I7 920 vs Phenom II 965 with an ATI 5870.(Finally!) - Page 3

Last message on previous page:

In my opinion for current games all the tested CPUs in my links are mostly the same in practice. You will get playable fps with all of them. Note though that in my second link CPUs are tested at stock speeds. At similar clockspeeds i7 is a faster CPU. Wether one feels like paying the price premium for that is their business. Phenom II has a price/performance advantage in my opinion. Especially when used at stock speeds. Though it decreases somewhat or might even disappear for overclockers.

Kettu that benchmark was done with Catalyst 9.6 right? The big improvement for Phenom II crossfire came with 9.8.

I'm sure Jaydee had a link somewhere...I'll try to find it.

http://www.pcgameshardware.com/aid [...] /Practice/

(Check out the Farcry bench)

http://www.xbitlabs.com/articles/v [...] html#sect0

Some more there. Catalyst 9.8 brought the Phenom's in line with the i7's for crossfire performance. It was never that much more powerful, for some reason ATI's drivers just worked better on the Nehalem than it did on the Phenom before 9.8.

At the time it was. It changed fast due to the flow of new tech but my old P4 system still can play TF2 maxed and get 30FPS. not as nice as my new system but still playable.

No one does play at low res anymore but it still is the only true way in a game to test a CPUs potential. It may be useless in higher res but still shows the current and future potential for a CPU.

I'm not saying Phenom II is not a good chip for future proofing for a time or even current gaming. Just that I doubt it will last as long as a Core i7 will. Thats what people went on about form AM2+. A drop in replacement, futureproofing. Seems like it only applies to certain aspects......

The very original crossfire was actually better than SLI in many ways. For one it did more than just upper half and lower half which did change performance. Second it could mix and match GPUs. You needed one CF GPU and the other could be an equal or lower GPU as it connected together via a dongle externally.

But that changed with the current CF and thus the performance did have a bit of a slide down.

And its not strange to see a GTX285 beating CF 5850s when nVidia is now pretty much the only GPU company being optimized for games. This allows for even older gen nVidia GPUs to keep up with and even outpace ATIs next gen GPUs.

ATI was there too. Now they stopped after being bought by AMD. not a great move since now they have to rely on driver optimizations which normally only bring 10-15%.

The Core i7 has the ability to feed more than two cards faster than most CPUs out there. Drivers don't make a major difference in EVERY game or for every OS/GPU. Most of the time it is GPU specific, game specific or OS specific. Normally its for the current and past gen.

My old HD2900 stopped getting major performance updates at about the 8.8 drivers. Sometimes they would be a all around update but rarely. Currently its the 4800/5800 series being updated in performance and its a per game.

9. 8 upped AMD platform based Crossfire performance, Intel is planning the same thing but its called cheating with Intel and just platform enhancements with AMD. Wont get into that though.

Still lets see 2 5870x2s in the game and see which CPU bottlenecks first on a clock per clock basis..... oh wait.... we have to do it stock since a Core i7 @ 2.66-2.8GHz keeeping up is good enough and clock per clock can't be anymore.

Man it just seems like no matter what its awlays changed to their perspective and what it should be. During the Athlon X2 days it was clock per clock. Now that Intel has the top teir its stock vs stock.

If its for a true comparison then yes. Clock per clock tells you that at this speed these CPUs do this. Of course turn Turbo off for a 100% fair comparison. In the enthusiast world we do OC.

But my main point is how certain fans of AMD tend to change their tune. When Intel became the better clock per clock and performance per watt CPU it became about value. And now its stock vs stock.

Its a one way street. I think be fair. Use dual channel for Core i7, turn off turbo and even SMT just to get a one on one IPC comparison. Then turn them all on to show what the CPU is fully capable of.

Its kinda like AMDs YouTube marketing. They take a decent AMD system price wise and then throw the balls to the wall highest end unnecessary Intel parts to make a giant price gap to make it more appealing. I know its marketing but if you put same price points then compare its different. A Core i7 920 keeps up with most Phenom IIs on a stock test. When OCed it tends to beat it (gaming).

But I guess some people don't want that to be seen.

Wow, how messed up did you get that? Intel started the clock speed nonsense when they were losing heavily on ipc. It's only now that IPC is so important to intel because AMD have such a clock speed advantage.

