Intel Bloomfield 2.66 GHz: First Comprehensive Evaluation

[tkpenalty]

Guys, remember that intel removed ALL limits on overclocking (the FSB wall, etc)...

 

[Morgoth]

well lets find out the limit of quikpath

 

[cray86]

MY 6700 @ 3.6ghz does super Pi 1Mil in 15 seconds... so is my proc slow or is the nehalem fast.

Either way, this architecture is not a step back, and the new hyperthreading has shown (in initial benchmarks) to be more effective the P4 HT. The processor has very low voltage, will scale to 8 cores easily, and frankly Video Editing will take full advantage of it (which I love... plus gaming isn't going to be hurt).

 

[cdawall]

Guys, remember that intel removed ALL limits on overclocking (the FSB wall, etc)...

well not completly true now you have the limits of intels integrated memory controller remember AMD's 1st try with that on clawhammer and s754?

 

[Rash-Un-Al]

MY 6700 @ 3.6ghz does super Pi 1Mil in 15 seconds... so is my proc slow or is the nehalem fast.

Actually, when I had a GO Q6600 @ 3.6 GHz, I was regularly obtaining 1M SuperPi times in the arena of 13.8 seconds. Are you sure nothing else is sapping your CPU cycles (or that your memory isn't operating at extraordinarily slow frequencies)?

Your processor is not particularly slow, and Bloomfield is not significantly faster. However, your SuperPi times -- which seem particularly slow -- may be exacerbating your perception of Bloomfield's performance advantage.

 

[Fitseries3]

first of all... the chip is running stock speeds... wait till you see it overclock before you start to complain.

second... this is the mid range bloomefield chip. it has 4 cores and NOT the 8 core's that the high end chips will have. the 8 core 16 threaded bloomefield will not perform 2x as good on single threaded apps but in multi-threaded apps... LOOK OUT! especially if these overclock anything like the 45nm wolfdale and yorkfields. people are quick to dismiss anything new for some unknown reason. perhaps it's jealousy? IDK... i know i cant wait to get one. i'll save my monies for the big daddy 8 core chip and whatever x58 board asus comes out with that has the best features. i'll have my phase done by then too so this outta be interesting.

 

[cdawall]

first of all... the chip is running stock speeds... wait till you see it overclock before you start to complain.

second... this is the mid range bloomefield chip. it has 4 cores and NOT the 8 core's that the high end chips will have. the 8 core 16 threaded bloomefield will not perform 2x as good on single threaded apps but in multi-threaded apps... LOOK OUT! especially if these overclock anything like the 45nm wolfdale and yorkfields. people are quick to dismiss anything new for some unknown reason. perhaps it's jealousy? IDK... i know i cant wait to get one. i'll save my monies for the big daddy 8 core chip and whatever x58 board asus comes out with that has the best features. i'll have my phase done by then too so this outta be interesting.

$10 says these are coldbugged like a mofo

but i have to say look at the scores next to my AMD's with an oc i'm running higher than any released AMD phenom by 200mhz and i can't come close to it performance wise

 

[Rash-Un-Al]

first of all... the chip is running stock speeds... wait till you see it overclock before you start to complain.

This is true -- it is performing at stock speeds. But, even if/when it does overclock to Wolfdale/Yorkfield speeds of 3.8 to 4.0 GHz, it will still have the same relative speed advantage (in this case 15.3% and 3.8% faster in SuperPi and single-thread Cinebench 10, respectively). I don't think posters are dismissing it as much as they are comparing.

Overall, those who have had a chance to play with Nehalem suggest a 20% IPC average advantage over Penryn, across a variety of applications (that's a 20% increase with slightly lower power requirements). And, I was excited enough to snag Wolfdale/Yorkfield for its average 6% IPC advantage over its 65nm cousins (not to mention the significantly reduced voltage requirements).

second... this is the mid range bloomefield chip. it has 4 cores and NOT the 8 core's that the high end chips will have. the 8 core 16 threaded bloomefield will not perform 2x as good on single threaded apps but in multi-threaded apps... LOOK OUT!

Yep, expect a multithreading monster the world has never seen!

 

[cray86]

Actually, when I had a GO Q6600 @ 3.6 GHz, I was regularly obtaining 1M SuperPi times in the arena of 13.8 seconds. Are you sure nothing else is sapping your CPU cycles (or that your memory isn't operating at extraordinarily slow frequencies)?

Your processor is not particularly slow, and Bloomfield is not significantly faster. However, your SuperPi times -- which seem particularly slow -- may be exacerbating your perception of Bloomfield's performance advantage.

Yeah, I just love hardware engineering and am looking with anticipation towards the new architecture. I am no fan boy and just love good chips

As for my own performance, I had nothing else running besides uTorrent. I could try with it off but I love The Office too much

Once Sony Vegas (FINALLY) gets a 64-bit version that can scale up to 16 rendering threads, Watch out!

 

[btarunr]

MY 6700 @ 3.6ghz does super Pi 1Mil in 15 seconds... so is my proc slow or is the nehalem fast.

Considering SuperPi is single threaded and is a pure math bench, you can assume that it takes a 3.60 GHz Core architecture prorcessor to beat a 2.66 GHz Nehalem.

