Phenom II and i7 <=> cache design for gaming

[cjl]

Really?

You don't think 3-4% is within normal variation (note: the normal variation I'm talking about is on a completely GPU limited scenario, so your clockspeed comparison is invalid). I bet if I run 10 Crysis benchmarks, it'll have more than 3-4% variation on several of them. It's perfectly normal.

I'll run several later if you want me to prove that the normal variation is at least that much.

 

[sighQ2]

I believe i7 was a dt cpu that was adapted to be a server. as for good server, all of the bogus and dubious trix that make it look good in biased benchmarketing, are disabled in any respectable IT usage. Example turbo trix - works when cores idle, single thread - as if a server is idle or single threading much. So much for turboid.

I don't see what's so convincing with the posted cjl benches above. An overheating i7 at 3.8 vs an old PhenomII 940 on ddr2 partially oclocked. You must think people are real stupid. And what's the test setup - who cares anyway.

Most of you seem real nervous about this discussion.

jsmitty - I am real tired of hearing about your eol 6600 as if it's the greatest thing - I suppose it snuffs out the i7 also. And I really like your dissertation on the glories of spintel junk followed by your comment:
= "I am not too sure how Phenoms L3 works in comparison." =
Hilarious. Thank you for that accurate information.

Nobody except jdj seems willing to consider the possibility that spintel is once again just telling you all how to think - and they are about to do it again with the latest paper launch of i5, conveniently timed to coincide with the release of Phenom II 965. How low they sink, again and again - and you support that dontcha.

I lack the expertise, and I daresay I am not alone in that, to fully address this cache question - but I admit it.

My understanding is that spintel is a cpu company and all they do is cpu - they don't do grafx. They want the cpu to do everything, cos that's how they make money. If, as jdj indicates, AMD demos that this cpu actually sux when accompanied by a hot grafx solution, then you can all invent a way to keep the world lagging behind the truth about where tek could be today if it weren't for spintel's profit plans, and antitrust scams. The truth will not be denied. And even spintel can't stop it forever - but they try - and you support that dontcha.

Anyway, the cpu can't do grafx any longer, and a cpu that can work with a gpu is going to cause some problems for the lies that seem to be preferred in this thread. That's the hi rez difference - the grafx is offloaded to the gpu. So the i7 can't control the world as you might prefer.

Yomama what are you afraid of? People having an intelligent discussion? and you scream troll, like some troll adolescent drunk spinner fan, brainwashed irreparably. Wake up and know that spintel treats you all like mushrooms. It's simply too obvious anymore.

Ever notice that intelligent people prefer AMD? How could that be so?

Anyway, this is just another case of gross dysfunction being used to subvert the healthy communication of truth and education. An obviously superior L3, hampered by spintel dominated code - nothing new there - and you support that dontcha.

Big changes coming - you all get to receive new orders, or get real quiet maybe - oh I wish.

Thx to kassler for opposing this dysfunctional messy place. here's to transcending the bottleneck scam.

 

[kassler]

At least now youre specific.
Then to really get down to it, wed need as many examples as we can find to make this a reasonable assumption, or fact, right? You have more benchmarks? That shows this disparity, even at other frequencies?

I am not looking that much for graphs. I don't think there are that much games out there now that shows this because the market for X3 or X4 cpu's with L3 cache is low. There will be few buyers if you design the game to need that type of cpu. Also DX9 and 10 don't work multithreaded. Most games will probably work well with 2 GB memory also.
I think you will see a more games this winter that will show this behavior. Some games are probably skipping X2 cpu's and move to X3 or X4 cpu's to be able to run the game. They may use more memory also and physics and AI is also evolving. DX11 is going to increase advantages to phenom too.
All this will take more use of memory to calculate the frame. More code and more data is needed.
The hit rate needs to be high on cache in order to get the best performance. A cache that doesn't evict data to soon is very important in this type of workload.

 

[JAYDEEJOHN]

I do know W7s global limits on MT is better than Vistas, which is better than xps
Also, DX11 will help in this as well.
cjl, remember tho, it was the clocks and the fps difference combined, thats why its 7%, not just 3 or 4%, but we need more data
@ sighQ2, actually, i7 IS a server chip, once IMC was done, just like K8, and K10 as well.
Its the scaling baby

 

[JAYDEEJOHN]

Just a guess here, but I suppose its possible, the more threading done thru the OS or the DX, the more it could work into the P2s favor, much like what we see with ATIs new drivers, where MT thru the driver has increased.
More calls means more access to cache. Might be possible, time wil tell

 

[cjl]

cjl, remember tho, it was the clocks and the fps difference combined, thats why its 7%, not just 3 or 4%, but we need more data

If it's completely GPU limited, as I suspect, you would not expect the clocks to make any difference in this case. That's why the relevant number is 4%, not 7%.

 

[sighQ2]

quote kassler = "All this will take more use of memory to calculate the frame. More code and more data is needed. "

Those mem accesses also add latency. If the L3 does not need to access mem, that's less latency - or even stutter.