It's no wonder you feel the way you do when you have that point messed up so badly jimmy.

Hm. It says right here on toms that the QX6850 / QX6800 (chosen because they are the same platform, so they would perform like a Q6600 OCed to those clocks, which you know) near your 3.0ghz Phenom IIs.

http://www.tomshardware.com/charts [...] ,1407.html

And I'm not saying the Q6600 is better than the Phenom II at all, that would be silly, Phenom II is better all around. But clock for clock it's mighty close.

There was no GHz "race". I know AMD was using the + sign for their models(2800+, 3000+, 3200+, etc), but the fact of the matter is different companies use different techniques. AMD use to have a higher IPC, now Intel does. The comparison needs to be stock vs stock because there's a price difference, and people want to justify the price difference when they pay for what they get. The i7 920 costs about $90 more than the 965 BE(now with the recent price drop). The 965 BE runs at 3.4GHz. Now suppose AMD came out with a processor that also costs $290. Although it might run at a much higher clock speed(3.8-4.0GHz), who's to say the i7 will still perform better?

If there was a Core i7 at 3.4GHz, it wouldn't be $300. It would a lot more than that.

Another way to test a CPU is to remove the GPU bottleneck and then test at high resolutions and high details. That's a realworld test. I prefer that.

That might be an interesting test from the technology buff point of view but I wouldn't call it fair. Not from consumers point of view atleast. Sure, all the technology buffs are consumers but most of the consumers are not technology buffs. And most people run their computers at stock clocks.

I agree and disagree.

In my opinion if you overclock one then the other one should also be overclocked. Both should be clocked as high as they can go and then compare them. That is a fair win for the i7. But I wholeheartedly agree that all around comparison is the way to go. Only that same clock speeds should be used when stock clocks are the same or they have the same overclock limit. I know you weren't trying to make an exhaustive list but for me the most important is price/performance comparison.

Links don't work.

I wasn't offering you an opinion I was attempting to tell you that way it is.

When I see the following statement:

I interpret it this way:

Massive improvements with Crossfire and CrossfireX setups in CPU limited scenes. This tells us that AMD spent some time optimizing CPU code (and I would also think it is safe to assume that they threaded a lot of this code to take advantage of more cores as the FarCry 2 results clearly show when comparing an E6600 with a Q6600).

But any way you look at it there is a GPU bottleneck present. We see this as being quite obvious as most of the results even out with the new Catalyst 9.8 drivers. They increase performance but in doing so hit the GPU bottleneck. This is all confirmed when we look at the Grid results.

Again, you're not objective in your approach therefore I am attempting to rationalize with you.

Another thing worth mentioning is the fact that QPi link is not compatible with the PCI Express Bus. The Intel IOH (x58) actually tunnels the signal from the PCI Express Bus to though the QPi link to the CPU and the same occurs in the other direction. You can view a patent filed by Intel right here: http://www.patentstorm.us/patents/ [...] ption.html

This adds a degree of complexity to the process (and some latency in the transfer) which could account as to why (when faced with a GPU bottleneck scenario) Intel systems lag AMD systems by a few Frames per Second.

Of course when we look at other CPU intensive applications.. Intel completely dominate. Gaming (the area where there is communication between the PCI Express Bus and the QPi link) is the only area we see this phenomenon rear it's ugly head.

All that having been said. A Core i7 Processor of the same clock speed is significantly faster than a Phenom II based processor. Gaming is the only area where we see these sort of discrepancies.

Could be that there is a driver issue with catalyst beta 8.66 that is holding back the crossfired HD5870.

If you look at these numbers i7 has a lot more to offer than ~75 fps. That is when you remove the GPU bottleneck.

http://www.tomshardware.co.uk/core [...] 673-6.html

I doubt that's the reason. The numbers tom's get with a single HD4870x2 are in the ballpark. It's just that when they add another HD4870x2 they significantly increase the fps they are getting. That suggests i7 is not the bottleneck there but that the graphics sub system is. It is more likely that there is a driver issue with HD5000 series cards.

