Core i7 Reviews

[Markfw]

For those users, they only need an E7200 or E8400, they don't need Nehalem, and the budget versions of it will be coming WAY before there are no more C2D's

jag87, I just don't see your point. Todays launch was high end, and it is the best. If you want low end, get C2D or wait....

 

[NoStateofMind]

Sorry guys just got in. Thanks Mark for updating the OP.

 

[Zap]

Originally posted by: MarcVenice
I wish gurud3 had included a qx9770.

Me too. Maybe I'll do my own benchies, and post 'em here. I finally have a decent (hopefully) x58 board at my disposal with a Core i7 965, and I have a 790i with a QX9770 also at my disposal. If I don't get anything else added to my plate at work (fat chance), I might be able to start benching Wednesday with single, dual and tri SLI GTX 280 in various 3Dmocks and maybe some stuff like COH.

Originally posted by: zsdersw
Pure gamers should be putting their money into GPUs, not CPUs.

What if you already have as many of the most expensive GPUs your system can fit... and still have budget left?

Originally posted by: Ocguy31
If the damn software could catch up to the hardware, (games) I would be all over this!

Well, therein lies the problem. Software needs to "catch up" to hardware by taking advantage of new features, but in reality software "catches up" to hardware by bloat-fitting into the extra performance.

 

[aigomorla]

wow... then when i post my gainestown build, ima get slaughtered on this forum.

I decided to 1 up i7, and pair them.

:X

 

[Nemesis 1]

All in all IC7 looks really good . Really good fot Intel.

1). IC7 in servers is clearly a monster.

2). Because of the TRI SLI results we know there is hidden gaming performance in this chip. Goodthing others tested with 790. SO. IC7 when next gen. gpu's arrive will set themselves apart from the rest.

3) Intel has successfully limiited AMDs Deneb to Less than $284. By pricing the mid level IC7 920 right were they wanted it.

4) O/C is great on these things. Easy and 4ghz on early sand. Is outstanding.

I would imagine that Intel will do as they have with Merom / Penryn . SO a great stepping is coming. Maybe! But I still wouldn't buy this system till I know more about the Hydra Chip. Witch allows Time to work out other bugs anyway. So the Smart buyer should wait till 1st qt. 09. Hard to believe that Penryn is now a low end chip isn't it?
I bet a few are having a hard time dealing with the facts. Hell I read 1 post somewhere that 1 guy who is $$$ conservative. He is also not an early adapter. But its a strange dude. You have to be strange to talk down Nehalem when you claimed to own C2D dual

C2D quad. Merom C2D Penryn thats 3 intel cpus in 2 years. I bet because He has Had C2D for awhile his M/B was big $$$$$. Its more than likely the 2nd M/B .
Yet he preaches cost of Nehalem to high. LOL! I have seen this from alot of people.

Hay keep your penryns there great. But not all have Penryn now. Nehalem is loooking good to these people . + Others. You have to admit guys. Its a little intriging. After the reviews. To have one big question Mark remaining. Thats gaming. If the hydra chip is as good as advertized. 4 ati 4000 series cards should bring all other cpus to their knees . way befor Nehalem runs out of Poop. So until we have a Hydra M/B with 4 mid range cards installed . Running Crysis and RC2. The gaming question has yet to be decided.

Got to love comparing Nehalem to a P4P. Only in the fact that Nehalem brought no real performance boost over the Penryn. Like Nortwood to the pressies. I thought the pressies were a step backwards myself. Nehalem is performing right exactly were intel said it would. As for apps. and the efficiency they run at . Its all up to the coders to use tools wisely. Once coders optimize for sec4.2 ThIS is a great chip.

 

[sunnn]

Originally posted by: sunnn

Originally posted by: cusideabelincoln
Sunn: Read Guru3D's multiGPU tests. They provide results between the 965 and E8400 at all different resolutions and different video cards.