 

[Rash-Un-Al]

Considering SuperPi is single threaded and is a pure math bench, you can assume that it takes a 3.60 GHz Core architecture prorcessor to beat a 2.66 GHz Nehalem.

Actually, not quite, as I discovered and pointed out earlier...

A Wolfdale/Yorkfield must be operating at approximately 3.05 GHz to match that 1M SuperPi time and at approximately 2.80 GHz to match that single-thread Cinebench 10 score (according to tests run personally).

In other words, (again, as I personally tested, moments ago) a Wolfdale/Yorkield @ 2.66 GHz achieves a 1M SuperPi time of approximately 18.274 sec. (versus 15.475 sec.) and a single-thread Cinebench 10 score of approximately 2936 (versus 3048). This suggests Bloomfield is 15.3% and 3.8% faster in SuperPi and single-thread Cinebench 10, respectively.

(Note: Yorkfield is only about 6% faster than Kentsfield)

 

[Tatty_Two]

Actually, not quite, as I discovered and pointed out earlier...
(Note: Yorkfield is only about 6% faster than Kentsfield)

And that speed difference at the same clock speeds is due to the increased L2 cache and some architectural efficiency tweaks.

 

[Morgoth]

first of all... the chip is running stock speeds... wait till you see it overclock before you start to complain.

second... this is the mid range bloomefield chip. it has 4 cores and NOT the 8 core's that the high end chips will have. the 8 core 16 threaded bloomefield will not perform 2x as good on single threaded apps but in multi-threaded apps... LOOK OUT! especially if these overclock anything like the 45nm wolfdale and yorkfields. people are quick to dismiss anything new for some unknown reason. perhaps it's jealousy? IDK... i know i cant wait to get one. i'll save my monies for the big daddy 8 core chip and whatever x58 board asus comes out with that has the best features. i'll have my phase done by then too so this outta be interesting.

the 8 cores version comes around q2 2009
first 8 cores nehalem cpu wil be the server version Beckton
Bloomfield is the overclock EE version
Lynnfield is the budget posible not overclock posible like bloomfield does

 

[InnocentCriminal]

Lynnfield is the budget posible not overclock posible like bloomfield does

Translation (correct me if I'm wrong) - Lynnfield is the budget offering, possibly not overclockable, unlike Bloomfield.

 

[Morgoth]

bloomfield got Unlocked Multiplyer and not lynnfield i think

 

[mk_ln]

whoa, having an unlocked multi...does that mean that it will be like previous extreme intel CPUs? (i.e. having MSRP of $999+)

 

[lemonadesoda]

Both the benches your focusing on are single-threaded. Look at the sandra, there's your feature improvements, and huge ones at that.

WRONG. CB10 multithread scores, and the CB10 multipler, are NOT single-threaded statistics. LOL. Is school out?

idiot quikpath and ht and multi cores doest play a role in Super PI

OMG what a rude fool. Read the post. The post was based on stats based on CB10. And FYI memory speeds affect superpi, and quickpath reduces latency, improves superpi. LOL. Is school out?

 

[farlex85]

WRONG. CB10 multithread scores, and the CB10 multipler, are NOT single-threaded statistics. LOL. Is school out?


OMG what a rude fool. Read the post. The post was based on stats based on CB10. And FYI memory speeds affect superpi, and quickpath reduces latency, improves superpi. LOL. Is school out?

Ok cinebench is multi-threaded, my bad. Your math is still way off in your last post, your deriving conclusions that don't really make sense so you can bash the stuff. QPI reduces increases bandwidth, but it's still in it's early stages. Still no triple-channel also, in fact, I believe they are running in 64-bit still (not sure, but previous threads indicated this). And you can't say the others w/o looking at sandra too.

 

[lemonadesoda]

Sorry, I could not see the Sandra scores. So my comments dont include those results. If they are positive, then great!

QPI reduces increases bandwidth

I find that is actually a pretty good take on the situation.

PS. If the results are majorly handicapped for any reason, I'm sure that would have been stated up front. But lets say it was forgotten. TODAY, the results for bloomfield are luke warm if not tepid. Those are the facts.

The stats are objective. Wishful thinking about what might be achieved later down the road has no place in denying those results we've just seen.

 

[Morgoth]

lemonsoda you deserve a headschot

 

[btarunr]

What's that?

 

[InnocentCriminal]

Headshot!

 

[Wile E]

Sorry, I could not see the Sandra scores. So my comments dont include those results. If they are positive, then great!

I find that is actually a pretty good take on the situation.

PS. If the results are majorly handicapped for any reason, I'm sure that would have been stated up front. But lets say it was forgotten. TODAY, the results for bloomfield are luke warm if not tepid. Those are the facts.

The stats are objective. Wishful thinking about what might be achieved later down the road has no place in denying those results we've just seen.

I don't find the results lukewarm at all. As such, it can't be stated as a fact. I was happy to jump on the 6% increase offered by the 45nm quad over my 65nm quad. This looks even more promising.

But still, I'll hold my final judgment until I see how they OC. The on die mem controller makes me a little nervous.

 

[InnocentCriminal]

The on die mem controller makes me a little nervous.

Why d'you say that, it can only help right?

 

[Wile E]

Why d'you say that, it can only help right?

Look how it effects AMD overclocking.