Why is Core i7 920 better than Phenom 2 955

[sighQ2]

I agree ... for the price it is hard to go past.

But it isn't a better CPU.

How? Why? - cos you say so?
This is worse than your previous explanation of architecture.

i7 is not core 2 at all. It's a copy of old AMD K8 architecture, and Phenom II is also based in K8.

spintel never used an IMC before i7. and their socalled quickpath is a copy of AMD's HTT - hypertransport.

The i7 has better ipc, prefetch, and L3 latency. However, I doubt it to equal a 955 at stock clock. Often people think it's better cos the i7-975 is the fastest thing on the planet - BUT COSTS 1000$$$$. The one people buy is not that - it's the bottom of the line 920, which still costs more than Phenom II. And people don't need quads anyway.

And there is something that has never been explained about actual user experience. Owners of Phenom II all say the same thing - the real world experience is "smooth". that is not heard re i7.

I say i7 is better in some ways. It also costs a lot more for a system with cpu, ram, and mobo all costing more money.

Phenom II is lower cost for cpu, and ram, and mobo. Phenom II also offers backwards compatibility to older AM2+ mobos offering previous AMD customers an optional economical and easy drop-in AM3 cpu upgrade = job done. spintel has nothing like that.

So if you look at only throughput, i7 is better. If you look at the real world picture, some very attractive features of AMD tend to dominate - unless you just don't know anything about AMD - which is true for way too many people.

And AMD's current Socket AM3 is rolling out now, and will be until at least the end of 2010. spintel has a rep of changing sockets very often, so the upgrade path is difficult and expensive.

 

[sighQ2]

phenom II is better value until the i7 price cuts come, and i5 will be just a bit cheaper than i7 is now so AMD doesn't really need to be worried.

I think Intel is making a mistake though by using larrabee as a GPU, if they used it as a CPU they would easily own the market but then I guess the EU would stomp down on them for taking out the competition.

This is a common illusion that some people readily buy into - it's called PLAYING THE FUTURE CARD.
This is not about the future. The i5 does not exist. It is not the topic. And it is a lesser cpu anyway. Nice attempt at advertising for spintel though.

AMD is not worried btw. In the future, you will see. And they are not worried now either.

 

[randomizer]

And there is something that has never been explained about actual user experience. Owners of Phenom II all say the same thing - the real world experience is "smooth". that is not heard re i7.

I heard this about Phenom II vs Core 2 Quad, although I haven't seen anything comparing this "smoothness" factor between Phenom II and Core i7. If both Core i7 and Phenom II are equally smooth then the IMC can probably be attributed to this (since Core 2 doesn't have one), otherwise I don't know.

 

[jimmysmitty]

How? Why? - cos you say so?
This is worse than your previous explanation of architecture.

i7 is not core 2 at all. It's a copy of old AMD K8 architecture, and Phenom II is also based in K8.

spintel never used an IMC before i7. and their socalled quickpath is a copy of AMD's HTT - hypertransport.

The i7 has better ipc, prefetch, and L3 latency. However, I doubt it to equal a 955 at stock clock. Often people think it's better cos the i7-975 is the fastest thing on the planet - BUT COSTS 1000$$$$. The one people buy is not that - it's the bottom of the line 920, which still costs more than Phenom II. And people don't need quads anyway.

And there is something that has never been explained about actual user experience. Owners of Phenom II all say the same thing - the real world experience is "smooth". that is not heard re i7.

I say i7 is better in some ways. It also costs a lot more for a system with cpu, ram, and mobo all costing more money.

Phenom II is lower cost for cpu, and ram, and mobo. Phenom II also offers backwards compatibility to older AM2+ mobos offering previous AMD customers an optional economical and easy drop-in AM3 cpu upgrade = job done. spintel has nothing like that.

So if you look at only throughput, i7 is better. If you look at the real world picture, some very attractive features of AMD tend to dominate - unless you just don't know anything about AMD - which is true for way too many people.

And AMD's current Socket AM3 is rolling out now, and will be until at least the end of 2010. spintel has a rep of changing sockets very often, so the upgrade path is difficult and expensive.