After reading the Legionhardware and Guru3D reviews, I must say I am impressed right now with Core i7's raw performance, and its performance for future video cards and gaming. I think Guru3D nailed the analysis by saying that games today are limited by software and 3D hardware. There is just no way Core i7 was going to deliver significant performance improvement when it is paired with any single video card, because even today's Core 2 processors are able to let a video card like an HD4870 or GTX280 deliver maximum performance at the most commonly played resolutions. Those multi-GPU results did blow my mind, and I hope everyone has a chance to look at them and read Guru3D's article.

In short: It's not Core i7's fault that it can't improve performance in today's games at today's commonly used resolutions using today's popular video card setups.

Now, off to read some more reviews!

Edit: To those worrying about overclocking: Reviews seem to show that a Core i7 @ 3.2 is plenty fast for even the 3-way SLI GTX 280s, so I imagine many people will be able to take an i7 920 up to 3.2 GHz and have a pretty damn fast system for not as costly as some of you are making it out to be, since a 3.2 GHz i7 does provide a noticeable performance improvement over a Core 2 Duo. I do wish, though, that Guru3D threw in a Q9650 for good measure.

re guru3d review: i don't know, they compared a ~$1000 monolithic-quadcore-with-imc-hypertrading-and-qpi-2nd-gen-45nm-process-with-triple-channel-memory-cpu to a ~$200-old-gen-and-dualcore-cpu and we can call it impressive? hah.

sorry quoting my own.
on guru3d still, i noticed they tossed 9770 in there:
@1024x768:FC2
965 139 fps
9770 79 fps
anandtech:
965 115.1 fps
9770 102.6 fps
something doesnt add up. how does adding gpu decrease 9770's performance? or even affect both cpu performance if its cpu bound?

 

[zsdersw]

Originally posted by: BTRY B 529th FA BN
Even AT is calling the new Nehalem a "Native" quad core: "Nehalem is Intel's first "native" quad-core design, meaning that all four cores are a part of one large, monolithic die"

i remember a while ago an argument on this subject

The argument was over terms like "true quad-core", not "native". "Native" is an appropriate way to describe certain quad-core processors, "true" is not.

 

[zsdersw]

Originally posted by: Zap
What if you already have as many of the most expensive GPUs your system can fit... and still have budget left?

Buy more games or, heaven forbid, spend the money on something not computer-related or.. even rarer still these days.. save the money.

 

[AmberClad]

Originally posted by: Zap

Originally posted by: zsdersw
Pure gamers should be putting their money into GPUs, not CPUs.

What if you already have as many of the most expensive GPUs your system can fit... and still have budget left?

A few Intel SSDs or 300gig Velociraptors?

In COH, I've run 3 vs 3 skirmishes where the other five armies are AI controlled. In TQ:IT, I'm running it with a mod that increases the size of monster mobs fivefold, and I can set the graphics and physics settings to max and still not see a significant slowdown.

What bothers me a lot more is sporadic hiccups whenever a new area is loaded, or how cumbersome it is to transfer inventory between a half dozen characters because of the time to load and unload each character from saves.

Not just this game either, but any game that has autosaving enabled (X3 and Heroes V, for example), it gets to be a drag after a while, to the point where you want to turn it off.

If you're running FSX or SupCom, then there might be a case for more CPU horsepower that only i7 can provide, over hard drive speed. And those are just my impressions based on what I play. Someone else is going to have different upgrade priorities.

 

[sunnn]

Originally posted by: JackyP

Originally posted by: sunnn

now comes the interesting part, at ~2.6Ghz, nehalem is faster by 7.3% but the improvement diminishes as the clock goes higher - only 6.2% @3.2Ghz. so its not hard to imagine that at some point, a well oclock penryn can match nehalem clock for clock. which can also be said for deneb, assuming it oclocks well.
i hope i make sense

No, you don't. I don't believe Nehalem has worse clock scaling than the other architectures, I think I've even seen tests showing the opposite.
That small difference can be explained by a. variability (there's a lot with nehalem, HT and turbo is quite dynamic and the boost may be different depending on temperature, model or other random variables - and it's a new platform after all) b. margin of error c. are you sure you're not comparing a nehalem to a 333 and 400mhz fsb penryn? (I'm too lazy to look up if you did).
Even if nehalem really loses 1% (!) clock for clock performance for a 600mhz change in clock it wouldn't necessarily lose another percent for the next 600mhz, and going by that number the penryn would need to clock very, very well, about 6800mhz to match nehalem.

