Getting to the bottom of the bottleneck question.

[happy medium]

The question of bottlenecking comes up often on this forum.
I thought it would be nice to have a sort of general guide for a balanced system for gamers.

To get the most out of you system you don't want to pair a stock dual core cpu with sli'd gtx285's or a crossfired 4890's with a 21 inch monitor.
It's a shame seeing people wasting their money like this.

http://forums.anandtech.com/me...=2304213&enterthread=y

Best cpu, gpu, and resolution for smooth gameplay at high detail.

Ultra high end: tri sli/Tri fire (GTX 285/280/275 or 4870/4890)
Best matcing cpu: Core i7 @ highest overclock
Minimum cpu: Quad core @ highest overclock
Recommended Resolution: 30 inch monitors +

High end: sli/crossfire (gtx 285,275,260 sli or 4870/4890 crossfired)
Best cpu: Core i7 @ 3.6, Quad core @3.8 or Phenom II @ 3.8
Minimum cpu: E8xxx @ highest overclock or Phenom II x3 @ 3.6
Res. 2560x1600 max, 1900x1200 min

High/med range sli/crossfire (gts 250/9800gtx or 4850/4770) or single gtx285 or overclocked 4890
Best cpu: quad core @3.0 , e8xxx @ 3.4, Phenom II @ 3.2
Minimum cpu: e8400 stock, e5xxx @ 3.4, Phenom x3 @ 3.1
Res. 1900x1200 (1gb version only), 1650x1050 min.

Med range sli/crossfire (88/9800gt, 4830)or single gtx 260 or 4870
Best cpu: Quad core @2.7, duel core e7xxx@ 3.0 , Phenom II x3 @2.7
Minimum cpu: dual core e5xxx@2.6
Res. 1900x1200 (1gb version or gtx only) 1650x1050 512mb version

Lower end Single cards (9800gtx, or 4850/4770).
Cpu: Athlon x2 @3.0, Pentium e5400 dual core.
Minimum cpu: Athlon x2 @ 2.5, Pent d @ 3.6
Res: Under 1900x1200

A great site to explore the bottleneck question here.

http://alienbabeltech.com/main/?p=4090


E8400 @ 3.0 vs i7 765 stock @ 3.2.....................

Brothers in Arms @ 1900x1200

4870 crossfire

62 fps vs 83 fps

4870 x4

63 fps vs 85 fps

260 sli....

55 fps vs 95 fps

280 sli ......

50 fps vs 97 fps

280 gtx 3x sli

53 fps vs 107 fps


http://www.guru3d.com/article/...e-performance-review/7

Call of duty 4 @ 1900x1200

e8400 @ 3.0 vs core i7 965 @ 3.2 @ 1900x1200

4870 crossfire

132 fps vs 124 fps

4870 x4

169 fps vs 219 fps

260 gtx sli

117 fps vs 117fps

280 sli

127 fps 132 fps

280 3x sli

145 fps 183 fps

http://www.guru3d.com/article/...e-performance-review/9

Far Cry 2 @1900x1200

4870 crossfire

50 fps vs 63 fps

4870x4

53 fps vs 94

260 sli

54 fps vs 77 fps

280 sli

55 fps vs 84 fps

tri 280's

52 fps vs 112 fps

http://www.guru3d.com/article/...-performance-review/11

Crysis Warhead @ 1900x1200

4870 crossfire

25 fps vs 32 fps

4870 x4

28 fps vs 35 fps

260 sli

38 fps vs 36 fps

280 sli

40 fps vs 50 fps

3x 280 gtx;s

35 fps vs 55 fps

http://www.guru3d.com/article/...-performance-review/15

3x 280 sli stock core i7 (3.2) vs overclocked @ 3.7

Brothers in arms

107 fps vs 115 fps

Call of duty 4

183 fps vs 183 fps

Far cry 2

112 fps vs 112 fps

F.E.A.R.