Come on, the i7 is better than Phenom II in gaming if you have a good GPU.

http://www.legionhardware.com/docu [...] id=807&p=0

Crysis warhead
i7 downclock to 2.1GHz__1920x1200 39fps
PII overclock to 3.6GHz__1920x1200 36fps

Company of heroes
i7 downclock to 2.4GHz__1920x1200__8xAA 113fps
i7 downclock to 2.1GHz__1920x1200__8xAA 108fps
PII overclock to 3.6GHz__1920x1200__8xAA 107fps

Left 4 dead
i7 downclock to 2.1GHz__1920x1200__8xAA 124fps
PII overclock to 3.6GHz__1920x1200__8xAA 120fps

Unreal Tournament 3
i7 downclock to 2.1GHz__1920x1200__8xAA 135fps
PII overclock to 3.6GHz__1920x1200__8xAA 133fps

World in Confilict
i7 downclock to 2.4GHz__1920x1200__4xAA__16xAF 55fps
i7 downclock to 2.1GHz__1920x1200__4xAA__16xAF 51fps
PII overclock to 3.6GHz__1920x1200__4xAA__16xAF 49fps

Far Cry 2
i7 downclock to 2.1GHz__1920x1200 85fps
PII overclock to 3.6GHz__1920x1200 79fps

As you can see i7 at 2.1GHz beat the PII OC at 3.6GHz. Now imagine the difference if the i7 was OC at 4.2GHz…


http://www.tomshardware.com/review [...] ,2360.html
AMD Phenom II X4 955 Black Edition (OC 3.7 GHz) with 2 x HD 4890
Intel i7-920 (OC 3.4 GHz) with 2 x HD 4870

HAWX
Phenom II 3.7 GHz __ 2x HD 4890 __ 1920x1200 very high 99fps
Intel i7 920 3.4GHz __2x HD 4870 __ 1920x1200 very high 113fps

World in Conflict
Phenom II 3.7 GHz __ 2x HD 4890 __ 1920x1200 very high 63fps
Intel i7 920 3.4GHz __2x HD 4870 __ 1920x1200 very high 88fps

far cry 2
Phenom II 3.7 GHz __ 2x HD 4890 __ 1920x1200 very high 71fps
Intel i7 920 3.4GHz __2x HD 4870 __ 1920x1200 very high 92fps

Prototype
Phenom II 3.7 GHz __ 2x HD 4890 __ 1920x1200 4xAA high 54fps
Intel i7 920 3.4GHz __2x HD 4870 __ 1920x1200 4xAA high 70fps

Fallout 3
Phenom II 3.7 GHz __ 2x HD 4890 __ 1920x1200 4xAA high 82fps
Intel i7 920 3.4GHz __2x HD 4870 __ 1920x1200 4xAA high 92fps

i7 (with slower graphic cards) shows a solid lead over the Phenom II(with faster graphic cards).

I think that I prove my point that the i7 is faster in games.

What is the graphics subsystem in those reviews? Is it comparable to HD4870x2 with catalyst 9.8?

Since you seem to hold the evidence from L4D higly, this should convince you:

http://www.tomshardware.co.uk/core [...] 673-7.html

No I hold anands benchmarks in much higher regard than THG's or those ludicrous legionhardware ones.

Now what you are not trust benchmarks from tomshardware and legionhardware?

Whose benchmarks are NOT strange. i7 @ 3ghz is the same as an i7 at 3.6ghz in those benches because of the GPU limit with those settings(1920x1200 4xAA or 8xAA).

I think there was clearly something *very* wrong with that reviewers Phenom II setup. Try using a bit of logic, if you actually believe a core2 duo at 2.33ghz is just as fast as a Phenom II @ 3.6 you have issues going beyond fanboyism.

Oh and what is noteworthy is that in that THG test i7 and pII are both at stock clocks. Not very flattering for the higher clocked pII when i7 beats and/or matches its performance.

Is there a difference between high and max other than nomenclature? You don't think 16x10 high settings and 4xAA is unreasonable do you?

Not flattering from performance point of view. But flattering from the price/performance point of view.

You are saying the 64fps for a Phenom II @ 3.6ghz and a core2 duo at 2.3ghz is because the game is gpu limited. That holds up pretty well, until you realise at the same settings a core i7 @ 3.6ghz is getting 80 fps. 16 more fps in a gpu limited game? How is that happening in a gpu limited game? Surely it must be cpu limited? But if that was the case then there would be a difference in the numbers.

