AMD Phenom II X6 1055T 2.8GHz 95W

[zanzabar]

Intel doesn't create threads, software programs do! What prevents you from running 8 threads on your super efficient thuban? You wanna eat your cake and have it too? "Real cores are better than fake cores," except when it comes to performance tests and then "fake cores" are the real deal again. Simple, i7 quad-cores can render 8 threads, well, because they can.

if u want to do something other than basic int work than real cores are better, just look at gaming benches or some floating point or 64bit int, HT s a detriment to performance. sure if u only encode video HT is great, but gaming and anything native 64bit there is no meaning full difference.

[OhNoes!]

You think it's a fee launch?Intel designed it that way and it costs die space.We have been over this 100x before. AMD's philosophy is to use more smaller cores while intel's is to use fewer big cores(one Nehalem core is 60% larger than one Shaghai core,is it 60% faster or as reality shows ~25% faster on average?).

It may not be free lunch, but you have two extra "real cores" plus about 500mhz x 6 (3GHZ ) advantage, and you're still complaining about some "fake cores."

[informal]

What 500Mhz advantage?What CPUs are you compering here? ANd where did I complain about fake cores?Where did I say fake exactly? Use more facts and stop putting words into other people's mouths.I said threads,and that will be the measuring unit when 2011 hit.

[OhNoes!]

What 500Mhz advantage?What CPUs are you compering here? ANd where did I complain about fake cores?Where did I say fake exactly? Use more facts and stop putting words into other people's mouths.I said threads,and that will be the measuring unit when 2011 hit.

No you didn't say it: An AMD exec. said it. Remember "Real men use real cores." ? I am comparing the almighty 1090t to a lowly 17 920 in overall benchmark tests conducted by many sites. Too many links have been posted for me to even bother. You haven't seen those?

[informal]

Let me quote you:

and you're still complaining about some "fake cores."

You said I was still complaining meaning you are lying... I haven't used that term in this thread.Period.I do not care who used it,I haven't.
Also "the many benchmark sites" are those that use mostly applications with poor MT optimizations(if at all) and games in their reviews?
How about techreport's word about 1090T and 930:

The scatter plot gives us a little more context and highlights another interesting matchup: that of the Phenom II X6 1090T versus the Core i7-930. While both processors perform roughly in the same ballpark in our test suite overall, the Intel chip requires relatively expensive X58 motherboards and triple-channel memory kits, while the AMD chip works happily in more affordable 890GX mobos (and even cheaper Socket AM3 offerings) with dual-channel RAM. The 1090T ends up looking somewhat more attractive as a result.

Lost Circuits tests can be found here.They also show Thuban performing between 930 and 940 in many tests.

[OhNoes!]

Let me quote you:

You said I was still complaining meaning you are lying... I haven't used that term in this thread.Period.I do not care who used it,I haven't.
Also "the many benchmark sites" are those that use mostly applications with poor MT optimizations(if at all) and games in their reviews?
How about techreport's word about 1090T and 930:

Lost Circuits tests can be found here.They also show Thuban performing between 930 and 940 in many tests.

It's de ja vu all over again. No, you don't require triple channel kits on x58. I'm currently running dual-channel and my read bandwidth is in excess of 20,000mbs/sec. I also said I wasn't addressing you with that remark; the thing is there is so much anti Intel fud out there, sometimes I try to address them in one post - when it makes sense to do so.

"Real world" is a phrase you've used though, right? So how come when it comes to thuban you want ALL benchmarks to be run where Thuban is strongest. Mainstream apps, including gaming is where it's at. You want a processor that does everything very well, not just in one or two apps that the average mainstream user has no idea about. Also, in highly MT optimized apps, budget core i7 keeps up very well with high end thuban. That is all I'm saying. By the way even with the i7 930 as reference, that's still 400 x 6 = 2.4GHZ advantage to thuban.

[informal]

Yes and in real world gaming tests Thuban rocks as well .
BTW,I was really reserved when I said it performed between 930 and 940.Just look at the LC results,Thuban manages to outperform 965 i7 and 940 i7 on many occasions,even Westmere...

[OhNoes!]

Yes and in real world gaming tests Thuban rocks as well .
BTW,I was really reserved when I said it performed between 930 and 940.Just look at the LC results,Thuban manages to outperform 965 i7 and 940 i7 on many occasions,even Westmere...

Uh oh, aren't we a bit picky? Anyway, thanks for your indulgence.

[informal]

No problem,any time mate

[savantu]

You think it's a fee lunch?Intel designed it that way and it costs die space.We have been over this 100x before. AMD's philosophy is to use more smaller cores while intel's is to use fewer big cores(one Nehalem core is 60% larger than one Shaghai core,is it 60% faster or as reality shows ~25% faster on average?).

?? Core size is irrelevant. Total die size and performance OTOH are very relevant.

How come a Shanghai takes 243mm^2 and Nehalem takes 246mm^2 ( according to Hans deVries) while Shanghai has 8MB total cache and Nehalem has 9 ??