266 fps vs 266 fps

Crysis Warhead

55 fps vs 55 fps

http://www.guru3d.com/article/...-performance-review/15


E8400 @ 3.0 vs qx9770 @ 3.2 vs core i7 965 @ 3.2

Brothers in arms 1900x1200

53 fps, 67 fps, 107 fps

Far Cry 2 @ 1900x1200

52 fps, 77 fps, 112 fps

F.E.A.R.

229 fps , 250fps, 266fps

Crysis Warhead

32 fps , 42 fps ,55 fps

3d mark Vantage 280 tri sli

E8400 - 12219
QX9770 - 19057
i7 965 - 25070


http://www.guru3d.com/article/...-performance-review/19


I don't recommend using dual card low end set ups. (9600gso, 4670 ect. ect. )
I recommend you buy a single card with similar performance.

Just a general guide. Any big mistakes or changes needed?


<Single card game benchmarks next week. June 1st 2009

 

[happy medium]

Is everyone asleep or just don't care?

 

[Qbah]

Well, I'd say people investing into an ultra-high or high end system know their sh!t and won't pair GTX285s with a 2GHz Core 2 Duo system.

I think if you want to have a pretty balanced system it's enough you get a C2D/C2Q 2.5-3.0GHz for single card setups (24" and lower screens). Actually, since Quads are so cheap these days, I don't see the point buying a C2D. Also, unless you're aiming for 25x16 screens, going dual cards in usually overkill.

Plus: "duAl" not "duel" and "their" not "there"

 

[happy medium]

Did you click the link at the top of my post and check out the benchmarks for 3d mark 2006? The guy went from a e8400 stock with dual 8800gtx's with 14,000 points to dual gtx 285's scoring 14,500.
If you look at his prior systems and scores he's been wasting money for years.

At this present time there are 4 post in video on the front page alone that deal with some kind of bottlneck question.

I believe it could help potential buyers of crossfire/ sli with weak cpu's also.

 

[cbn]

Good thread. I like these type of discussions.

I think for most people knowing what CPU/memory/GPU give the highest FPS from a performance/dollar ratio matters alot.

 

[Golgatha]

High/med range sli/crossfire (gts 250/9800gtx or 4850/4770) or single gtx285 or overclocked 4890
Best cpu: quad core @3.0 , e8xxx @ 3.4, Phenom II @ 3.2
Minimum cpu: e8400 stock, e5xxx @ 3.4, Phenom x3 @ 3.1
Res. 1900x1200, 1650x1050 min.


On this setup, I would recommend not including 1900x1200 (1920x1200), as several of those cards only have 512MB of memory on them. You'll run into a lot of situations with triple buffering, AA, and AF enabled that will dog down because you'll run out of video memory. For 1680x1050 resolutions, any of those setups would be fine though.

 

[Golgatha]

Originally posted by: happy medium
Is everyone asleep or just don't care?

Oh, and at 2am. Yes, I am asleep

 

[cbn]

Originally posted by: happy medium
Did you click the link at the top of my post and check out the benchmarks for 3d mark 2006? The guy went from a e8400 stock with dual 8800gtx's with 14,000 points to dual gtx 285's scoring 14,500.

But was this 3D Mark 2006 score run using the free trial version? (ie, 1280x1024 with no AA)

I'll bet if this same guy was able to choose higher resolution and AA he would start to see a much bigger difference with those 285 GTXs compared to the 8800s.

 

[Golgatha]

I would edit the resolutions to 1920x1200, 1920x1080 (1080p), and 1680x1050. Also, for the medium and low end categories, I would just put 1680x1050 or less. Some of those cards aren't going to run 1920x1200 well at all in modern games.

 

[cbn]

With respect to VRAM limitations Is this because increased challenge on the GPU slows things down to the point where the VRAM becomes more and more filled (ie, data is filling the ICs faster than it is leaving)

But at high frame rates this spillover into memory doesn't progressively accumulate (due to the time factor).

So is it correct to say if someone wanted really high frame rates (for full use of 120 Hz display) VRAM size isn't nearly as important? The way I see it VRAM size is most important only when the tasks on the GPU threaten to slow it down. Is this right or wrong?

