Why is Core i7 920 better than Phenom 2 955

[verndewd]

IPC if it hasnt been said already. Intel has 3 amd has 2 which is 30%? faster clock for clock. Longer instruction pipe does math faster. AMD used to rule at fpu processing but with the added instruction per clock it basically wails on amd bad.

If AMD added 1 IPC they be stomping Intel into the ground right now. IMO And on a severely dated process tech to boot :lol:

Clock speed: Given in megahertz (MHz) or higher (GHz), the clock speed determines how many instructions per second the processor can execute. Also called clock rate, the speed at which a microprocessor executes instructions. Every computer contains an internal clock that regulates the rate at which instructions are executed and synchronizes all the various computer components. The CPU requires a fixed number of clock ticks (or clock cycles) to execute each instruction. The faster the clock, the more instructions the CPU can execute per second.
Clock speeds are expressed in megahertz (MHz), 1 MHz being equal to 1 million cycles per second.
1 MHz = 1 million clock cycle
1 GHz = 1 billion clock cycle

http://www.escotal.com/cpu.html

need to keep in mind that AMD isnt using ANY next gen metals to increase electron ballstics, its still polysilicon silicon oxide and germanium. For the speeds and processing power they are getting with P2 thats saying alot about the design.

 

[keithlm]

Think about what you write before you get all excited again please.

Actually I'm not the one getting bent out of shape because somebody is pointing out obvious discrepancies. That would be you.

 

[JAYDEEJOHN]

Its also a lil to do with cache/memory sub systems in i7. i7 handles multi card better, single, not so much, or as much. CS,W7 and DX11 could/may help level the playing field here, or, it could even widen the gap, time will tell.
I agree, IPC helps, but if it were only IPC, wed still see huge gaps in perf using single card, in non bottlenecked by gpu games, which we dont. I also agree seeing the speeds/thermals from AMDs latest is somewhat amazing, and am looking forwards to their implementation of HKMG, even their first gen go round should/could be impressive

 

[Pailin]

Wow. You picked an article that shows 3dMark06 and a bunch of game that are already bottlenecked at both x8 and x16 for your "proof".

What you need to do is focus on something that moves a LOT of textures. Oh wait... your article shows one: The MS FlightSim. If I look at that then I discover that I was WRONG. That shows a 17%-21% difference between x8/x8 and x16/x16. So my 3%-5% number was a too low.

That showed that on bottlenecked games it won't matter much between x8 and x16. But on some things it could matter by as much as a 20%. (My personal experience moving from x16/x8 to x8/x8 was a only 6% difference in 3dMark Vantage GPU scores. Other scores varied in results. Ergo my conservative 3%-5% figures.)

Thanks for showing me that my numbers were WAY too conservative.

Actually I'm not the one getting bent out of shape because somebody is pointing out obvious discrepancies. That would be you.

Oh dear, you Really don't know when to stop digging do you


Your whole response had an overly sarcastic tone, and SHOUTING to emphasis your points while showing very inaccurate information to highlight your points - points that anyone would see are wrong... as if having to access system RAM wouldn't slow down FPS... am surprised it was only by 3.4fps tbh

You have sadly degenerated to now just plainly snapping back with more plainly wrong responses in relation to your own posts - inciting unnecessary responses and lowering the tone of this thread, please grow up or go away...

as I said before we are trying to constructively build our knowledge together and have fun - not get at each other like some playground argument


Sorry I felt I had to say this guys... am quite done with this now.

 

[keithlm]

Oh dear, you Really don't know when to stop digging do you


Your whole response had an overly sarcastic tone, and SHOUTING to emphasis your points while showing very inaccurate information to highlight your points - points that anyone would see are wrong... as if having to access system RAM wouldn't slow down FPS... am surprised it was only by 3.4fps tbh

You have sadly degenerated to now just plainly snapping back with more plainly wrong responses in relation to your own posts - inciting unnecessary responses and lowering the tone of this thread, please grow up or go away...

as I said before we are trying to constructively build our knowledge together and have fun - not get at each other like some playground argument


Sorry I felt I had to say this guys... am quite done with this now.

