Thuban 1055T/1090T Previews, info and Reviews

[Sam_oslo]

We all, including those of us on Intel-platform, should be happy that AMD is making progress. Because, a little more push, and it will be flying cheap 6-core 32nm Intel CPU all over the place, soon.

[JumpingJack]

We all, including those of us on Intel-platform, should be happy that AMD is making progress. Because, a little more push, and it will be flying cheap 6-core 32nm Intel CPU all over the place, soon.

This is the best take on it, Thuban will (aside from winning the hearts of many enthusiast) likely induce Intel to populate lower price points with 6 core Gulftown.

[radaja]

Can everyone involved just quit the Intel vs AMD in this thread? It's a preview thread of processors from AMD. Let's keep it to that.

+1,that goes for both sides.it takes two points of crazy views to keep this crap going

We all, including those of us on Intel-platform, should be happy that AMD is making progress. Because, a little more push, and it will be flying cheap 6-core 32nm Intel CPU all over the place, soon.

exactly right.AMD kicking ass is a good thing.amazing that some dont see things this way

[SEA]

Yes, indeed. HT is an architectural feature to improve utilization of the pipeline/execution units, it does not change the overall results, 4 cores vs 6 cores, why would one turn something off that ups performance in most cases?

Not sure if I understand your point. (Or you - mine?)

Yes, It adds performance. There is dedicated hardware for HT within every i7 core that IS part of extra core. So it is not just 4 cores.
For example, here pure 4 cores in cinebench 11.5 have only 80% of performance compared to case with HT on.

[mstp2009]

Not sure if I understand your point. (Or you - mine?)

Yes, It adds performance. There is dedicated hardware for HT within every i7 core that IS part of extra core. So it is not just 4 cores.
For example, here pure 4 cores in cinebench 11.5 have only 80% of performance compared to case with HT on.

His point is that the die space required to add HT is minimal. Far far less that the % increase in performance it gives for multi-threaded applications.

[Sam_oslo]

I remember Intel didn't have any plans for a cheap i7, but suddenly announced the i7 920 just 2 months before the release of i7 line-up. Some has speculated that the rumors about a competitive Deneb forced Intel to release a cheap i7. That move killed the Dened before berth, actually.

I guess AMD has learned a lesson there, and therefore kept the cards about 1055T/1090T pretty tight. That's why Intel has released 980x without any cheap locked 6-core 32nm, I mean.

Intel has to counter the 1055T/1090T now, because these early results shows these can shake the midrange marked. Intel can of course release some really good and cheap i7 (cheap 935?, and 865?), but can't (traditionally doesn't want to compete on price), and therefore will have to release a few cheap locked 32nm 6-core, hopefully, I think.

[JumpingJack]

Not sure if I understand your point. (Or you - mine?)

Yes, It adds performance. There is dedicated hardware for HT within every i7 core that IS part of extra core. So it is not just 4 cores.
For example, here pure 4 cores in cinebench 11.5 have only 80% of performance compared to case with HT on.

I see your point, you are arguing Intel is the first on the market with a true 8 core on the desktop-- AMD ,and most anyone with an objective take on the technology, would disagree

My point is that it takes 6 amd cores to get to the performance of 4 intel cores, it's really that simple. HT is an architectural feature that enables one intel core to populate and keep the dispatch stations full, yielding more instructions retired per clock or more work done per unit time.

HT is not an extra core (though it is recognized as an extra core in the OS), it is a method of interleaving two distinct contextual threads through the pipeline that would otherwise not be populated -- HT extracts parallelism at the thread level in addition to the instruction level. Most processors (AMD/Intel/IBM power) that are superscalar OoO driven are actually very inefficient. Even with a 4 issue design, Intel cannot overcome the logistical latency inherent in branch misprediction, cache misses (long memory calls), and ambiguations, as such cores, even at 'full load' spend more time idle than they spend actually performing an execution. AMD is at the same disadvantage, the actual IPC of AMD cores never approach 3 just as the actual IPC of an Intel core never approaches 4.