It's gpu limited...but not in the case of the i7?


But wait the hilarity continues...

As if that wasn't enough, the Phenom II @ 2.2ghz was equally as fast as the i7 @ 3.2ghz at 2560x1600 resolution. Does that make the Phenom II a far superior cpu? Surely by your logic it must?


You wanna know what was up with those Fallout 3 benchmarks? The reviewer alluded to it.

This is a well known issue with Fallout 3. Even when you turn off vsync ingame it doesn't always work unless you change the .ini file manually. I know this because I had to do it myself.

Two things:

1. This discussion has been about phenom II vs i7 not i5. Atleast that's what I have been focusing on.

2. In that test they are using a single HD5850. Meaning that they are most likely bottlenecked by the graphics subsystem. But I for one have consistently been focused on situations where there are no GPU bottlenecks. Please keep that in mind in the future when you post links to reviews.

I did post you the facts. Your inability to comprehend them is frustrating. Two things I mentioned. First there is a GPU bottleneck present for the reasons I listed above. Secondly there is latency added to the communications between the PCI Express Bus and the QPi Link. I hypothesize that these two elements together are responsible for the results you're seeing which are:

1. Improved CPU code helps slower CPUs do better but doesn't help Core i7 much which indicates that the Core i7 is already hitting the maximum performance possible with that particular GPU.

2. In the event where both the Core i7 and Phenom II are bottle necked by a GPU (or two or three) you would generally expect them to perform identically but this is not the case. In the event of a GPU bottleneck the added latency on the X58/Corei7 platform which is a result of this tunneling process shows up. In these cases the AMD machine is more likely to pull a few frames ahead of the Intel machine.

It has nothing to do with Caches, Predictors etc. In all of these areas the Core i7 is far superior to the Phenom II. It simply has to do with a communications latency. READ THE PATENT. If you don't understand the patent then don't argue with someone who does.

I can prove to you that it has nothing to do with caching performance: http://www.realworldtech.com/page. [...] 015436&p=6
And I can also prove to you that it has nothing to do with Branch Predictors: http://www.realworldtech.com/page. [...] 015436&p=5

It could have something to do with uops though:

Something worth noting is that these graphs were of Core 2 vs. K8. Core i7 is very similar to Core 2 but has a few new tricks (which include QPi, HT and an IMC). The fact that we are not seeing this occur with Core 2 Quad based systems would seem to indicate that it is something that affects Core i7. We know it's not HT (because it's disabled in those tests), we know it's not the IMC (because it performs remarkably well) therefore it leaves us with one possibility (QPi). And thus you have why I brought up the Intel patent (Communications method used to tunnel information from the PCI Express bus to the QPi link and then to the CPU and back).

PS. When improving the CPU code of a driver (making it threaded or optimizing it) it will be optimized for ALL applications making use of the driver (this includes GRID). The reason ATi did not mention any performance increases for GRID is because there aren't any. The new CPU code does not have an effect on that game which means that A. The game is likely already fully threaded or optimized and/or B. The game is not GPU bottlenecked and scales as intended.

And I find it hilarious that every single one of you completely ignore every anand bench, any reason why the phenom II's do a lot better there?

@ Elmo, I couldnt care less *what* the reason is, all that matters is that there *is* a reason why the i7 regularly fails at realistic gaming resolutions.

I also couldn't care less about 1024x768 resolution 'gaming' benchmarks because nobody games at those resolutions with this hardware - yet it is consistently brought up by intel fanboys as 'proof' of the i7's superiority. Are you telling me you'd rather have a cpu that wins at low res, low settings compared to high res, high settings because it sure as hell looks to me like you would.

And then we have benchmarks like on that Fallout 3 bench where the reviewer suspected vysnc was on yet didn't actually bother to check it out properly, and you expect me or anyone else to believe those garbage results?

All I have used is plain facts and logic and you idiots simply refuse to accept it because it goes against everything you've been spoonfed from the start. First it was 'triple channel memory' then 'hyperthreading' then 'crossfire/sli' that was supposed to be the reasons for intels superiority, but every new series of benchmarks fix the issues with Phenom II so now it is the only sensible gaming cpu choice.