You cannot base any meaningful extrapolation on such a small difference.

ok i looked up anand's gaming review again. nowhere did he mention about turbo being off. i think turbo is on by default. if turbo kicked in during the test, then at stock voltages, anand wasnt really comparing them clock for clock. there might be a good chance that they have same ipc gamewise.

 

[sunnn]

Originally posted by: Ichigo
Okay, let us ditch the subtlety of my previous post.

Check out Guru3d's multi-GPU benchmarks. Core i7 is *ridiculous* in games using Crossfire/SLI.

guru3d's numbers are questionable at best. check anand's fc2 numbers for comparison. 965 is nowhere near the almost 2x performance over 9770 its trying to project.

 

[Idontcare]

Originally posted by: zsdersw

Originally posted by: BTRY B 529th FA BN
Even AT is calling the new Nehalem a "Native" quad core: "Nehalem is Intel's first "native" quad-core design, meaning that all four cores are a part of one large, monolithic die"

i remember a while ago an argument on this subject

The argument was over terms like "true quad-core", not "native". "Native" is an appropriate way to describe certain quad-core processors, "true" is not.

And it's not like Anand is infallible - he also wrote this in the same article concerning IMC:

"Nehalem does spice things up a bit in the memory department, not only does it have an integrated memory controller (a first for an x86 Intel CPU)"

http://www.anandtech.com/cpuch...howdoc.aspx?i=3448&p=5

I was under the impression Intel's first integrated memory controller was released long ago on a 386SL chip.

So if Anand is prone to a gaffe such as this, surely he could plausibly make a gaffe here or there regarding the "native" quad-core debate.

Not slighting Anand, or the side-topic of native vs. true, just saying one article on the interwebz makes not the final word or truth or definition on any topic if you really are trying to keep an open mind about things.

 

[Denithor]

Originally posted by: Markfw900
3D rendering ? like maybe render farms ? Can you say new farm anyone ?

Hey, Mark, are we going to see a bunch of used Folding boxes on the FS forum shortly?

Originally posted by: Idontcare
In conclusion: meh.

That's pretty much my take on Nehalem as well. Intel definitely has this architecture aimed squarely at servers & taking back share in that highly profitable chunk of the market. For typical desktop users (even hardcore gamers included in this group) there's a pretty low ROI for the considerably higher cost of motherboard & memory versus the current C2Q chips.

One comment though, DDR3 has already started falling significantly. There's a 4GB DDR3-1333 G.Skill kit at newegg for about $130 (only twice the cost of equivalent DDR2 kits). If the i7 920 is actually priced around $300 at launch it will be actually cheaper than lesser-performing Q9550 processors (by about $75). The only part that would be considerably more expensive will be the x58 motherboard, which will probably cost twice what an equivalent X48 board does.

 

[Markfw]

No, I was referring to the comanies that make movies. I thought maybe one of them would be converting, since this is such a much faster chip.

My folding farm is doing fine, and if I did upgrade, it would be to get 2 more video cards.

 

[IntelUser2000]

Uhh, guys?? Hello??

Look at this: http://www.anandtech.com/cpuch...howdoc.aspx?i=3448&p=5
Then this: http://www.anandtech.com/cpuch...howdoc.aspx?i=3448&p=6

Compare Triple Channel DDR3-1600 (9-9-9-24) with Dual Channel DDR3-1066 (9-9-9-20) in actual application benchmarks.

Looks like the Dual Channel DDR3-1066 is faster in some cases than Triple Channel DDR3-1600. I'm pretty sure that the Triple Channel DDR3 might do better in higher end and higher load app benchmarks and Tri SLI.

For those of us that has their finances and heads on the ground, there's no difference between Dual and Triple channel.