No because the PH II 955 needs to be at 3.2GHz to get close to the Core i7 920 thats at 2.66GHz. And even at 2.66GHz the Core i7 still beats the 955 in most everything apart from gaming where the bottleneck becomes the GPU until you add more than 2.

And before you talk, you should actually learn bcause everything you have said is all wrong. Core i7 is in no way like K8. In fact the only similarities are the fact that they both have a IMC. But other than that the way the IMC and the intercore links work are completely different. Core i7 is the same core as a Core 2 but with IPC enhancements and a IMC. Why you think Core i7 is K8 is beyond me. Its not. In fact it still has its roots in the P III Coppermine with the cache design.

And actually yes, Intel did use a IMC before. Way before AMD bought DEC Alpha. In a 486 to be exact. Then in many other CPUs. AMD was the first to implement in it a way that it was usefull to the server market and since AMD just takes what they make in the server market and apply it to the desktop market, thats why they had a IMC on the desktop.

QPI is nothing like HTT. Yes it connects the cores together and accesses the memory really fast but if you look at the specs and how it works they are nothing alike. But of course you wouldn't know that, you jusy go based off of your own thoughts.

You doubt a Core i7 920 can match a 955 at its stock speed?

Huh. Running at 2.66GHz and nearing a 30s better time....

And in gaming for having a lower clock speed it keeps up pretty damn well.

And that never will be explained. Guess what? My experience on my Q6600 is always smooth even when I am gaming and switch between intensive apps. But guess what. Thats all in a persons preference. If they build a system it will normally be a system they are proud of and like.

Um.... didn't AMD just change the socket to AM3 which is not unable to support anything but a AM3 CPU? And a AM3 CPU is only supported by AM2+ leaving AM2 dead and as well that AM2+ mobo needs a BIOS update with the CPUs code in order to work or it wont. So in the end its the same ball game. In fact LGA775 lasted longer than AM2 and AM2+. LGA 1366 and 1156 is planned for future sockets until it becomes impossible to fit a CPU in there, say 8+ cores. AMD will be in the same market when they add more cores. In fact isnt their 6 core CPU planned to be a MCM and not a true hex core? Didn't AMD bash Intel for doing a MCM quad that was easily competing with their true quad?

Huh.

Either way you have a lot of your facts wrong here. Seriously stop drinking the kool-aide. AMD has Phenom II, which is great for sub $1K machines. But if you plan a non gaming or multicard gaming setup Core i7 is the way to go.

 

[JAYDEEJOHN]

Thats my thoughts as well. IMC design. Annand said P2 seemed smoother, and later on said the i7 had the same attibutes, so, theres 1 confirmation, in a roundabout way

 

[JAYDEEJOHN]

I totally disagree with the bottleneck of the gpu thing. If thats the case, we wouldnt see higher clocks with higher fps, even tho at stock both P2 and i7 show the same fps.
If raising the clocks on the cpu gains fps, then its cpu limited, not the other way around. And many a game shows this, and yet people claim theyre gpu limited
The much faster inner communication found in i7s help with a needed faster workload, which you find on multi card setups. If the data is streamed at a slower pace, as seen in single card setups, and not as complicated, it doesnt benefit the i7s faster inner communications. Raising the clocks will help with the rest of the data, thereby giving higher fps even on single card setups.
Need to seperate the 2, as they arent the same

 

[jimmysmitty]

The only time I don't have a "smooth" experience is when its an age old machine running newer apps. Like the PCs at my work. they are age old Pentium 4 Westmeres running very memory intensive apps. Of course it wouldn't be smooth.

But put those same apps on my current machine and its smooth as butter. The older hardware gets, the harder it is for the machine to keep up with newer software. Its why I wont install vista on my old machine even though it can easily support it.

