i7 Nehalem Processor Questions

[Neil Jones]

What size cache does the 3.2 GHz Core i7-965XE or the 2.93GHz Core i7-940 have?

Is the Nehalem processor going to have any significant advantages over a Core2Quad Q9550 BX 2.83 GHZ @ 1333 with 12MB Cache?

I use an application called Mastercam on a daily basis. Some users of Mastercam report the biggest benefit for Mastercam seems to be from a big cache and high FSB speeds.

 

[jimmysmitty]

Well Nehalem will no longer have a FSB but rathe QPI which will link it to all the devices and is much faster than the FSB is. It will have only 256K L2 cache but a 8MB L3 cache.

Core i7 should be faster than C2Q in single threaded applications by about 10% depending on the app and has been seen up to 40% in multithreaded apps. Of cours we haven't had a conclusive benchmark series yet but should within the next month or so.

It will also include SSE4.2 which is the rest of the SSE4 instruction set.

In all, Core i7 will be faster but by how much depends on the apps you use and if they take advantage of more than just one sore.

 

[Freightshaker]

64mb level 1 cache
256kb level 2 cache per core...unshared
8mb level 3 cache shared
clock for clock i7 is showing large gains verses yorkfield proccesors.
2.93 seems to out perform or stay even wt q9770.
or to put it simply a $500 to $600 Nehalem should out perform a $1400 to $1600 penryn.
what that means as far as the 2.66 entry i7 est $325.00 retail and the
q9650 I don't know?

personal feelings on the subject:
q9650 might be a sub $300 chip by JAN 09

 

[Guest]

^ bullsh*t

core i7 brings little to no improvement over yorkfield clock for clock...

in synthetic benchmarks... yes synthetic not even real world... core i7 gets maybe 20 % bonus in non heavy memory usage programs... and less than a 5 - 10 % boost in single threaded applications

as for gaming core i7 has shown in preliminary benchmarks to be equal to or WORSE than yorkfield... core i7 is not needed except for professionals who use multi-threaded apps, or programs that use a lot of memory bandwidth

your best bet is to just get a q6600 and OC the hell out of it with a xigamatek s1283 cooler

 

[jimmysmitty]

^Actually its not BS when they show 40% gains in highly multithreaded apps.

We should start seeing finalized chips and drivers that should show the actual gaming performance. Remember that some of these game tests were done without the correct drivers which could cause some problems since those drivers include the driver for the PCIe link.

But only time will tell.

And to Freight it should be 64K of L1 cache. No way they could fit 64MB in that small space.

 

[krazyk12]

thog delete your post, thx

dont listen to him, doesnt know what hes talking about.

 

[Shadow703793]

DO NOT START A FLAME WAR!

 

[Freightshaker]

ROFL yup that would be one big chip omg!!!

Your point about the drivers is truly valid.
The early benchmarks....you know the ones you can't find any more.
all of those were done using engineering samples
on X58 protos that were not ready for prime time.
Nehalem performance could be %5 to %15 better
than that.
(%76 of all statistics are made up on the spot!)


my question: what are we going to see wt Penryn prices?
when the Q9450 hit the scene Q6700 prices went under $300
now a BX80562Q6700 can be had for $255 to $269
oem for$239

 

[jimmysmitty]

I think I read Intel plans another price cut soon and then probably a few months after Nehalems release Penryn and Yorkfield will drop in price again.

Then we will see in mid to late 09 Penryn will start to be ramped down and Nehalem will be ramped up and then the 32nm shrink will be started probably early 2009 and they will do the same with releasing ES chips to the reviewers for a preliminary idea of what they are able to do then it will be out by November of 2009. by that time Penryn will probably be at Conroes price point or lower.

Then in 2010 we will see Sandy Bridge which should be up to 16 or 32 cores. They haven't released much info on it yet but if it is 16 core it would make 32 threads.

I am more interested in Harem (might be wrong on the code name). It is supposed to be a very advanced SMT allowing 4 threads per core meaning a quad would have 16 threads. That should be interesting to see.

 

[JAYDEEJOHN]

What weve seen is, the gains in single threadedapps, the vast majority of apps today, maybe 7% on average over penryn. Multi threaded, its alot more. Gaming has shown it does slightly better in a few games, even in a couple and worse in a couple. If you run multithreaded apps, its a perfect cpu. If youre a gamer, from what weve seen, theres cheaper solutions out there.

