Why does phenom perform better than an intel quad at high resolution in Lost Planet?

[BenchZowner]

Since you guys want more on it.
I'll run some ( only Core 2 Quad at the moment though ).

[JumpingJack]

also memory bandwidth plays a role.

Not as big a role as you would think.... I did AMD at DDR2-667, 800, 1066 and the runs varied by about 5-8 FPS.

The QX9650 actually lost 2 FPS at DDR2-1067

Jack

[Devil's Prophet]

IMO both cpu's have their advantages over each other.

The way I see it is that both cpu's have their disadvantages over each other likewise.

I just want a perfect rig, damnit.

I like brute power, but I also like multicore smoothness.

[AlabamaCajun]

For the most part all I've seen is Phenom running a little smoother in a number of gamed and some intels doing most things faster. In most cases the intel showed more power.

T
I just want a perfect rig, damnit.

NO! Perfection is not achievable else we can't build a better-one tomorrow

Jako :p

At least we are reading and showing out presence while trying to lighten the atmosphere a little. In your defence, most of the time we would be more interested in results but most of it turns into arguments backed by shaded internet facts. The train wreck was cool.

[Jakko]

Not as big a role as you would think.... I did AMD at DDR2-667, 800, 1066 and the runs varied by about 5-8 FPS.

The QX9650 actually lost 2 FPS at DDR2-1067

Jack

Thanks for that info, that makes it even stranger.
Basicly the only area in which the phenom had a distinct advantage was the mem BW...

I guess it could it even be the 790FX chipset vs the X48.
But then, they both offer PCIE *16, so that shouldn't matter.

[JumpingJack]

Thanks for that info, that makes it even stranger.
Basicly the only area in which the phenom had a distinct advantage was the mem BW...

I guess it could it even be the 790FX chipset vs the X48.
But then, they both offer PCIE *16, so that shouldn't matter.

More strangeness .... I re-ran a 1280x1024 run using an HTT multiplier to set the HTT link to 800 Mhz (that's down from 2000 Mhz):

Here is the data for both runs:

Phenom 9850 2.5 GHz 2.0 GHz NB 2.0 Ghz HTT and DDR2-1066


Phenom 9850 2.5 GHz 2.0 GHz NB 800 MHz HTT and DDR2-800


The data at 10280x1024 in the article is 2.5 GHz Core 2.0 GHz NB 2.0 GHz HTT and DDR2-800

Neither snow or cave show much of a huge difference, Snow gets about 8 FPS (EDIT at low 'CPU limited regime' resoluionts) and generally better with DDR2-1066 but the HTT at 800 has no effect.

jack

[SkullCracka]

More strangeness .... I re-ran a 1280x1024 run using an HTT multiplier to set the HTT link to 800 Mhz (that's down from 2000 Mhz):

Here is the data for both runs:

Phenom 9850 2.5 GHz 2.0 GHz NB 2.0 Ghz HTT and DDR2-1066


Phenom 9850 2.5 GHz 2.0 GHz NB 800 MHz HTT and DDR2-800


The data at 10280x1024 in the article is 2.5 GHz Core 2.0 GHz NB 2.0 GHz HTT and DDR2-800

Neither snow or cave show much of a huge difference, Snow gets about 8 FPS (EDIT at low 'CPU limited regime' resoluionts) and generally better with DDR2-1066 but the HTT at 800 has no effect.

jack

i have noticed that low ht and low nb doesnt make much difefrence either.
is it that the ondie ht,nb is such low latency that it is still just as fast as the memory and still has a gerater or equal bandwidth. not sure. but i did find it interesting.

in Tonys 1000mhz thread he mentioned this once. that he noticed the cpu take off and get better clocks with lower ht and nb. less heat probly in tony's case. but i dont know im just a dumb newb

[JumpingJack]

i have noticed that low ht and low nb doesnt make much difefrence either.
is it that the ondie ht,nb is such low latency that it is still just as fast as the memory and still has a gerater or equal bandwidth. not sure. but i did find it interesting.

in Tonys 1000mhz thread he mentioned this once. that he noticed the cpu take off and get better clocks with lower ht and nb. less heat probly in tony's case. but i dont know im just a dumb newb

Yeah, in terms of an OC I can understand this... but from a pure computational flow of 1' and 0', it is hard to figure out why Phenom supports higher FPS in a GPU bound mode... it is interesting nonetheless.

Of all the contemporary games out there Lost Planet is one of the 'neater' ones to fool around with from a bench perspective. Between LP and UT3, they do the best job of utilizing quads.

EDIT: PS -- I have only ever met one person in the enthusiast forums I would catagorize as 'dumb', you are not that person
Jack

[Jakko]

Ok, it's either the phenom or the 790FX.


The snow benchmarks at resolutions around 1600*1200 are mostly gpu limited right?
So this means it shouldn't matter much, going from a phenom 2.5 Ghz, to a dual core K8 at say 3.2 Ghz.
If the AMD system then still benches higher than the intel quad core system, it has to be a chipset > GPU connection thing, giving the AMD setup the advantage.