 

[happy medium]

I chose 1900x1200 because most review sites use it and it's close to 1080p.

Other changes made for resolutions. Added 1gb for 1900x1200.

Thanks for the input.

 

[happy medium]

Originally posted by: Just learning

Originally posted by: happy medium
Did you click the link at the top of my post and check out the benchmarks for 3d mark 2006? The guy went from a e8400 stock with dual 8800gtx's with 14,000 points to dual gtx 285's scoring 14,500.

But was this 3D Mark 2006 score run using the free trial version? (ie, 1280x1024 with no AA)

I'll bet if this same guy was able to choose higher resolution and AA he would start to see a much bigger difference with those 285 GTXs compared to the 8800s.

Good point, but he is still using a stock e8400 with dual 285 gtx's.
That's a waste.

 

[Munky]

Why are we using 3dmock as any sort of meaningful benchmark? I stopped caring about synthetic scores a long time ago, and instead only look at performance in actual games and apps. If you want to explore this topic with more empirical data, we can run some benchmarks on various systems.

For example, I just upgraded my system and I also ran a bunch of benches with some modern games like Farcry2, Mass Effect, Bioshock, COD4, WIC, Stalker CS. I compared an oc'd Opteron 165 + 8800gt, same cpu + 4890, and then an oc'd Phenom2 x3 + 4890. In most cases a 4890 provided a noticeable improvement over a 8800gt, even when combined with the old cpu. However, there are definitely some cases where moving to the faster x3 cpu increased the gap significantly. And we are talking about 1920x1200 gaming with high settings, not some pitiful 1024x768 low quality settings than nobody uses.

 

[Golgatha]

Originally posted by: Just learning
With respect to VRAM limitations Is this because increased challenge on the GPU slows things down to the point where the VRAM becomes more and more filled (ie, data is filling the ICs faster than it is leaving)

But at high frame rates this spillover into memory doesn't progressively accumulate (due to the time factor).

So is it correct to say if someone wanted really high frame rates (for full use of 120 Hz display) VRAM size isn't nearly as important? The way I see it VRAM size is most important only when the tasks on the GPU threaten to slow it down. Is this right or wrong?

As I understand it, the reason the system slows down is because the GPU can't get the texture data it needs fast enough. If the video RAM fills up, the textures start getting cached to system RAM, which is way slower than getting the textures from video RAM, and the GPU is basically waiting on data to calculate constantly. This is experienced as framerates crashing to the point that the game will be unplayable (aka the game gets very choppy). One can run out of video RAM at 1680x1050 (for 512MB cards) if triple buffering and extremely high levels of AA and AF are used in modern games as well, but these cases are few and far between, and can be remedied by simply turning down the AA.

If someone wants high framerates, they need a powerful GPU(s) and enough memory to hold all the texture data locally on the video card.

This is why 4-way SLI/Crossfire X don't make sense with 512MB cards (not enough video memory to cache textures at the resolutions and settings capable by the multiple GPUs), and weak GPUs paired with ridiculous amounts of video RAM are just a waste of good RAM (e.g 9500 GT with 1GB onboard video memory).

 

[Golgatha]

Originally posted by: munky
Why are we using 3dmock as any sort of meaningful benchmark? I stopped caring about synthetic scores a long time ago, and instead only look at performance in actual games and apps. If you want to explore this topic with more empirical data, we can run some benchmarks on various systems.

For example, I just upgraded my system and I also ran a bunch of benches with some modern games like Farcry2, Mass Effect, Bioshock, COD4, WIC, Stalker CS. I compared an oc'd Opteron 165 + 8800gt, same cpu + 4890, and then an oc'd Phenom2 x3 + 4890. In most cases a 4890 provided a noticeable improvement over a 8800gt, even when combined with the old cpu. However, there are definitely some cases where moving to the faster x3 cpu increased the gap significantly. And we are talking about 1920x1200 gaming with high settings, not some pitiful 1024x768 low quality settings than nobody uses.