On the single thread, pure per clock advantage might not be a lot. But it has definite advantages in some places and overall it comes out to be higher, even without the Turbo Mode.

Turbo Mode isn't a gimmick either. It works, and it works well. It rewards you for having a cooler system and it works with overclocking. You can raise the limit if you wish.

Why do some people hate hyperthreading so much?? Just because it only impacts in apps that you don't care about?? When did games in high settings ever benefitted from a faster processor?? Why get a faster CPU at all? It's the same BS and argument.
-Anyways Aceshardware has interesting point about Hyperthreading. In regards to cache and memory latency, the L1 and L2 cache latency increases but the L3 cache latency decreases. Which is why some games that gets out of L2 cache size manages to perform better with hyperthreading on.

Core i7 isn't Core 2+HT+IMC. Core i7 fixes the weaknesses that's existing on the Core 2 architecture. In multi-threading, your app will scale better even if it doesn't scale beyond 4 threads because the core communication is faster, it has memory controller on-die.

This iteration of Nehalem, Bloomfield is in all regards high-end platform. Sure, the 920 is cheap, but the platform itself is not.

The more sane versions will come with almost no sacrifice when Lynnfield comes next year. No sacrifice, because its gonna be clocked same and Tri channel isn't faster than Dual channel 99% of the time.

Intel introduced 3.0GHz QX9650. But then they came with Q9650, which is same thing but with cheaper price. I bet you same will happen with Nehalem.

It'll also run cooler not having extra memory channel and QPI. And QPI doesn't help on single socket systems

 

[Idontcare]

Originally posted by: PC Surgeon
Well its almost launch time and at least one review site has posted their Core i7 benchmarks.

Anandtech + Everything You Need to Know about Intel's New Architecture

Benchmark Reviews

Tweak town

Legion Hardware

Overclockers

Guru3D + Guru3D Multi-GPU Benchmarks

PC Perspective

Hexus

T Break

Trusted Reviews

Tech Gage

Toms Hardware

HardOCP

Madshrimps

Xbit

OC Aholic (Googlenglish)

Golem (Googlenglish)

IT Examiner

Planet X64

Hot Hardware

APC

Au Ja (Googlenglish)

Driver Heaven

TechSpot

Lost Circuite

Custom PC

Tech Radar

Canard PC

PC Inpact

Extreme Tech

Hartware (Googlenglish)

Matbe (Googlenglish)

Tweakers.net (Googlenglish)

Bit Tech

Revioo (Googlenglish)

Hardware.fr (Googlenglish)

Neoseeker

OCAU

HCW

Frosty Tech

CRN

B3D i7 965

B3D i7 940

B3D i7 920


Thanks harpoon84, Cheesetogo.

Added more
Markfw900 Anandtech Moderator

Sheesh! :shocked:

Can anyone point me to the "best" review out there which is going to help me understand:

(1) why Nehalem L1$ latency was increased while it's size remained the same relative to Penryn?

(2) why Nehalem 8MB shared L3$ cache runs so slow...2.66GHz...and has such piss-poor latency...compared to the nearly same size (6MB) shared L2$ on Penryn which runs faster clockspeed and MUCH lower latency?

(3) all that work to transition from domino to static CMOS and the power savings are where? I expected vastly better power consumption numbers, vastly better.

(4) Thread scaling performance results? 4 threads with affinity locked to two cores (so 2 threads run "native" and two threads run SMT), etc? Where's the data on these kinds of tests?

So perhaps I am missing the data and would appreciate a gently nudge towards the "best" review to educate myself further on these subject matters.

I read thru Anandtech and Tom's reviews. Tom's sure did a shitload of apps and processors for comparison. It left my impression of AT's review to be, er, lacking in breadth just a skosh...

 

[aigomorla]

Originally posted by: AmberClad

A few Intel SSDs or 300gig Velociraptors?

you need a 500 dollar controller for benifit of those.

I am on the later, 6 Raptors:
http://i125.photobucket.com/al...aigomorla/IMG_1145.jpg
http://i125.photobucket.com/al...aigomorla/IMG_1146.jpg

i expect to move to the former Intel SSD's.