If this test scores lower than the intel system, it has to be a phenom architecture thing.

[JumpingJack]

Ok, it's either the phenom or the 790FX.


The snow benchmarks at resolutions around 1600*1200 are mostly gpu limited right?
So this means it shouldn't matter much, going from a phenom 2.5 Ghz, to a dual core K8 at say 3.2 Ghz.
If the AMD system then still benches higher than the intel quad core system, it has to be a chipset > GPU connection thing, giving the AMD setup the advantage.

If this test scores lower than the intel system, it has to be a phenom architecture thing.

That is reasonable....

[BenchZowner]

Alright, let's see what we've got here:

System Specs:
CPU: Intel Core 2 Extreme QX9650 ( always at 333MHz FSB ) [ QDR1333 ]
Mobo: Asus Striker II Extreme 790i Ultra SLI [ BIOS 0704 ]
RAM: CellShock DDR3-1800 2GB Kit
VGA: Single GeForce 9800GTX 512MB Ref. Card ( Normal Stock Clocks )

Normal RAM Timings: A-A-8-7-6-21-1T-7-35-10-25-35-9-7-74
Tweaked RAM Timings: E-E-7-7-6-21-1T-1-1-5-21-1-1-1-74

Game Settings:
Set 1 = JumpingJack's settings + Default Drivers Settings ( matched with JJs ) @ 1280x960
Set 2 = Real-life gaming settings, Full Details ( All settings maxxed along with 4xAA 8xAF and Transp. Anti-Aliasing ) + Modified Drivers Settings ( see attached pics ) @ 1280x960

CPU Frequency Tests - JJ's Settings - CPU Multiplier adjustment method

Complimentary VGA Overclock Chart

As you can see the Snow test is GPU Limited at these settings.
The Cave test is totally CPU Limited.

Memory Performance Chart - "Normal RAM" vs "Tweaked RAM" - JJ's Settings

As expected, the results in the first test ( Snow ) remain totally unchanged since it's totally bound to our GPU's performance, while the performance in the more CPU/System Limited Cave test increased.

Real-Life Gaming, Full Game Details, Drivers set for max image quality, 1280x960 4xAA 8xAF Transp. AA On
CPU Frequency Tests - Full Game Details, Drivers Max IQ, 1280x960 4xAA 8xAF Transp AA On - CPU Multiplier adjustment method


Memory Performance Chart - "Normal RAM" vs "Tweaked RAM" - Full Detail Settings, Drivers IQ, 1280x960 4xAA 8xAF Transp AA Settings


The game is purely GPU Limited with these settings like I told ya

[JumpingJack]

The game is purely GPU Limited with these settings like I told ya

Nice data... as it should be... the GPU is the determinant when playing at 'contemporary' resolutions with the eye candy up.

DX10 data is nice to see for this to compare.

[IvanAndreevich]

How about sticking a lowly X2 into the 790i to see if this anomaly still persists?

[thebluemeanie1]

Maybe you could redo this test but with the Intel CPU at 1600 FSB.

[Daveburt714]

How about sticking a lowly X2 into the 790i to see if this anomaly still persists?

Pretty sure you can't put an X2 in a 790i...

[JumpingJack]

Maybe you could redo this test but with the Intel CPU at 1600 FSB.

Harder to do... would require a complete re-vamp of the test. at 400 Mhz, 2.4 or 2.6 is the closes I could get to 2.5... but, what I could do (and did) was go to 200 Mhz on the system clock -- use a multi of 12.5 and try again... within a few FPS, no change ... Snow at 1280x1024 came to 106, Cave was about 87 ... so overall that is not a huge modulator.

[Jakko]

The game is purely GPU Limited with these settings like I told ya

At both JJ's settings, and your settings, the snow level is gpu limited.
I don't think your settings are "real life settings" though, anyone would turn up the eye candy, but not at the cost of having smooth gameplay.

It is weird how the phenom system performs better in some gpu limited scenarios.
This means that the same issue could arise on a quad crossfire system that uses settings closer to your high quality settings.

[JohnMike]

The game is purely GPU Limited with these settings like I told ya

Excellent work!

But most of the games nowadays is GPU limited. Enable all settings and playing with high-resolutions, CPU power becames useless. Maybe bandwith or latency (HTT or FSB) does make small diference.

In this scenario GPU limited, having an Intel or AMD CPU doesn't matter in game performance.

[BenchZowner]

I don't think your settings are "real life settings" though, anyone would turn up the eye candy, but not at the cost of having smooth gameplay.

1) Most people don't like playing games without AA & AF at all ( especially in resolutions lower than 1920x1440 ).
2) I buy high end graphics cards to play games, so yes, I like max details @ game, and highest AA/AF possible ( adjusting depending on performance ).
3) Coming in a bit...another chart.