 

[jimmysmitty]

I totally disagree with the bottleneck of the gpu thing. If thats the case, we wouldnt see higher clocks with higher fps, even tho at stock both P2 and i7 show the same fps.
If raising the clocks on the cpu gains fps, then its cpu limited, not the other way around. And many a game shows this, and yet people claim theyre gpu limited
The much faster inner communication found in i7s help with a needed faster workload, which you find on multi card setups. If the data is streamed at a slower pace, as seen in single card setups, and not as complicated, it doesnt benefit the i7s faster inner communications. Raising the clocks will help with the rest of the data, thereby giving higher fps even on single card setups.
Need to seperate the 2, as they arent the same

When a CPU of equal speed to a Core i7 920 is showing 40% less performance with the same amount of GPUs then yes it is CPU bottlenecked. TBH, unless you plan a Core i7 I don't recommend more than a 4850/4870X2 due to the fact that its useless.

 

[JAYDEEJOHN]

Same argument. Going thru the bus, as C2Ds do, we have a slower data path, even if the instructions are read and processed a lil quicker on the C2D, theres gaps, and it isnt as smooth. If I can, Ill find a nice link, its heady stuff dealing with DX11, where it shows exactly how when anything else enters in, your syncronisation gets huge gaps in the out put levels. Its the nature of cpus, that do all calls, whenever.
Having an IMC, it is just too quick to detect these other calls gaps, as theyre quicker as well, but visually, we can detet them

 

[JAYDEEJOHN]

Wrong. A single i7, when oceed gives higher fps with a single gpu. Its NOT GPU limited, not at all, just because the P2 has the same fps. It doesnt mena a thing
Heres what Im addressing
"No because the PH II 955 needs to be at 3.2GHz to get close to the Core i7 920 thats at 2.66GHz. And even at 2.66GHz the Core i7 still beats the 955 in most everything apart from gaming where the bottleneck becomes the GPU until you add more than 2. "
Simply not true. If you ocee either of those cpus, in most cases, the fps goes up, showing cpu limitations
Just look at benchmarks of games in low res. At 12x10, the i7 and P2 are closer together in alot of games using a single card. Why is that? Dont say its gpu limited

 

[sighQ2]

No because the PH II 955 needs to be at 3.2GHz to get close to the Core i7 920 thats at 2.66GHz. And even at 2.66GHz the Core i7 still beats the 955 in most everything apart from gaming where the bottleneck becomes the GPU until you add more than 2.

And before you talk, you should actually learn bcause everything you have said is all wrong. Core i7 is in no way like K8. In fact the only similarities are the fact that they both have a IMC. But other than that the way the IMC and the intercore links work are completely different. Core i7 is the same core as a Core 2 but with IPC enhancements and a IMC. Why you think Core i7 is K8 is beyond me. Its not. In fact it still has its roots in the P III Coppermine with the cache design.

And actually yes, Intel did use a IMC before. Way before AMD bought DEC Alpha. In a 486 to be exact. Then in many other CPUs. AMD was the first to implement in it a way that it was usefull to the server market and since AMD just takes what they make in the server market and apply it to the desktop market, thats why they had a IMC on the desktop.

QPI is nothing like HTT. Yes it connects the cores together and accesses the memory really fast but if you look at the specs and how it works they are nothing alike. But of course you wouldn't know that, you jusy go based off of your own thoughts.

You doubt a Core i7 920 can match a 955 at its stock speed?

http://media.bestofmicro.com/3/F/212667/original/image015.png

Huh. Running at 2.66GHz and nearing a 30s better time....

http://media.bestofmicro.com/3/G/212668/original/image016.png

And in gaming for having a lower clock speed it keeps up pretty damn well.

And that never will be explained. Guess what? My experience on my Q6600 is always smooth even when I am gaming and switch between intensive apps. But guess what. Thats all in a persons preference. If they build a system it will normally be a system they are proud of and like.

