Anandtech benches Nehalem

[duploxxx]

All tests run with high latency and singlechannel? :x

as if this matters most in nehalem and k10 architectures..... btw none of these benches are high memory bandwidth related. cache latency on nehalem is very good and that is exactly what k10 is missing, the 3th memory channel will only be added value in server platform.

same for all the tests, scores are very good but then again only high tests that in old p4 days ran very well with hyperthreading

btw you already changed your statement from nehalem will consume less/equal against current york to better performance/watt....

[Jimmer411]

Why would they do that? That would be like testing a high performance V8 with 7 spark plugs removed.

Seeing as most software isnt multi threaded yet possibly? Im interested in seeing some single threaded performance also

[Shintai]

as if this matters most in nehalem and k10 architectures..... btw none of these benches are high memory bandwidth related. cache latency on nehalem is very good and that is exactly what k10 is missing, the 3th memory channel will only be added value in server platform.

same for all the tests, scores are very good but then again only high tests that in old p4 days ran very well with hyperthreading

btw you already changed your statement from nehalem will consume less/equal against current york to better performance/watt....

Did you read the article? Also what part is the hot thing currently? CPU or PCIe NB?

Plus everything is still a good 3 months atleast from release.

[ceevee]

Looks great , however Anand said:

Note that Nehalem continues to be faster than even the fastest Penryns available today, despite the lower clock speed of this early sample.

This is not true. I ran some of the benches he lists just seconds ago and was able to beat his Nehalem scores with a QX9770 set at stock 3.2Ghz. (not by much though, although I am stuck atm to also run my RAM way below stock settings due to MB issue)

[Shintai]

Maybe he talked about release frequency. Perhaps 3.33Ghz.

[bowman]

I didn't convert to Intel in 2006 or later - I haven't even tried a Core 2 platform, Conroe or Penryn.

What do you guys think it'll be like going from a 90nm AMD Athlon to this?

[Sr7]

I didn't convert to Intel in 2006 or later - I haven't even tried a Core 2 platform, Conroe or Penryn.

What do you guys think it'll be like going from a 90nm AMD Athlon to this?

Should be pretty damn fast, but at a certain point (maybe starting with Nehalem) the performance gains pretty much become imperceptible for everyday use..

I.e. you can tell the difference between 2 seconds and .5 seconds easily

Then things go from .5 to .25 seconds.. okay you notice.

Then things go from .25 to .125.. hmm less likely to notice.

Soon enough it's imperceptible, and you don't have enough workload to justify/exhibit the 2x performance gains anymore.

[ceevee]

Maybe he talked about release frequency. Perhaps 3.33Ghz.

clock for clock Nehalem is going to be huge

what I took issue with was the statement that this ES was faster than any Penryn today.

Quick example:

Nehalem result vs. stock qx9770 result (with terrible RAM settings)

[duploxxx]

clock for clock Nehalem is going to be huge

what I took issue with was the statement that this ES was faster than any Penryn today.

Quick example:

Nehalem result vs. stock qx9770 result (with terrible RAM settings)

this is not a clock/clock comparison.

let them shut down SMT and see if it is clock/clock faster and for cinebench with a mem controller it doesn't matter how good or bad the settings are it is cache related where nehalem puts a nice score.

[gallag]

this is not a clock/clock comparison.

let them shut down SMT and see if it is clock/clock faster and for cinebench with a mem controller it doesn't matter how good or bad the settings are it is cache related where nehalem puts a nice score.

Hurting? 2.66 VS 2.66 is not clock for clock?

[duploxxx]

Hurting? 2.66 VS 2.66 is not clock for clock?

4 physical against 4 physical + 4 virtual in benches where smt is actually working....

thx but i prefer real world unless proven that this time SMT is actually good for daily use.

[Shintai]

Duploxxx is on stage one of the kübler-ross model

[ceevee]

If you do the Cinebench math based on my stock result then 2.66Ghz Nehalem = 2.91Ghz Penryn. If you do it based on Anand's result then 2.66Ghz Nehalem = 3.29Ghz Penryn. So maybe this is why he said it was faster than any stock Penryn?

However it begs the question what the hell kind of screwed up mess did he use to get those Penryn benchmarks. They are terrible but he obviously knows what he is doing, so I could only think he lowballed the Penryn results on purpose to make the story more sensational and dramatic.

[bowman]

4 physical against 4 physical + 4 virtual in benches where smt is actually working....

thx but i prefer real world unless proven that this time SMT is actually good for daily use.

Hyperthreading was less useful back then when every desktop CPU was single core, nearly all apps were written for that, and the architecture was gimped to begin with..