Your "having to access system RAM" argument doesn't actually pertain to the the issue being discussed. The issue is that changing the speed from x8 to x16 causes a difference in performance in many applications. In the link you posted this change is seen for the MS Flightsim on the same system with the same memory subsystem where the only variable being changed is the PCIe speed. Since that is the only variable being changed we can easily ascertain that the change in performance has nothing to do with anything other than PCIe speed.

Whether or not the video card has 512M or 1G of ram is also not important. In fact using the lower amount of memory will only reveal the differences sooner if massive amounts of textures are being moved. It is TRUE that more GPU ram is better; but not relevant to a discussion of x8/x16 differences.

My goal was to show discrepancies in the links. I did so. If you choose to ignore the discrepancies that is your choice. You pretend to advocate discussion but then you get emotional because I have pointed out these discrepancies.

And you should probably take your own advice.

 

[Pailin]

that was a much more constructive response

But shows you did not read/remember all the posts...

as I said earlier:
"The extra stress put on the x8 bandwidth of additional access to system memory due to too low a card memory to hold the texture distort the result... "

you said:
"Whether or not the video card has 512M or 1G of ram is also not important. In fact using the lower amount of memory will only reveal the differences sooner if massive amounts of textures are being moved. It is TRUE that more GPU ram is better; but not relevant to a discussion of x8/x16 differences. "

And clearly what you just said is somewhat contradictory stating:
"Not important. Important. It is TRUE that more GPU ram is better; not relevant to x8/x16"

a card with inadequate Video memory will have to share main system memory, which will add a Far greater strain over the x8 bandwidth than normal use of a strong GFx card.

So is Totally relevant to a discussion of x8/x16 differences:
1) A low memory SLI card configuration will cripple the x8 in very select games like Microsoft Flight Simulator X which use Huge textures - or High res with lots of AA
2) Maybe a larger memory cards will be ok - though 1) shows there may not be so much leaway

Would be interesting to know how much extra bandwidth that shared system memory makes up !?



From searching about, most sites say you are ok on x8 with most cards in SLI until you hit a 2 4870X2or better setup, then you will be wanting X16 (with enough onboard RAM it seems you should be ok in most things)

Personally I would make sure to get a x16 MB (on both PCI Express lots) for SLi or xFire in any newer purchases for some future proofing.

 

[theholylancer]

would love to see the author of the article to raise their hand in this thread

his article is being quoted here and elsewhere and because of the very fudgy look of it is generating heated debates and flamewars across the boards...

if he is doing a follow up i hope to god that he is doing it with caution. Esp. when his first article seems to be justifying a previous claim rather than a really fair assessment.

 

[JAYDEEJOHN]

Ummm, why is it then that at 16x mode, it seems to be enough? Just a question....

 

[gamerk316]

Higher ICP + More Cores = Faster.

 

[JAYDEEJOHN]

I know theres more info moving faster on 16x, but the amount produced by the 512mb of gddr3 seems to be fine at 16x, while at 8x , it floods it. Wouldnt 1 gig gddr3 have the same result?

 

[theholylancer]

I know theres more info moving faster on 16x, but the amount produced by the 512mb of gddr3 seems to be fine at 16x, while at 8x , it floods it. Wouldnt 1 gig gddr3 have the same result?

i think the logic is this:

with enough ram (be it 1 gig, 2 gig, or 2 mb), all the work done is contained on the card, and it only needs the PCIE to comm with the cpu and what nots about instruction for it to do, not having to go to main memory for data storage for internal use.

with not enough ram, main memory is used, and then the PCIE is being acted on as if it is like a HTT or QPI (comparable, but obviously not lol) bus offering comm with main memory and thus a x8 speed is really having an effect on performance since it would be like EDO ram vs DDR3.

Think of the ram onboard the gpu like cache on a CPU, if you get a cache miss in the cpu it is likely a ~100x slower process to pull information from RAM into the cache and then use it, and now imagine that the bus in between the ram and the CPU is gimped, making that cache miss even more of an issue that causes it to be 1000x slower, then you have problems if all of your data and instructions begins to be in the RAM and not the cache since it is so small that it can't hold enough.

To solve this is to either use more cache (more ram on the gpu) so cache miss is less or use a better link like QPI or HTT rather than older design with MMU off the CPU (x16 PCIE 2.0)

 

[keithlm]

Wouldnt 1 gig gddr3 have the same result?