Um.... didn't AMD just change the socket to AM3 which is not unable to support anything but a AM3 CPU? And a AM3 CPU is only supported by AM2+ leaving AM2 dead and as well that AM2+ mobo needs a BIOS update with the CPUs code in order to work or it wont. So in the end its the same ball game. In fact LGA775 lasted longer than AM2 and AM2+. LGA 1366 and 1156 is planned for future sockets until it becomes impossible to fit a CPU in there, say 8+ cores. AMD will be in the same market when they add more cores. In fact isnt their 6 core CPU planned to be a MCM and not a true hex core? Didn't AMD bash Intel for doing a MCM quad that was easily competing with their true quad?

Huh.

Either way you have a lot of your facts wrong here. Seriously stop drinking the kool-aide. AMD has Phenom II, which is great for sub $1K machines. But if you plan a non gaming or multicard gaming setup Core i7 is the way to go.

You obviously do not know AMD. And I similarly could care less about anything spintel. So we are the same. You have your preference and I have mine.

Your indo re spintel is well practiced. You info re AMD is not quite so good.

No problem. If I want to run iTunes really fast, I will get an i7. If I wanna have fun gaming, I know what to do.

sigh
.

 

[jimmysmitty]

You obviously do not know AMD. And I similarly could care less about anything spintel. So we are the same. You have your preference and I have mine.

Your indo re spintel is well practiced. You info re AMD is not quite so good.

No problem. If I want to run iTunes really fast, I will get an i7. If I wanna have fun gaming, I know what to do.

sigh
.

Actually in a comparison of released info, Intel tends to release more than AMD. Intel has set roadmaps that they have been meeting for the past 4+ years while AMD has changed theirs quite a bit. So its true that its hard to truly know AMD as well as Intel with a lack if information.

I actually only have one true preference with PCs and that is with my GPUs and that will always be ATI. In fact I plan on getting a 4870 soon because it will increase performance and their drivers are much more stable. Only thing I care out CPU wise is when it comes to my new systems and price range performance which is always $1500 in which each time I build a machine tends to be Intel as it would be right now with a Core i7 920.

What I posted was just an example. You can pull up the review and see that only in gaming does the Core i7 running at 2.66GHz not beat the 955 but rather keeps up with it. You said the Core i7 920 at stock could not keep up wuth a 955 that runs at 3.2GHz which was fals as was a lot of your post. I just proves you wrong, which you don't like.

And you say you don't care about Intel, which shows with your lack of knowledge on Core i7s design and how QPI vs HTT works, but still you always have to show hate against them with a bias towards AMD. I try not to as its pointless in the end. In the case of the OP, he was asking why a Core i7 is faster than a Phenom II 955 even though the clock speeds are different by almost 600MHz, which simply is explained by the fact that Intels arch for Core i7 is better.

 

[JAYDEEJOHN]

I agree with alot of what jimmysmitty has to say, except one thing, turbo. Let real clocks be real clocks here. Make no claims such n such is running at 2.66 or whatever, when no one knows how fast its actually running unless turbo is off. With it on, is it on for the whole duration? Part of it? Only the highest demanding points? 1 core? 2 cores? Like I said, disable SMT and turbo, run them head to head, and youll truly see what any serial/single threaded app perf will be
That way, theres a true IPC showing, as no 2 i7s will run alike, even in like systems running the same apps

 

[jimmysmitty]

I agree with alot of what jimmysmitty has to say, except one thing, turbo. Let real clocks be real clocks here. Make no claims such n such is running at 2.66 or whatever, when no one knows how fast its actually running unless turbo is off. With it on, is it on for the whole duration? Part of it? Only the highest demanding points? 1 core? 2 cores? Like I said, disable SMT and turbo, run them head to head, and youll truly see what any serial/single threaded app perf will be
That way, theres a true IPC showing, as no 2 i7s will run alike, even in like systems running the same apps

True in a way but when it comes to a CPU why would you disable a feature? It would be like lowering a HTT link to match a FSB speed making it a fair comparison.

TBH, most of the apps that use a single core probably would have gotten a speed boost but from what I know most of the time the highest it will go is 2 steppings so that puts it at 2.93GHz which is still shy of 3.2GHz by about 10%. beating a CPU with a 10% clock speed advantage by 20-30% is pretty interesting when you look at the arch of each.