I can also add some extra features like :
-3 channel IMC vs. 2
-HT
-more advanced power management

And in the end, they have the same die size, while the performance difference is 20-40% in favor of Nehalem at the same clockspeed.

[Manicdan]

?? Core size is irrelevant. Total die size and performance OTOH are very relevant.

how is core size not relevant? the process of the cache is vastly different and very space consuming for AMD. if they used the same method Intel did would you still be able to use "theyre the same size" argument?

EDIT: i think to better clarify, Informal was talking about the cost to add an extra thread, while you are just talking about the overall result of the chip.

[informal]

The total die size mostly depends on the L3 sram cell size and intel (currently) reigns supreme here. Core logic die area in Nehalem is vastly larger than that one in Shanghai,but still Shanghai does not have 60% less performance than comparable Nehalem chip(the advantage varies and is between 20 to 25% ,on average,depending on the workload selection). Nehalem does have 3ch. IMC which is clearly advantage for server workloads.AMD dealt with this advantage when they transitioned to Opteron 6000 series.

[richierich]

"Real world" is a phrase you've used though, right? So how come when it comes to thuban you want ALL benchmarks to be run where Thuban is strongest. Mainstream apps, including gaming is where it's at. You want a processor that does everything very well, not just in one or two apps that the average mainstream user has no idea about. Also, in highly MT optimized apps, budget core i7 keeps up very well with high end thuban. That is all I'm saying. By the way even with the i7 930 as reference, that's still 400 x 6 = 2.4GHZ advantage to thuban.

Budget core i7? High end Thuban? Then why is the i7-930 MSRP $289.99?

[Bobsama]

X6-1055T ($200) = i7-930 ($290) > i5-750 ($200)

[gosh]

You want a processor that does everything very well, not just in one or two apps that the average mainstream user has no idea about. Also, in highly MT optimized apps, budget core i7 keeps up very well with high end thuban.

Intel has built in strong prefetchers for data (they have them on the L1 and L2 cache). They also have development tools that help to optimize data for those prefetchers. Using that you can get a higher hitrate for data in the L1 and L2 cache if data is ordered well.
Applications that are used in performance tests or when games are only drawing vertices these prefetching comes in handy, but most real world applications don't get that much advantage from these prefetchers.

Six cores are much better than 4 cores even if Intel has some tricks (80% market share makes a lot easier to add special tricks) to make those 4 cores to work a bit harder by not making them wait for data as much as AMD does in some scenarios.

[wuttz]

Intel has built in strong prefetchers for data (they have them on the L1 and L2 cache). They also have development tools that help to optimize data for those prefetchers. Using that you can get a higher hitrate for data in the L1 and L2 cache if data is ordered well.
Applications that are used in performance tests or when games are only drawing vertices these prefetching comes in handy, but most real world applications don't get that much advantage from these prefetchers.

thank you for pointing this out.
does this mean the data here is squat diddly representative of real apps?
(how does one COPY data faster than you can READ it? )

[OhNoes!]

thank you for pointing this out.
does this mean the data here is squat diddly representative of real apps?
(how does one COPY data faster than you can READ it? )

Are you serious? Educate me, why not?

[richierich]

Um when I run everest benchmark my copy speeds are also higher than my read speeds.

[Sn0wm@n]

so if i get it straight.. amd has 10x less $$$ for R&D ... has cores with 60% less die space yet they perform only 25% slower .... and people still think amd is a cheapo cpu maker??? LOLL drug is good in some part of the world

[BeepBeep2]

thank you for pointing this out.
does this mean the data here is squat diddly representative of real apps?
(how does one COPY data faster than you can READ it? )

That's how much it can read and write at the same time combined.

[OhNoes!]

so if i get it straight.. amd has 10x less $$$ for R&D ... has cores with 60% less die space yet they perform only 25% slower .... and people still think amd is a cheapo cpu maker??? LOLL drug is good in some part of the world

I don't know what to make of this post. So let's see, AMD doubles the size of their cores = double core performance? Stop grasping at straws. If VIA gets as much sympathy as AMD does, they may overtake AMD.

[informal]

Snowman,the 60% die space refers to core logic die area(the core level stuff).Since intel has much better L3 sram cell density they can pack more cache in less space,so the total die area of Shanghai and Nehalem is similar(while one Nehalem core is massively larger than one Shanghai core :24.4mm2 Vs 15.3mm2)

[Chumbucket843]

copy is faster because multiported SRAM can read or write every half cycle. it would be 2x read speed but policies for accessing cache prevent it.

[gosh]

thank you for pointing this out.
does this mean the data here is squat diddly representative of real apps?

That times the speed for L1, L2 or L3 cache

(how does one COPY data faster than you can READ it? )

Ask lavalys, they have the code (latency for the L3 cache should be slower)

[Sn0wm@n]

Snowman,the 60% die space refers to core logic die area(the core level stuff).Since intel has much better L3 sram cell density they can pack more cache in less space,so the total die area of Shanghai and Nehalem is similar(while one Nehalem core is massively larger than one Shanghai core :24.4mm2 Vs 15.3mm2)

alright..thanks for the clarification.. but still its o a good thing that amd isnt too far away with such a difference in logic core size