It is weird how the phenom system performs better in some gpu limited scenarios.

Since it's only 1 game, and 1 test, and JumpingJack himself is puzzled and can't say something about it, and he is unsure, as I am as well, it's not right to say something especially considering it as a fact.

[Jakko]

1) Most people don't like playing games without AA & AF at all ( especially in resolutions lower than 1920x1440 ).
2) I buy high end graphics cards to play games, so yes, I like max details @ game, and highest AA/AF possible ( adjusting depending on performance ).
3) Coming in a bit...another chart.

True, but most people also don't like to play anything with lower fps than 30.

Since it's only 1 game, and 1 test, and JumpingJack himself is puzzled and can't say something about it, and he is unsure, as I am as well, it's not right to say something especially considering it as a fact.

Phenom performing better in some gpu limited cases, is a fact. What did I say that wasn't right?
Also it wasn't one test, it was several tests at several resolutions.

[BenchZowner]

True, but most people also don't like to play anything with lower fps than 30.

I've got SLI for this game ( if you're talking about the low FPS @ my charts ).
I also have better single cards.
Even if I couldn't make it run smooth with any of those ( even if one of those cards that's faster than any current dual-GPU card or SLI/CF combination ) I could easily drop the AA to 2x and game fine

I wouldn't drop my details however.
Would you play Crysis @ Low ?
It looks like FarCry...maybe even worse

Phenom performing better in some gpu limited cases, is a fact. What did I say that wasn't right?
Also it wasn't one test, it was several tests at several resolutions.

Where ?
And please do not tell me about low to medium game settings without AA/AF, because most people do not play games at low, and surely not me.

Hold on now please, more charts coming soon.

[BenchZowner]

Here's my last piece for now.
Let's see how our FSB, memory, and cores count affect our gameplay quality with "gaming" settings ( Full Details, + AA/AF + Transp. AA )


As you can clearly see, Lost Planet uses about 2 cores.
Having a single-core CPU will result in lower FPS in some occasions ( not all the time ).
Lower FSB or higher FSB doesn't make a difference ( add that in the lower FSB tests here we also had the memory running at a lower frequency than in the higher FSB tests ) [ 200MHz FSB & RAM @ DDR3-800 8-7-6-21 vs 333MHz FSB & RAM @ DDR3-1333 8-7-6-21 ]
Our memory's performance also does pretty much nothing in real-life gaming settings.
Same settings DDR3-800 8-7-6-21 relaxed sub-timings vs DDR3-2000 7-7-6-21 fully tweaked = exact same performance.

So a Quad-Core Core 2 45nm @ 4GHz ( 333MHz FSB ) & RAM @ DDR3-1333 7-7-6-21 fully tweaked ram has the same performance with a Dual-Core Core 2 45nm @ 2GHz & RAM @ DDR3-800 8-7-6-21 relaxed sub-timings...

[Devil's Prophet]

And please do not tell me about low to medium game settings without AA/AF, because most people do not play games at low, and surely not me.

I do, because my Intel GMA900 is too weak to play at anything better than lowest settings. At least in more recent games like HL2. I mostly enjoy my oldies at max res/settings, but's because they are REALLY old.

Maybe next year I'll be doing some gaming on my main rig again instead on my weak but very cool laptop.

[jas420221]

I do, because my Intel GMA900 is too weak to play at anything better than lowest settings. At least in more recent games like HL2. I mostly enjoy my oldies at max res/settings, but's because they are REALLY old.

Maybe next year I'll be doing some gaming on my main rig again instead on my weak but very cool laptop.

Yes...you would not be in "most".

[JumpingJack]

Here's my last piece for now.
Let's see how our FSB, memory, and cores count affect our gameplay quality with "gaming" settings ( Full Details, + AA/AF + Transp. AA )

As you can clearly see, Lost Planet uses about 2 cores.
Having a single-core CPU will result in lower FPS in some occasions ( not all the time ).
Lower FSB or higher FSB doesn't make a difference ( add that in the lower FSB tests here we also had the memory running at a lower frequency than in the higher FSB tests ) [ 200MHz FSB & RAM @ DDR3-800 8-7-6-21 vs 333MHz FSB & RAM @ DDR3-1333 8-7-6-21 ]
Our memory's performance also does pretty much nothing in real-life gaming settings.
Same settings DDR3-800 8-7-6-21 relaxed sub-timings vs DDR3-2000 7-7-6-21 fully tweaked = exact same performance.

So a Quad-Core Core 2 45nm @ 4GHz ( 333MHz FSB ) & RAM @ DDR3-1333 7-7-6-21 fully tweaked ram has the same performance with a Dual-Core Core 2 45nm @ 2GHz & RAM @ DDR3-800 8-7-6-21 relaxed sub-timings...

Could you repeat this with 640x480 and low details, no fSAA.... to see how the core count scales when the full responsibility of the load is on the CPU. I will do the same for DX9 under XP.