But a multicore app will not go more than one step up which I think is probably 2.8GHz, still 400MHz lower than 3.2GHz. Thats about 12.5% of a advantage and yet it is still beat by a nice margin. Normally when OCing a Core i7 people and sites disable Turbo so that has no effect there.

SMT is great but only for apps that truly can utilize more than 4 cores all together and is an advantage yes but its what makes Core i7 great for that area for its price. It will also help with the advent of multicore gaming. I think SMT does better this round than HT did because of Core i7s QPI. Before the FSB would get saturated trying to do two threads but now there is no longer that bottleneck. Plus the other improvements they have made since most tests show that SMT does give a considerable performance increase in every aspect.

Core i7s arch is quite amazing. Even when you OC it and turn off Turbo it still manages to outperform anything except in gaming, mainly in games where CPU performance is minimal. Even in Source based games where I play the most the benefit is not that much better than a C2Q at the same clock speed unless you have like 3/4 GPUs, which I will never have.

 

[JAYDEEJOHN]

Remember tho, as each app is firing off its data, some is MT, some is not, but later, it is again. As this happens, turbo kicks in 1 step or 2, also depending on a i7 single chip efficiency and its thermal enviroment as well, so there always going to be variations. Always.
As for SMT and turbo, people seem to never make the distinction, yet at the same time say its only 2.66, which isnt so, and its obvious which apps are MT and which ones arent, when SMT brings nothing to the table, and findings are much closer between non i7s and i7s.
My point is, people say @ 12x10, games are gpu bottlenecked, when a P2 is matching a i7 in fps, or that a i7 is 2.66 beating a 3.2 are wrong. People that dont know whats going on with each processor gets the wrong idea, and its misrepresenting a true comparison

 

[jimmysmitty]

12x10 is normally the breach between CPU and GPU bottlenecks. Normally a game becoms a CPU bottleneck at lower res where the CPU does more work wne GPU bottlenecked at higher res. Of course current gen GPUs can scale normally to 1920x1080 easily before bottlenecking with full AA and AF. After that res thats where a GPU will truly be a bottleneck and is when more GPUs are needed to increase FPS.

In that case a Core i7 does truly shine, and if you remember the THG benchmark on it with a quad CF setup the 2.66GHz Core i7 easliy kept up with a 3.2GHz Core 2 Quad and anything above the 920 flat out beat it. But with the power of the Core i7 and a quad CF setup its truly hard to say if there is a bottleneck because most games don't even push the full potential of either, be it graphics or physics really.

of course that may change in the coming year with DX11 where current gen will get pushed further than currently (yes even more than Crysis because thats just a badly optimized engine) and the next gen CPUs and GPUs will be overkill again.

At least for me what I have now is overkill since I don't really play the latest and greatest and stick to my Source based games and will. My next build will probably come with HL3 which will probably include a new version of Source hopefully built upon DX11 and Raytracing mixed with Tesselation and Rasterization. By then I hope Intel is stirring things up in the GPU market to help lower prices of GPUs to all time lows.

 

[theholylancer]

Remember tho, as each app is firing off its data, some is MT, some is not, but later, it is again. As this happens, turbo kicks in 1 step or 2, also depending on a i7 single chip efficiency and its thermal enviroment as well, so there always going to be variations. Always.
As for SMT and turbo, people seem to never make the distinction, yet at the same time say its only 2.66, which isnt so, and its obvious which apps are MT and which ones arent, when SMT brings nothing to the table, and findings are much closer between non i7s and i7s.
My point is, people say @ 12x10, games are gpu bottlenecked, when a P2 is matching a i7 in fps, or that a i7 is 2.66 beating a 3.2 are wrong. People that dont know whats going on with each processor gets the wrong idea, and its misrepresenting a true comparison

Hmm about turbo, can someone explains to me how a eVga X58 claims that if you enable it on its mobo, it would make turbo set to on for all time and that the multi will be 21 regardless of what happens (with all power saving features disabled, which happens to be true if you went with the default setup)? is this a nice little mobo hack or is that built into i7s?