 

[Freightshaker]

I've read the X58 chipset is much less expensive than the overpriced X48 chipset
That should leave room for Mobo makers to release less expensive midrange boards?
I mean a low end X48 from ASUS, GIG or DFI can be had for $220 to $250
Shouldn't partners be able to stick a good X58 on the market for less than $250?

or should I go back to treatment for all those drugs I'm doin?

 

[jimmysmitty]

^They wont. They want to make money so they will start them off at the same price as the X38/X48s started at and thats normally $300 for the uber top end. Then you will see the lower high ends with the X58 that have less features but as long as they have 6 memory slots and 2 x16 PCIe slots its basically the same just not all the features most people will never use.

 

[JAYDEEJOHN]

The 58 boards will probably end up a little more than the 48s only because theyre 8 layers on the pcb vs 6 for the 48s, but having a cheaper chipset off sets that mostly

 

[Freightshaker]

OK now the big question:
I am waiting on my build until the I7 release
even though I probably will go ddr2 wt a penryn.

am I nutzy or what?

 

[roadrunner197069]

I'm waiting on I7 myself. I got half the money and will have the rest by the first of the month.

 

[Freightshaker]

.
.
.
..................... OK I'm not NUZTY or we are Nutzy.

 

[jimmysmitty]

Well Core i7 is going to be nice. Mainyl because on the IMC and QPI. That coupld with SMT and better IPC will result in a smoother, faster experience.

That and the fact that gaming seems to be trying to head multithreaded , if the preliminary results are tru then you will see much better gains compared to Penryn. I think one game is Lost Planet since it is multithreaded sees a big boost from Core i7 thanks to improved IPC and SMT.

From what I have heard and seen VALVes next game Left 4 Dead, which looks fantastic graphics wise for a 4+ year old game engine, might have multithreading and that would mean VALVe would update the Source engine for all of their games with this allowing better performance since Source is a very CPU limited game. In fact in TF2 there is a console command to make the game support more than one core but it is unstable. People who use it have seen anywhere from 50-100% performance improvements too.

But as I said Core i7s performance improvement over Penryn will mainly depend on how highly threaded the app/game is.

I myself am waiting for the 32nm shrink. Not sure why but it would seem the 32nm shrink will be a bit better since most of the early bugs will be taken care of, it will run cooler and they will also use a more advanced version of the HK/MG tech.

 

[JAYDEEJOHN]

I think youll be happy, and down the road even happier, once multithreaded apps start showing up. I hear they oc nicely, a few small bugs in them, but anything brand new is gonna have a few things not working optimally, but from what Ive heard, just normal stuff. New bios, things like that. To me, it looks good, just pricy, and I dont like cutting edge, the wounds heal slowly on cutting edge with me. Once you get it, youd better let us know how its working tho

 

[jimmysmitty]

Well I was just on Asus website to check out the BIOS updates for my board when I noticed in the downloads section you can select LGA1366 and their first X58 mobo will be called the P6T Deluxe. Its using the Marvell Yukon LAN, which ahs always been good and the normal Soundmax 8.1 digital onboard audio and the ICHR10 which is supposed to be better than the ICHR9 southbridge. Time will tell.

But I find it strange that thats already up but the mobo spec isnt.

 

[Stupido]

@jimmysmitty: you mentioned in the previous post that TF2 has a console command to make it use multicore... I'm playing TF2 quite a lot... I have Q6600 and I see only one core utilized 100% flat...
thanks!

 

[Sable Wanderer]

i know this is a stupid question
so Core i7's use socket 1366?
and also will 775 CPU work with a 1366 motherboard

 

[cjl]

Yes, the initial i7's will use 1366. No 775 won't work - the socket is physically different (1366 contacts vs 775).

 

[jimmysmitty]

@jimmysmitty: you mentioned in the previous post that TF2 has a console command to make it use multicore... I'm playing TF2 quite a lot... I have Q6600 and I see only one core utilized 100% flat...
thanks!

Its in all Source based games. The command in mat_queue_mode 2, the default is -1. VALVe has been working on multithreading since HL2: EP2 but they didn't finalize it yet since it is still unstable. Some people see gains some don't,

When I play TF2 my main core (core 0) will see about 30% usage and the other 4 cores will jump between 10-15% usage. The only time it hits 100% is at the main menu. VALVes Source engine uses the CPU for particle and physics (Havock for physiscs) so once they have it stableized and add the update you will see more use out of the cores.

 

[Sable Wanderer]

Yes, the initial i7's will use 1366. No 775 won't work - the socket is physically different (1366 contacts vs 775).

Thanks

 

[Stupido]

@jimmysmitty
Thanks