This is a vastly improved Core 2 with SMT, and it'll debut in a market where even every single low to mid-end laptop and desktop PC has a dual core CPU.. So I'm thinking it'll see much more use.

[s0lid]

wants to see ocing!
just hopes these lga1366 chips and mobos won't be too expensive

[ceevee]

wants to see ocing!
just hopes these lga1366 chips and mobos won't be too expensive

I wonder if the memory controller will introduce another dreaded cold bug like the old AMD chips had.

[Syn.]

This makes me wonder, now that they finally implemented everything AMD invented so far what will they do next?

Intel is like Apple they take loads of good ideas, combine them together and then they do them right.

Performance leadership: Intel > AMD
Innovation leadership: AMD > Intel

Them two working together makes things so much better, if only Intel wasnt so touchy about market share.

[gallag]

4 physical against 4 physical + 4 virtual in benches where smt is actually working....

thx but i prefer real world unless proven that this time SMT is actually good for daily use.

Sorry man, I just don't see the point in not allowing advantages. Would you want to disable sse4 as well? I mean if they have put smt on there cpu's to make them faster then why b1tch about it?

[Origin_Unknown]

i see people saying the end is nigh for amd....
well guess what? thats bad for us if its true - with amd gone why would intel be bothered improving? i dont see any other competetor around...

[informal]

If you do the Cinebench math based on my stock result then 2.66Ghz Nehalem = 2.91Ghz Penryn. If you do it based on Anand's result then 2.66Ghz Nehalem = 3.29Ghz Penryn. So maybe this is why he said it was faster than any stock Penryn?

However it begs the question what the hell kind of screwed up mess did he use to get those Penryn benchmarks. They are terrible but he obviously knows what he is doing, so I could only think he lowballed the Penryn results on purpose to make the story more sensational and dramatic.

Hmm interesting point there.Could be the case,but we need to see the actual independent confirmation(or negation) with the retail production units.At least we now roughly see what kind of numbers Nehalem produces@2.66Ghz,so users with Penryn can compare and see for themselves,as you did.

[gallag]

This makes me wonder, now that they finally implemented everything AMD invented so far what will they do next?

Intel is like Apple they take loads of good ideas, combine them together and then they do them right.

Performance leadership: Intel > AMD
Innovation leadership: AMD > Intel

Them two working together makes things so much better, if only Intel wasnt so touchy about market share.

Man are you for real. Did Intel not use an imc first anyway.

Looks like there is going to be a lot of sore fanboys about here before the nehalem dust settles .

[duploxxx]

Sorry man, I just don't see the point in not allowing advantages. Would you want to disable sse4 as well? I mean if they have put smt on there cpu's to make them faster then why b1tch about it?

what did you do? joined the IT world in 2006? in most cases you were better of shutting down hyperthreading performance wise... and in worst cases it was even mandatory unless you wanted locked databases.

Duploxxx is on stage one of the kübler-ross model

sure keep on posting.... and most of it keep on dreaming, i am confident that i will work on this new architecture before you can even pre-order them, hence in your case it will always be dreaming and posting.

[M.Beier]

I hope Mr.Anand was sleepy...

CPU / Everest Ultimate 4.50 Memory Read Memory Write Memory Copy Memory Latency
Nehalem (2.93GHz) 13.1 GB/s 12.7 GB/s 12.0 GB/s 46.9 ns
Core 2 Extreme QX9650 - Penryn - (3.00GHz) 7.6 GB/s 7.1 GB/s 6.9 GB/s 66.7 ns

That would really really suck... I want 50GB/s read.... So he should have got 25GB/s before me being satisfied..

[gallag]

what did you do? joined the IT world in 2006? in most cases you were better of shutting down hyperthreading performance wise... and in worst cases it was even mandatory unless you wanted locked databases.

I don't follow you, You said that it was not a clock for clock comparison and when asked for your reasons why, you went back it time to the pentium4 architecture. Whats up with that?

[Shintai]

I hope Mr.Anand was sleepy...

CPU / Everest Ultimate 4.50 Memory Read Memory Write Memory Copy Memory Latency
Nehalem (2.93GHz) 13.1 GB/s 12.7 GB/s 12.0 GB/s 46.9 ns
Core 2 Extreme QX9650 - Penryn - (3.00GHz) 7.6 GB/s 7.1 GB/s 6.9 GB/s 66.7 ns

That would really really suck... I want 50GB/s read.... So he should have got 25GB/s before me being satisfied..

The testingboard only have SINGLECHANNEL working. Plus higher latency than another board. But they couldnt keep the other board long enough. SO all tests are done on the faulty.