I'm asking here since you seems to have lots of info regarding the turbo mode... like does the little Pentium 2 like chip inside the i7 monitors the heat output and then decides if its turbo or not, and hence the mobo just feeds it with some sort of false temp data (if thats possible, since there are heat sensors onboard), or is it s "guaranteed" no hands OC, as long as the chip it self can take it, it's on type of deal?

 

[JAYDEEJOHN]

No, most games are cpu limited at 16x10 still, with some even at 19x12.

Still not getting my point. At 12x10, games are not bottlenecked by gpus, at all, not 4890s, or 260s or even 4850s for the most part.

People saying so need to study further, or ask someones opinion they respect for clarification. 3 years ago, yes. Its what Ive been saying all along. The gpus are outpacing the cpus. This coming gen of cards, using the same old games, but using the best cpus and gpus, those gpus will now allow for higher fps at higher resolutions.
Whats ths mean? It means that gpus can ask more from the cpus, and itll also mean, that the new cpus will bottleneck more games . What was considered 3 years ago as a decent resolution was 12x10. Now its just fodder for todays cards. Next gen? even easier yet, which as anyone knows, theres a balance rquired between the cpu and the gpu, but as that balance becomes more and more tilted, it eventually will lead to limitations. Good thing is, we have increased our resolutions, thereby putting the usage on the gpu, or asking more from it.
Best way to tell if a game is truly cpu bottlenecked is to have several cards tested on the same machine, if those fps are all the same? Its the cpu. If a 4850 is only 2 fps different than a 4870, its a cpu limitation, plain and simple, and at 12x10, youre going to see alot of it too

 

[JAYDEEJOHN]

Its a hack, but also a i7 feature. Its allowing the cpu to go up 1 step at all times, while at the same time, it has to shut down the cpus power saving modes, as theyre tied in together with the variable clocking, including turbo.
So it locks turbo in a plus 1 step mode, and doesnt downclock

 

[theholylancer]

Its a hack, but also a i7 feature. Its allowing the cpu to go up 1 step at all times, while at the same time, it has to shut down the cpus power saving modes, as theyre tied in together with the variable clocking, including turbo.
So it locks turbo in a plus 1 step mode, and doesnt downclock

ok thanks for that, I was wondering what the heck they did with the thing to make it basically your multi + 1 so you can OC the cheaper stuff easier with a higher multi

I need to try some old school A64 trick like lowering the multi, put a huge divider, and then set bclk sky high and hope the uncore keeps up (or put divider there as well on the QPI and other parts) lol

 

[fazers_on_stun]

You obviously do not know AMD. And I similarly could care less about anything spintel. So we are the same. You have your preference and I have mine.

Your indo re spintel is well practiced. You info re AMD is not quite so good.

No problem. If I want to run iTunes really fast, I will get an i7. If I wanna have fun gaming, I know what to do.

sigh
.

Hey, Zooty - didn't you get banned over at AMDZone?? 😀

Seriously, dude - if you couldn't preach to that audience of fanbois zealots, what do you think your chances are here at Tom's?? "Spintel" ... LOL!

 

[randomizer]

Turbo is not always a feature you need to disable. On my motherboard it needs to be explicitly enabled. 😉

 

[theholylancer]

Turbo is not always a feature you need to disable. On my motherboard it needs to be explicitly enabled. 😉

lol on my mobo I had to disable it since currently my CPU will not be stable in Linx if I do 200*21 unless I put in 1.375 ish V into that thing, I'm quite happy at 200*20 with 1.33V lol

 

[Guest]

hi,

the new phenom is not that bad:

look here:

http://www.maxxpi.net/pages/reviews/rev.php

cu

 

[JAYDEEJOHN]

Turbo is not always a feature you need to disable. On my motherboard it needs to be explicitly enabled. 😉

And on others NOT 😱 😱 😱
http://www.sav-computer.com/les-news-f4/oc-genie-tout-savoir-en-4-diapos-t553.html
[:jaydeejohn:4] [:jaydeejohn:4] [:jaydeejohn:5]