3DMark is useful in that it can easily show you when you're getting no benefit in games by overclocking your CPU (e.g. a single 4850 with a Core 2 Duo vs Core i7 will get the same score at high resolutions), as the scores get GPU limited at some point. As far as telling you how well a certain game will play though, I agree with you, it is completely useless.

 

[happy medium]

Originally posted by: Just learning
Good thread. I like these type of discussions.

I think for most people knowing what CPU/memory/GPU give the highest FPS from a performance/dollar ratio matters alot.

Thanks that was exactly the idea.

Some games require more cpu and some gpu. I wish there was a way to integrate this with specific titles because when choosing the right video card the claus "It depends on games you play " comes up alot also.

 

[OCGuy]

What good are i7 or Yorkie on games that dont use more than 2 or 4 real/virtual cores, or scale poorly even if they do?

 

[happy medium]

Originally posted by: OCguy
What good are i7 or Yorkie on games that dont use more than 2 or 4 real/virtual cores, or scale poorly even if they do?

I'm gonna compile a list of popular games and the best cpu,gpu combination this weekend.

See what games scale better with quad core vs dual core and sli/crossfire too.
Mabe I can get some help from the community?

For instance we all know that quad core cpu's fair much better then dual core cpu's in GTA 4 and Far Cry 2 and that Crysis requires both a fast cpu and a high end video card.

 

[Barfo]

DuAl Core.
DuAl Cards.

OP gives me headaches.

 

[wchang99]

I'm really appreciating this information (as poster of one of the threads you might be talking about).

I feel better knowing that my upgrade path, with help from people, would be spot-on (am I really at the very bottom? Hehe, guess I am).

 

[SickBeast]

IMO the CPU requirements are too steep in this guide. CPU limitations tend to be overblown.

 

[cbn]

Originally posted by: happy medium

Originally posted by: Just learning
Good thread. I like these type of discussions.

I think for most people knowing what CPU/memory/GPU give the highest FPS from a performance/dollar ratio matters alot.

Thanks that was exactly the idea.

Some games require more cpu and some gpu. I wish there was a way to integrate this with specific titles because when choosing the right video card the claus "It depends on games you play " comes up alot also.

To me knowing what games require more CPU and memory is higher yield.

I mean as long as the video card is at least quasi decent then resolution, AA/AF and detail settings can be adjusted to regain frame rates.

P.S. Since getting my 4770 yesterday I am now convinced Anti-aliasing effects might be overly hyped up. Or maybe I haven't sampled enough games? (most of what I have are demos) For most people on a budget simply being able to run native resolution on a LCD (without slowing the game down) is probably the highest yield benefit. After that maybe detail settings.

 

[fffblackmage]

so my comp won't be balanced if I get myself a 4890 and I'm still using my e6320? =/

seems odd if i upgrade to a e5x00 since I'm getting less cache. and then it just seems too much $$$ if get an e8x00. =/

 

[cbn]

Originally posted by: fffblackmage
so my comp won't be balanced if I get myself a 4890 and I'm still using my e6320? =/

seems odd if i upgrade to a e5x00 since I'm getting less cache. and then it just seems too much $$$ if get an e8x00. =/

It would be balanced if you were planning to use more resolution/detail settings/AA/AF along with that 4890.

To me "balanced" means the demands you have on your video card aren't excessive or too lignt (either of which affect GPU computational times relative to the CPU computational times).

 

[cbn]

http://www.anandtech.com/video/showdoc.aspx?i=3517&p=8

http://www.anandtech.com/video/showdoc.aspx?i=3517&p=9

Speaking of this CPU vs GPU bottleneck issue here is an example of how it can be game dependent.

Comparing 1680x1050 resolutions for Fallout 3 and Crysis warhead it appears they both put very similar demands on the quad core test CPU but Crysis is much harder on the Graphics card than Fallout 3 is.

So alluding to what Happy Medium said earlier Generalizing what is a balanced system can be confounded by unbalanced demands different games put on systems.

What is a balanced system for one game can be unbalanced for another game @ same resolution.