Yes: That is why the 512M/1G argument is not relevant to the discussion of whether x8/x8 is slower than x16/x16 and whether a serious comparative benchmark should allow one brand to have the benefit of the faster access protocol while the other uses the slower.

It doesn't matter what actually causes the performance difference. Since the only variable changed was the PCIe speed then that is what created the change in performance. As such this is a variable that should be removed if you want to do a comparative benchmark. (At least if you want it taken seriously.)

Since it is a small gain that doesn't affect all applications some people are willing to accept questionable results. (This is basically the exact same logic used awhile ago to claim that DDR2-800 is just as good as DDR2-1066 in a comparative benchmark; some people think that the results are "good enough".)

 

[JAYDEEJOHN]

So, "To solve this is to either use more cache (more ram on the gpu) so cache miss is less or use a better link like QPI or HTT rather than older design with MMU off the CPU (x16 PCIE 2.0) " means that 8x is fail using the same components. Isnt that purely a 8x restriction then? I know not in all scenarios, but in some, it is, which plays to the point, is 8x enough? What if all future games somehow start doing this? My point is, keith has a valid point also. 8x fails in this scenario, while 16x works, and thus his/my point.
You could argue in some minute way the cpu in some instances could do it all, and blame any number of lessor findings on any part of it, tho, would that be a full assessment?
The better link is obviously 16x, as we see no loss . I could just as easily say, use a better card. But that also doesnt address the issue fully. Thats why games shouldnt be used as qualifiers for cpus etc, as they vary in usage of your entire system, and not the cpu alone, as well as the gpu.
And since having more ram would "cache" out, explain why it wouldnt still flood the 8x slot? It obviously isnt doing so on the 16x. same HW being used, are the calls so slowed down then? And, if so, why? Because of the 8x slot right?

 

[theholylancer]

So, "To solve this is to either use more cache (more ram on the gpu) so cache miss is less or use a better link like QPI or HTT rather than older design with MMU off the CPU (x16 PCIE 2.0) " means that 8x is fail using the same components. Isnt that purely a 8x restriction then? I know not in all scenarios, but in some, it is, which plays to the point, is 8x enough? What if all future games somehow start doing this? My point is, keith has a valid point also. 8x fails in this scenario, while 16x works, and thus his/my point.
You could argue in some minute way the cpu in some instances could do it all, and blame any number of lessor findings on any part of it, tho, would that be a full assessment?
The better link is obviously 16x, as we see no loss . I could just as easily say, use a better card. But that also doesnt address the issue fully. Thats why games shouldnt be used as qualifiers for cpus etc, as they vary in usage of your entire system, and not the cpu alone, as well as the gpu.
And since having more ram would "cache" out, explain why it wouldnt still flood the 8x slot? It obviously isnt doing so on the 16x. same HW being used, are the calls so slowed down then? And, if so, why? Because of the 8x slot right?

we are comparing 1300ish computers, not the best of the best kind, so in that sense 8x may not fail as bad as you think, older PCIE 1.0 parts had the fasted speed of 8x and they lived, they lived happily lol



now looking at the new and improved testing with two ati cards and just cpu mobo and ram difference seems good to me, and damn that I7 is not stretching its legs at that speed lol..

 

[JAYDEEJOHN]

Notice also that they both now have 16x 16x slots as well

 

[theholylancer]

yeah, but I'll bet it's not that issue that was causing all that FUD, from what I see, its the benchmark used (and the gfx card used, differing brands that is) and maybe the memory thing since the memory BW has changed somewhat

I do wonder why the hell they didn't OC the sky out of that 920... even a C0 should be able to hit 3.6-3.8 at the least and chances are they have a or can get a D0 in it (but that may be cause for ppl flaming him for using D0 when ppl may not be getting D0 and hitting a 4 Ghz OC)

And as far as I know, 3.7 on a 955 is pretty high end already, but that one I'm not too sure....