I am gonna get myself 4 x 2.5inch so i can put them on a sata backplate like this:
http://i125.photobucket.com/al...gomorla/Harddrives.jpg

and have fun with them on my areca controller.

 

[dmens]

Originally posted by: Idontcare
Sheesh! :shocked:

Can anyone point me to the "best" review out there which is going to help me understand:

(1) why Nehalem L1$ latency was increased while it's size remained the same relative to Penryn?

you really think a review is going to give you that information? pretty sure its confidential forever.

(2) why Nehalem 8MB shared L3$ cache runs so slow...2.66GHz...and has such piss-poor latency...compared to the nearly same size (6MB) shared L2$ on Penryn which runs faster clockspeed and MUCH lower latency?

because it has to go through the L2 first and through a 4-way queue? as for the clockspeed, you can increase the UCLK ratio in the BIOS.

(3) all that work to transition from domino to static CMOS and the power savings are where? I expected vastly better power consumption numbers, vastly better.

your expectations were misplaced.

(4) Thread scaling performance results? 4 threads with affinity locked to two cores (so 2 threads run "native" and two threads run SMT), etc? Where's the data on these kinds of tests?

So perhaps I am missing the data and would appreciate a gently nudge towards the "best" review to educate myself further on these subject matters.

I read thru Anandtech and Tom's reviews. Tom's sure did a shitload of apps and processors for comparison. It left my impression of AT's review to be, er, lacking in breadth just a skosh...

yeah i agree. guess i have to wait for the server tests (which should be quite the shocker).

 

[neosapien]

I've seen benchmarks for 3-way SLI Geforce GTX 280s and quad-crossfire Radeon X2 4870s on Nehalem, but for those of us who don't have $1,100-$1,200 to spend on GPUs alone, I'd like to see benchmarks comparing these types of systems with budget SLI (god, how weird is it to put those two words together) graphics:

1- Core 2 Duo E8400, Geforce 8800GT or 9800GT 512MB SLI
2- Core 2 Duo E8400, Radeon HD 4850 512MB Crossfire
3- Core 2 Duo E8400, Radeon HD 4850 1GB Crossfire (does the 1GB memory actually improve performance on a 4850? I have no idea)
4- Core i7 920, Geforce 8800GT or 9800GT 512MB SLI
5- Core i7 920, Radeon HD 4850 512MB Crossfire
6- Core i7 920, Radeon HD 4850 1GB Crossfire

 

[Tempered81]

Originally posted by: neosapien
I've seen benchmarks for 3-way SLI Geforce GTX 280s and quad-crossfire Radeon X2 4870s on Nehalem, but for those of us who don't have $1,100-$1,200 to spend on GPUs alone, I'd like to see benchmarks comparing these types of systems with budget SLI (god, how weird is it to put those two words together) graphics:

1- Core 2 Duo E8400, Geforce 8800GT or 9800GT 512MB SLI
2- Core 2 Duo E8400, Radeon HD 4850 512MB Crossfire
3- Core 2 Duo E8400, Radeon HD 4850 1GB Crossfire (does the 1GB memory actually improve performance on a 4850? I have no idea)
4- Core i7 920, Geforce 8800GT or 9800GT 512MB SLI
5- Core i7 920, Radeon HD 4850 512MB Crossfire
6- Core i7 920, Radeon HD 4850 1GB Crossfire

Well, Anand did 4870 Crossfire and GTX260 Sli.

 

[neosapien]

Radeon HD 4870 Crossfire and Geforce GTX 260 SLI aren't exactly low budget, though. That's why I mentioned Geforce 8800/9800GT and Radeon HD 4850, you can get them for <=$230 and $300 in SLI and Crossfire, respectively. That's pretty cheap, a good bargain for the performance. The 260 and 4870 in SLI/Crossfire would cost around twice as much.

 

[Idontcare]

Originally posted by: dmens

(4) Thread scaling performance results? 4 threads with affinity locked to two cores (so 2 threads run "native" and two threads run SMT), etc? Where's the data on these kinds of tests?