People say Phenom II is smooth. They said that about Phenom I also.

@jdj - my info says it was originally a desktop chip, adapted for server - not sure what you mean by "IMC was done" - but you know more than I do.

I am more of a generalist; actually an impressionistic generalist. Kinda like riding a motorcycle - after all the talk, it's a feel thing, or a "seat of the pants" thing, or a "real world experience" thing. I'm big on that experiential stuff, after all the talk - that's how I know I am alive.

Also, I find it interesting that kassler is told he has to provide proof thru benchmarks or whatever. People are not similarly encumbered when spouting endless claims about the marvels of spintel junk.

Did I say junk? Yes I did. O well. Some obvious things don't require proof. Then again, maybe it has already been proven by those willing to consider the possibility - and then explored it in a room more conducive to such investigation. Perhaps I can research at such a room and present something of value here re proof. haha. Do you think such proof would be well received? Or would I be labelled troll also? I suppose there's one way to find out.

 

[kassler]

quote kassler = "All this will take more use of memory to calculate the frame. More code and more data is needed. "

Those mem accesses also add latency. If the L3 does not need to access mem, that's less latency - or even stutter.

Yes but cache is much more important and latency memory on phenom is good. What i7 is good at for memory is bandwidth. bandwidth in games is not that important. Bandwidth is more important on servers (mostly databas servers)

 

[JAYDEEJOHN]

But Crysis is both cpu and gpu limited, the skies the limit here, as well as a lil gddr limited.
I know its only 4%. M$ claims only 10% best showing. ATI claims more with DX11, but is that a combo in a perfect scenario with W7 etc?
Point is, if hes right, and he could be, the lack of i7 compared to P2 could play a role in MT moreso on the SW end, as i7 has more guts, but may get choked out, whereas P2 doesnt have as much power, but if more is actually coming in, it may handle it better, but not by alot, because its not as powerful.
Im just guessing here, thats why my earlier post on the ATI drivers.
If access is increased thru SW, and its a heavily fose of scattered data like games are, who knows?

 

[cjl]

The i7 was primarily designed as a server chip, hence the tri channel memory and massive multithreading optimizations. In addition, its best performance is actually in server benchmarks - in some cases, a 2P Nehalem server approaches a 4P Opteron server for performance for example.

As for smoothness, I've seen the same claim made about the i7. Having used a (first gen) Phenom system for a bit, as well as my current i7 as well as a Q6600 rig, and quite honestly, all of them felt smooth.

Oh, and if you want anyone to take you remotely seriously, stop calling Intel "Spintel". It makes you look biased and unintelligent.

 

[JAYDEEJOHN]

And Im not talking BW either. Functions and calls, as both cpus meet the requirements BW wise

 

[JAYDEEJOHN]

Theres cycles used to place it all in the L3, thats my other point. I dont knwo the latency differences, thats why I asked

 

[sighQ2]

spintel? I said nothing about what you said? where did you get the idea I was biased or unintelligent? Do you always assume that when people don't respect the bogus biased benchmarketing you present as factual proof of inherent superiority to the detriment of a more innovative and courageous company? Or is it just popular to mimic others who are also led astray by spintel spin? And how can I believe anything you say since your bias is made severely obvious by your constant placation. If that's offensive, you created the platform; I just dance on it. It's not very secure. I think I will sit this one out. Perhaps another will respond to your appeals.

Sorry, I must go now; I have additional plans for my busy day; maybe later.

 

[cjl]

I never said you were necessarily biased or unintelligent. I said that your use of the petty and unnecessary name for Intel makes you appear that way. Your constant statements in favor of AMD and against Intel seem to indicate a pretty strong anti-Intel bias though...

As for bias? I honestly don't have any for one company or the other, aside from whatever performs best for the money when I build. For example, I used an 8800GTX in a build for a friend a while back, but have 4870x2s in my own system. I have an i7, but I have recommended Phenom IIs and Core 2 Quads to friends in the past.

Oh, and don't tell me that Intel only leads on synthetics. Some of my matlab scripts run more than twice as fast on my system compared to my friend's Core 2 Quad, and more than 1.5x the speed of another friend's PhII (I believe it's memory bandwidth limited, as that best explains the results).

 

[JAYDEEJOHN]

cjl is not biased

 

[kassler]

Here is one test that is testing on very low resolution and then increase to higher resolutions

LOW: http://www.techspot.com/review/162-amd-phenom2-x4-955/page10.html
HIGH: http://www.techspot.com/review/162-amd-phenom2-x4-955/page11.html

 

[laughingjack]

ok now im gunna chuck this into the mix.....

yes i want smooth FPS gaming. ( anything more than about 50fps is wasted IMO. and since i live in a 50Hz world, power supply-wise, this would work nicely). was playing JointOps but looking towards ArmA2.

however, i do 3D rendering as well (lightwave3d) that relies solely on cpu throughput. i need as much calculating power and threads as i can get. so which 'i' series do i go for? the 1366 or 1156?

what i would like is this;


cpu> i7 920 (oc'ed prolly), but with a view to upgrading it.

mobo> at least x58 and preferably mini ATX form factor. this because i lug the box around all the time and id like small and light.

ram> triple channel, for the i9 upgrade path and 3Gb (6 when i go to 64bit).

slots> 1 pci-e 16x, as i really dont think a multi card setup is actually worth the frames my monitor cant show me.

