Nehalem performance

[perdomot]

Wanted to ask if anyone has seen any benchies on Nehalem performance, particularly compared to Penryn and AMD.

 

[Tempered81]

compared to Penryn I'll bet a 2.66ghz Bloomfield = 3.6ghz Yorkfield - or 4.1ghz Phenom.

 

[perdomot]

That's one hell of a jump in performance. Aside from encoding times, where else can you expect to see the power used?

 

[Idontcare]

Originally posted by: jaredpace
compared to Penryn I'll bet a 2.66ghz Bloomfield = 3.6ghz Yorkfield - or 4.1ghz Phenom.

Which apps are you specifically thinking of when casting this ranking? (serious question)

Games, multi-threaded, single-threaded, winrar, sandra?

There are so many segments of applications on the desktop that I think it is prudent to specify.

Personally I have no idea, no expectation. All I know is the last rumor I heard was that we should expect 10% clock-for-clock improvement in single-threaded apps and 20-50% clock-for-clock improvement for multi-threaded apps. (relative to Yorkfield)

 

[Hugh H]

Here are a few benches for you:

Anandtech

Tom's Hardware

 

[Idontcare]

Originally posted by: Hugh H
Here are a few benches for you:

Anandtech

Tom's Hardware

And I bet those scores are below what the retail systems will punch-out given the platform maturity learning curve that was still going on back then (June and July).

For anandtech's tests the mem controller wasn't fully functional IIRC.

 

[minmaster]

is it going to run hotter than the Q9450 as it seems to consume more power?

 

[Tempered81]

Originally posted by: Idontcare

Originally posted by: jaredpace
compared to Penryn I'll bet a 2.66ghz Bloomfield = 3.6ghz Yorkfield - or 4.1ghz Phenom.

Which apps are you specifically thinking of when casting this ranking? (serious question)

something like divx/xvid, quake4, cod4. Real life benches, not synthetics, or bandwidth tests.

as hugh pointed out - in divx encoding yorkfield is performing at ~ 30% slower than bloomfield. 140/202 = 70% clock for clock. Assuming their performance scales linearly, 2.66ghz/3.6ghz = 73%. Just a rough estimate

 

[Idontcare]

Originally posted by: jaredpace
something like divx/xvid, quake4, cod4. Real life benches, not synthetics, or bandwidth tests.

as hugh pointed out - in divx encoding yorkfield is performing at ~ 30% slower than bloomfield. 140/202 = 70% clock for clock. Assuming their performance scales linearly, 2.66ghz/3.6ghz = 73%. Just a rough estimate

Cool, I like the numbers. Thanks.

Originally posted by: jaredpace
Assuming their performance scales linearly, 2.66ghz/3.6ghz = 73%. Just a rough estimate

And for a Bloomfield system with tri-channel DDR3 pumping an IMC feeding an 8MB L3$...if that doesn't produce one of the more linear scaling performance curves with clockspeed then I don't want to reach into my checking account to buy the system that would.

About the only thing that could be missing from a bloomfield x58 system is an integrated Areca controller card with 6x Raid-0 SSD's and 4GB controller cache. (who doesn't have $5k to drop on their hard-drive array?)

 

[Fox5]

Nehalem is going to shine for anything multithreaded (especially past the 2 threads point) when compared to Core 2 Duos/Quads. I think single/dual threaded performance is going to be pretty close.

Why? Well, Nehalem gives up L2 cache in exchange for that IMC, and I bet that L2 cache is still faster than system memory with the IMC. Since the core 2s have more L2 cache and share it among two cores, then for single and dual threaded benchmarks, I think they'll compare very well. Past that, they'll be hitting system memory more and more (not to mention the advantage of hyperthreading in nehalem) and lose way more performance.

 

[Idontcare]

http://www.xtremesystems.org/f...3215736&postcount=1115

 

[z1ggy]

Originally posted by: minmaster
is it going to run hotter than the Q9450 as it seems to consume more power?

I hear they with consume about 10% less power.