-46fps
Phenom II X4 965 BE (C3 stepping) [email protected] =29-52fps
Core i5 [email protected] (stock speed)=20-50fps
Core i5 750 [email protected]

(source: Custom PC Issue 078 March 2010, p41)

 

[GJ02]

What if I was to xfire 5870's with my q9650 @ 3.6?
Bottlenecked CPU here?

 

[johny dolan]

http://www.guru3d.com/article/core-i7-multigpu-sli-crossfire-game-performance-review/

 

[arc@css]

So myself, for example, wouldn't be bottlenecked running say two 5870's in xfire along side a quad cpu clocked at 3.4

Should be able to eek out many more months with my current rig then.

 

[GJ02]

whats the cpu ur using ?

 

[arc@css]

Me? I am using a Q6600 and currently have a 4870X2.

 

[GJ02]

Well from what I can make out from the 3dguru bench, at lower resolutions yes it does make a considereable difference, but at higher res it evens itself out, so I'm thinking that an i7 upgrade isn't required when I have a q9650 at 4.0 to run 5870 xfire

would you agree

 

[Rroff]

Yea that's the way I understand it too. I think I was in same arguement with drunkenmaster before, and he's dead set on the idea that games are ALWAYS GPU limited.

But test has shown that even a single 5870 would be faster with a faster CPU when running the same game at the same settings:

HD5870, 4GB Corsair 800Mhz DDR2, 4GB Corsair 1,600Mhz DDR3
Crysis (DirectX 10, 64-Bit, High) at 1,680x1,050

[email protected] (stock speed)=12-35fps
Q6600 [email protected]=24-45fps
[email protected] (stock speed)=19-36fps
E8400 [email protected]=27-48fps
Phenom II X4 965 BE (C3 stepping) @3.4Ghz (stock speed)=23-46fps
Phenom II X4 965 BE (C3 stepping) [email protected] =29-52fps
Core i5 [email protected] (stock speed)=20-50fps
Core i5 750 [email protected]

(source: Custom PC Issue 078 March 2010, p41)

Theres plenty of CPU limited games and plenty of CPU limited scenarios in games that are generally GPU limited - not always something benchmarks show as they might not include that section of the game.

Well from what I can make out from the 3dguru bench, at lower resolutions yes it does make a considereable difference, but at higher res it evens itself out, so I'm thinking that an i7 upgrade isn't required when I have a q9650 at 4.0 to run 5870 xfire

would you agree

Thats plenty enough to push 5870 crossfire at 1920x or higher res...

 

[marc512]

Usely expensive i7 motherboards will support 2 PCIe slots at x16 for sli and crossfire if thats what your meaning. Normal crossfire/sli boards will do 1 at x16 and 2 at x8

 

[Squark]

My mobo does 2@x16 or 4@x8 and that's an AM2+ board

 

[mmj_uk]

I doubt it; you can't offload the processing of those extra pixels to the CPU... and if you could it would be HORRIBLY slow. The only way it would pay off is if you;re doing some processing on the GPU (like AI, procedural generation, physics) that can be off loaded to a spare CPU core to free up GPU resources. (as I understand it)

I think you misunderstood me I meant that a Phenom is a sufficient CPU for running with Crossfired cards and isn't a major hinderance currently (compared to i7).

Most games at the moment probably use 50% of each core so the CPU is sitting idle a lot of the time.

 

[Marine-RX179]

Theres plenty of CPU limited games and plenty of CPU limited scenarios in games that are generally GPU limited - not always something benchmarks show as they might not include that section of the game.

I know. But Custom PC do their testing in the under same conditions, so I'm pretty sure they test it under the same environment in game.

I also remember seeing a test where the 5770 was tested at a game under the same condition with i5 750 stock speed (2.66GHz) and i5 750 overclocked to 3.8GHz. The result was that both frame rates were identical, so I'm guess it is suggesting that i5 750 at stock speed is already fast enough to push 5770 to 100% of it's capability, and pushing the clock speed of the i5 750 further wouldn't make the frame rate go any higher at that particular. But if I'm not mistaken, in the the case of heavy CPU dependant games with more data needed to be sent from the CPU to the GPU, the frame rate of the stock i5 750/5770 pairing would most likely drop in frame rate more/earlier than the overclocked i5 750/5770 pairing.

 

[mmj_uk]

I also remember seeing a test where the 5770 was tested at a game under the same condition with i5 750 stock speed (2.66GHz) and i5 750 overclocked to 3.8GHz. The result was that both frame rates were identical, so I'm guess it is suggesting that i5 750 at stock speed is already fast enough to push 5770 to 100% of it's capability, and pushing the clock speed of the i5 750 further wouldn't make the frame rate go any higher at that particular..

Yep all that overclocking from 2.66ghz to 3.8ghz would do in most current games is reduce the overall cpu utilization from say 75% to 40% whilst still giving the same frame-rate, until CPU's are utilized more in game it will continue to be the case that basically any 3ghz quad is sufficient for most games (including C2Q).

 

[rolypolyman]

I also remember seeing a test where the 5770 was tested at a game under the same condition with i5 750 stock speed (2.66GHz) and i5 750 overclocked to 3.8GHz. The result was that both frame rates were identical, so I'm guess it is suggesting that i5 750 at stock speed is already fast enough to push 5770 to 100% of it's capability, and pushing the clock speed of the i5 750 further wouldn't make the frame rate go any higher at that particular. But if I'm not mistaken, in the the case of heavy CPU dependant games with more data needed to be sent from the CPU to the GPU, the frame rate of the stock i5 750/5770 pairing would most likely drop in frame rate more/earlier than the overclocked i5 750/5770 pairing.[/QUOTE]

This.

 

[HangTime]

Thats not "reducing impact" though its giving the GPU "busy-work", nice looking "busy-work"

Thanks for the explanation though makes more sense when put like that.

OK, perhaps "reducing the impact" was a poor choice of words.

What I was driving at, is that (unless you have very high end graphics hardware) you can often engineer a situation where you aren't cpu-limited anymore. You can crank up the AA to make use of some of that extra graphics power.

Contrast this with situations where you are GPU limited. Pretty much the only option is to lower the screen resolution (unless it's a case of insufficient VRAM, in which case you'd lower the textures). But with TFT monitors changing res is generally a bad idea since you really want to be running at native res. Obviously for those with CRTs they have the best of both worlds because they have total flexibility over resolution and can tailor it to suit.

In other words, when cpu-limited you at least have the option to increase graphical quality while maintaining similar performance levels. When you are gpu-limited, there isn't much you can do to make things better at that performance level, except maybe bump up physics or something like that, and even that is starting to get phased out thanks to PhysX etc.

Ultimately when it comes to (most) gaming you get better value from spending an extra £100 on your gpu compared to an extra £100 on your cpu. The kind of settings most enthusiasts like to play at (~1080p with AA and AF, everything on 'high') usually lend themselves to gpu-limitations and something like a Phenom II X4 paired with a 5850 will likely outperform an i7 with GTX260 in a lot of games.