+lots of USB's and sata2 (with SAS, maybe).


my current box has lasted me nearly 5 years and still going strong. P4 3.02G,1Gb, radeon9800Pro, Xp32bit.
i would like a comparable life from the new stuff if poss.

cheers!

 

[fazers_on_stun]

could it be that you ar disipointed when someone explains how the cpu works and you will understand that phenom is a good game cpu? if you have spent a lot of money on intel

"Explains"?? What I see is your cut & paste from some article, plus a whole lotta assumptions about game code structure with zero proof or other credibility, and a large leap of faith. What kinda explanation is that??

 

[JEskandari]

I am talking about crysis 1920x1200
i7 @ 3.8 vs Phenom II @ 3.64 = 0.044 or 4.4%
29.42 fps vs 30.41 = 0.034 or 3.4%

4.4% + 3.4% = 7.8%

And remember that i7 L3 cache runs faster compared to Phenom II

This is NOT within normal variation. If you put up ten other computers and run same test you will notice same behaviour

I wonder about this preview

and consider in this test phenom 955 overclocked to 3.7 and use 2 x hd4890 while core I7 overclocked to 3.44 and use 2 x hd4870

 

[fazers_on_stun]

No ones explained why P2 trashes anything Intel, period. This may answer if, in 1 game, this theory actually works .....or not heheh

Oh yes, this theory by the world-famous software scientist Kassler, who can look at a couple crude bar graphs posted here & determine precisely which frames are CPU bottlenecked & which ones are GPU bottlenecked, with his super-science X-ray vision no doubt ... 😀.

 

[JEskandari]

I wonder about this preview
http://media.bestofmicro.com/T/F/216195/original/PhII vs. i7 Crysis.png
http://media.bestofmicro.com/T/G/216196/original/PhII vs. i7 Crysis VH.png

and consider in this test phenom 955 overclocked to 3.7 and use 2 x hd4890 while core I7 overclocked to 3.44 and use 2 x hd4870

as you see when the core i 7 face the same disadvantage and even face the video disadvantage the result are same.
i say there is some problem with xrisis itself

 

[spud]

How is you math?

If you take 100 frames. 95 of those frames are bottlenecked by the gpu, 5 is bottlenecked by the cpu. do you think you will get big differences?

Im pretty sure now you have no clue on how a game engine and a gpu for that matter prep those frames. Games arent similar to a TV broadcast with a constant predictable content stream. Variables such as FOV, shader effects, mesh (poly) counts, AI, physics, shadows, light sources, sound computation, and user input to name a small fraction of whats actually being done to a frame prior to rendering. You really really need to sit down and actually read/comprehend then clearly articulate those findings, because as of this moment its quite clear you dont understand what your trying to debate.

Word, Playa.

 

[fazers_on_stun]

The problem is that you don't want to understand

Nope - the problem is that you have not provided any plausible proof - all we have is your mumbo-jumbo, arm-flapping explanation based on no evidence (another poster had to supply that for you). If you explain it any arm-flapping harder, I would not be surprised to see you start rising into the air ...

 

[kassler]

Nope - the problem is that you have not provided any plausible proof - all we have is your mumbo-jumbo, arm-flapping explanation based on no evidence (another poster had to supply that for you). If you explain it any arm-flapping harder, I would not be surprised to see you start rising into the air ...

If you know that someone is wrong, than you must know what is right?
Please explain. And explain good

 

[fazers_on_stun]

People say Phenom II is smooth. They said that about Phenom I also.

I am more of a generalist; actually an impressionistic generalist. Kinda like riding a motorcycle - after all the talk, it's a feel thing, or a "seat of the pants" thing, or a "real world experience" thing. I'm big on that experiential stuff, after all the talk - that's how I know I am alive.

Ah yes - it must be laundry day at the AMD troll swamp - they're all coming out with their dirty laundry! Run for the hills, folks -- it's Zooty the sasquatch! 😀

I guess now we'll be subjected to the smoothier argument, seeing as how the double-cheeseburger one doesn't apply to the i7...

BTW, I thought the "smoothier" argument was based on the HTT used by the K8 - K10.5. Oops, now that i7 has a QPI, well it must be something else!! Lessee now, 3 decoders vs. 4 - naw, that doesn't sound good... Howzabout cache - yeah, that's the ticket!! Let's trot out the superior cache argument, ignoring the fact that Intel's L3 is faster and all that stuff...