The fact is Intel can push more work per core than AMD can, in order for AMD to reach this level of performance, they need 6 cores since each core produces less useful work than an Intel core.

This brings up the point mentioned above, Thuban will be ~ 350 mm^2, Nehalem is on the order of 260 mm^2 (quad core) both at 45 nm. It cost AMD more to get to this level of performance, it is that simple. That makes no difference to you and me, if AMD is willing to eat the margin and charge 300 bucks for a 6 core, so be it... cost to AMD is irrelevant to the end user.

Jack

[Vozer]

I can't believe core size has become an excuse for AMD's CPUs performance.

A CPU isn't made of only cores you know. Why don't you say 758M transistors or 258 mm2 perform much worse than 731M transistors or 263 mm2?

And then HT...

HT is not an extra core (though it is recognized as an extra core in the OS)

I wonder what people might say if Intel used, say, 4-way SMT.

[bloodbanger]

Saw pre-order prices here in Belgium from 275-320 euro pricerange.Can i better order a 920 then,cause that one is 235??

[generics_user]

Saw pre-order prices here in Belgium from 275-320 euro pricerange.Can i better order a 920 then,cause that one is 235??

just wait for the reviews

i guess that turbo + more cores will havbe no problem in pushing the 1090T beyond the 920s but only reviews will tell

additionally we're going to see the first 890FX boards soon

[RaV[666]]

Saw pre-order prices here in Belgium from 275-320 euro pricerange.Can i better order a 920 then,cause that one is 235??

Its most probably pre release price gouging, unless youre talking about 1090T then 275E should be right where its at.
However 1055T and 1035T which are basically the same chip should be going for 170-200E.
The 1090T is unlocked multiplier highest binned one like i7 975EE/ 965EE.

[MrMojoZ]

Why didn't I go Thuban?

Because you've been attacking AMD in threads for months now?

[Xoulz]

I see your point, you are arguing Intel is the first on the market with a true 8 core on the desktop-- AMD ,and most anyone with an objective take on the technology, would disagree

My point is that it takes 6 amd cores to get to the performance of 4 intel cores, it's really that simple. HT is an architectural feature that enables one intel core to populate and keep the dispatch stations full, yielding more instructions retired per clock or more work done per unit time.

HT is not an extra core (though it is recognized as an extra core in the OS), it is a method of interleaving two distinct contextual threads through the pipeline that would otherwise not be populated -- HT extracts parallelism at the thread level in addition to the instruction level. Most processors (AMD/Intel/IBM power) that are superscalar OoO driven are actually very inefficient. Even with a 4 issue design, Intel cannot overcome the logistical latency inherent in branch misprediction, cache misses (long memory calls), and ambiguations, as such cores, even at 'full load' spend more time idle than they spend actually performing an execution. AMD is at the same disadvantage, the actual IPC of AMD cores never approach 3 just as the actual IPC of an Intel core never approaches 4.

The fact is Intel can push more work per core than AMD can, in order for AMD to reach this level of performance, they need 6 cores since each core produces less useful work than an Intel core.

This brings up the point mentioned above, Thuban will be ~ 350 mm^2, Nehalem is on the order of 260 mm^2 (quad core) both at 45 nm. It cost AMD more to get to this level of performance, it is that simple. That makes no difference to you and me, if AMD is willing to eat the margin and charge 300 bucks for a 6 core, so be it... cost to AMD is irrelevant to the end user.

Jack

How is cost irrelevant?

Slap this baby down on my AM3 board... off and running! It's convienant and cheap.

[OhNoes!]

Because you've been correcting mis-statements about AMD hardware by fanboys in threads for months now?

Fixed.
My statement about turbo core, whether you agree or not, sadly seems to be true, as some users in the AMD section are finding out.

Everything I've posted on XS is easily accessible by anyone (even guests), just click on my username and provide us with your proof. Unless you consider objective statements to be "attacking AMD."

[Riska]

The nda ends april 27 at 6 in the morning cet.