So perhaps I am missing the data and would appreciate a gently nudge towards the "best" review to educate myself further on these subject matters.

I read thru Anandtech and Tom's reviews. Tom's sure did a shitload of apps and processors for comparison. It left my impression of AT's review to be, er, lacking in breadth just a skosh...

yeah i agree. guess i have to wait for the server tests (which should be quite the shocker).

Here's a test that shows a glimpse into thread scaling:
http://www.pcper.com/article.p...=634&type=expert&pid=9

And I love how they show in this graph that when it comes to thread scaling the number of memory channels really does make a difference:
http://www.pcper.com/article.p...634&type=expert&pid=12

(that's so cool!)

 

[IntelUser2000]

Looks like its not really Hyperthreading that affects multi-thread performance increase but rather the core itself. The multi-core performance scaling must be much better on Core i7 than Core 2. Increased bandwidth, better thread synchronization, L3 acting as a buffer for multi-threading.

 

[Idontcare]

Originally posted by: IntelUser2000
Looks like its not really Hyperthreading that affects multi-thread performance increase but rather the core itself. The multi-core performance scaling must be much better on Core i7 than Core 2. Increased bandwidth, better thread synchronization, L3 acting as a buffer for multi-threading.

Actually if you crunch the numbers to cast them as true "thread scaling" numbers as most folks in the field would do then you get this:

Threads.....i7-965 w/o turbo.....QX9770.......Amdahl for 81% parallelized code
....1...........1.0...........................1.0................1.0
....2...........1.68.........................1.68..............1.68
....4...........2.44.........................2.48..............2.55
....8...........3.57.........................2.44..............3.43

So what is really amazing to me is that QX9770 scales 1 to 4 threads in this application no better or worse than Nehalem scales.

4 threads on QX9770 gives you 2.48x speedup while 4 threads on the i7 965 gives you 2.44x speedup.

The extra 4 logical cores on Nehalem are quite functional though, clearly, as they yield an additional 1.13x speedup (i.e. 46% faster with 8 threads than with 4 threads) over the single-thread performance when all 4 SMT cores are put to full use.

So it is quite the opposite conclusion IMO...SMT is clearly value-add in this application but the migration from FSB to QPI+IMC+3-channel+monolithic yielded zero improvement (realtive to that of 1600MHz FSB of QX9770 and dual-channel) to the interprocessor communication speeds as far as this application is concerned.

edit: I meant to add this nugget: this manner of scaling where both processors score the same scaling despite radically differing methods of core-to-core communications is typical of serial-code limited course-grained applications.

The scaling limitation is not interprocessor communication speed but rather that only 81% of the code in this application is parallelizable (yes thats a word). I added the Amdahl scaling results to the table above.

The remaining speedup scaling discrepancy can be attributed to interprocessor communication not being infinitely fast...and would need to have Almasi and Gottlieb's modification to Amdahl's law added to scaling equation.

 

[IntelUser2000]

To Idontcare:

You do live up to the name. Go look up at Xbitlabs and Tomshardware results. Compare the i7 benchmarks against QX9770 and see the HT-off results against QX9770. Most of the benefits come from the core changes

Xbitlabs

Core i7-965 SMT off/Turbo off vs Core 2 Quad QX9770: http://www.xbitlabs.com/articl...-core-i7_10.html#sect0

There are significant advantages to be had by CPU core changes alone, average benefit from non-SMT changes and SMT changes(not including PCMark and 3DMark) shows 8% each.

Tomshardware

Core i7-965 SMT on vs Core 2 Quad QX9770:
http://www.tomshardware.com/re...7-Nehalem,2057-11.html

Core i7-965 SMT off/on comparison:
http://www.tomshardware.com/re...7-Nehalem,2057-12.html

Substantial amount of performance comes from core changes.

Cinebench shows no difference because its a real world app that doesn't reflect majority of other real world apps.

Indifferent to cache change(<1% difference), when lots of apps will be affected by it
http://www.nordichardware.com/.../?skrivelse=514&page=5