 

[JAYDEEJOHN]

Cant remember, was Turbo off? If not, then clocks were similar, or even in favor of i7.
The data showed 5% using 8x, so thats a small part of the difference. The other was the oc on the P2, and how it was oceed, the games used, and the memory. Everyone was down on him, as if he could rearrange the boutique build any. He was stuck with what they gave him, while on the other end, they had the ultimate balance setup

 

[theholylancer]

Cant remember, was Turbo off? If not, then clocks were similar, or even in favor of i7.
The data showed 5% using 8x, so thats a small part of the difference. The other was the oc on the P2, and how it was oceed, the games used, and the memory. Everyone was down on him, as if he could rearrange the boutique build any. He was stuck with what they gave him, while on the other end, they had the ultimate balance setup

yeah about the turbo, I haven't read anything about that as well, if it is on, then the 920 would be around 3.6 and that's more like it.

I think the conclusion about if you are CF/SLI with good cards then i7 is the only way to go, and if you are on single card/ weak sli/cf then amd is better bang for buck is still a valid one.

Also, this time, only games were tested, wonder what happens if they test it with more things like rendering or some other stuff.

 

[shinku_sai]

Hahahaha!!! 5.9???? My son's computer is running 4GB RAM, Athlon X2 5000+ and an HD 4670 and gets the same score!! Your computer is badly over-rated.... LOOKS CAN BE DECEIVING my friend!!!!!!

 

[Dekasav]

5.9 is the most you can get.... and it's not hard to achieve.

 

[cjl]

Hahahaha!!! 5.9???? My son's computer is running 4GB RAM, Athlon X2 5000+ and an HD 4670 and gets the same score!! Your computer is badly over-rated.... LOOKS CAN BE DECEIVING my friend!!!!!!

5.9 is maxxed out. My i7 @4.2 + 4870 quadfire gets a 5.9 as well, but that doesn't mean that my setup is even remotely similar to your son's in performance.

 

[jimmysmitty]

and it has a slightly higher IPC (instructions per clock)

Not slightly. Core i7s IPC is much higher than Phenom II. Phenom II is slightly betetr (in some/most areas) than kentsfield and still a bit behind Yorkfield in IPC.

Core i7 can easliy double performance in some apps over a Yorkfield quad of equal clock speed.

I just want to ask the question above...Core i7 clockspeed is lower, the amount of cache is the same but why does it performs better in most apps? I wonder how, I wonder why...

For most non gaming apps a Core i7 is betetr because of its arch. Its faster in not only the cache and prefectsh design but also in the inter core and memory communications which a lot of those apps depend on more than clock speed. Also the L3 serves as a memory bank that retains all the processes so if the CPU needs to repeat them it can go to the L3 instead of RAM which is faster. As well it has a repeater that will see a continuous loop and then save the end part and keep going thus cutting off a lot of steps.

In gaming they are even in most single and dual card solutions. but when you get into the relm of multi card (three and four) solutions, Core i7 has the bandwidth to feed them more than any CPU before. Thus giving its performance boost a better margin. That could be due to QPIs super fast connection being able to throw more data at each PCIe lane.

Overall its truly hard to say but its shown in numerous tests that a Core i7 is a pretty good chip. It can do a lot of tasks in about half the time of a equally clocked Core 2 Quad and others that don't use multicore technology tend to see a 30% perofrmance boost.

Its just a damn good chip and I for one hope Intel keeps going because this just means more good chips in the future and if AMD ever gets fully (they are partially competative, ATI is more competative than them now) competative again it will mean lower prices for better performance boosts for us.

 

[JAYDEEJOHN]

Id point out, any huge difference you see between Yorkys,P2's and Kentsfields vs i7 isnt only IPC. Actually, IPC has the least impact. Its turbo and SMT for apps that can use it, thats why theres such a disparity between i7 benches, because some apps are just too serial for SMT to have any effect, and shows how close Yorkys are to i7 in IPC

 

[mcx2500]

Newegg is implementing new price cuts from AMD and selling the Phenom II 955 for $210 and in a combo with an ASUS AM3/AM2+ for $225. If you sell off the ASUS motherboard for around $50-60 then your end price for a new AMD 955 can be as low at $140-150!!!

Now THAT is an unbeatable bargain.

 

[JAYDEEJOHN]

And whats the future upgrade path for i7? Hex cores? Out of a 1000 people, count on 1 hand the amount of people needing a hex core.
The 965s are coming, thats why theres a price drop on all the P2s, new king in town, taling top price, everything else moves down