The first leaks on the cpu has flaws in them look at the wprime in the 6.7ghz thread it is 5.4 sek at 6.4ghz bad it is because the mobo 790fx only is running at 4 cores or the win is an old install my run was at 5.8ghz stock nb and no tweaks and it was at 4.2sek running at all 6 cores so i will think with 6.4 ghz and 4.5ghz nb and way tweaked mem 3 sek i possible

[crazydiamond]

Fixed.
My statement about turbo core, whether you agree or not, sadly seems to be true, as some users in the AMD section are finding out.

Everything I've posted on XS is easily accessible by anyone (even guests), just click on my username and provide us with your proof. Unless you consider objective statements to be "attacking AMD."

the super-hero who goes around & "corrects mis-statements about amd hardware by fanboys in threads"

ahhh weve heard all about you

so anyways what is this 'mis-statements' youre referring to from this thread?

also what statements about turbo core 'sadly seem to be true, as some users in amd section are finding out'?
what weve found out is you get 500mhz free boost when the cpu is using 3 threads or less, and even more boost when you up htt. people running w/ 3.8ghz 6 core w/ 4.4ghz turbo. please explain to all of us why you think this is a bad thing again. and also what users are finding what out? your statements are not very clear.

[Manicdan]

Fixed.
My statement about turbo core, whether you agree or not, sadly seems to be true, as some users in the AMD section are finding out.

which thing?

[Mechromancer]

6am Central European Time = 11pm Central Time (USA). I'll stay up a little late tonight it seems .

[informal]

Never mind ohnoes, his 1st account got banned 4 months ago,now he is back to "correct" and "educate",again.

[Hornet331]

How is cost irrelevant?

Slap this baby down on my AM3 board... off and running! It's convienant and cheap.

Read his statemenat again, he said the production costs of a X6 is irrelevant to to consumer, since the consumer only pays the retail price.

[SEA]

HT is not an extra core (though it is recognized as an extra core in the OS), it is a method of interleaving two distinct contextual threads through the pipeline that would otherwise not be populated -- HT extracts parallelism at the thread level in addition to the instruction level
[..]
Jack

Sorry Jack, I still have to disagree in one thing. Programmers have to program the application to be explicitly parallel to utilize HT. Single threaded apps won't benefit from HT at all. Neither can you expect that some your busiest thread would do more.
So it is exactly like additional core: you have to do your program parallel. HT does not add any performance to one thread. At all.
I would consider a core with HT like some 1.x core. (say 1.25 of one core)

[Hornet331]

^
you just have described what HT is... its a way to increase the utilisation of one core (execution units). It possible to have as many SMT threads on one core as the developer of the cpu thinks it will do good. Best example is niagra from sun, with the ultrasparc T2 they have 8 threads per core and they gona double that with ultrasparc 3 to 16 threads per core. Just that it possible to execute more then one thread per core (simultaneously) doesn't make it a 1.xx core...

[MrMojoZ]

Fixed.
My statement about turbo core, whether you agree or not, sadly seems to be true, as some users in the AMD section are finding out.

Everything I've posted on XS is easily accessible by anyone (even guests), just click on my username and provide us with your proof. Unless you consider objective statements to be "attacking AMD."

Your posts read like FUD and you target AMD threads, whether you agree or not, sadly seems to be true. I cite the quoted post as an example.

[ajaidev]

Sorry Jack, I still have to disagree in one thing. Programmers have to program the application to be explicitly parallel to utilize HT. Single threaded apps won't benefit from HT at all. Neither can you expect that some your busiest thread would do more.
So it is exactly like additional core: you have to do your program parallel. HT does not add any performance to one thread. At all.
I would consider a core with HT like some 1.x core. (say 1.25 of one core)

Well not quite, multi threads app's will use multiple threads but can also use same resources at same time, in this case HT will bring down the SOE.

HT is good for multi threads app's that dont use the same resources at one time. If you take a example of three AMD cores "deneb class" vs two INTEL cores "nehalem class" the SOE and performance is real world multi thread app's is very close.

http://www.anandtech.com/bench/Product/118?vs=83

[tbone8ty]

at the egg

1055t $209.99

1090t $309.99