The benefit of Intels hyperthreading is the extremely small die usage. Its essentially free.
No its not, you need a wider execution unit and much more in the front-end in order to have any meaningful performance out of it.
No its not, you need a wider execution unit and much more in the front-end in order to have any meaningful performance out of it.
Yeah, I meant L1I.I assume you meant L1I, not L1D. Where are you hearing that the icache will be partitioned per core? This is the first I've heard something like that.
You're almost undoubtedly right. I was claiming that because the way the cache is arranged on die, at least with that leaked (supposedly) Excavator photo, it's in two separate arrays, with one nestled next to each decode stage.Bulldozer/Piledriver has a 2-way set associative 64KB instruction cache like all of its predecessors going back to the K7 (and even the K6 had 2-way set associative 32KB instruction cache). The natural extension to get 96KB would be to move to a 3-way set associative instruction cache; incidentally this is what Cortex-A57 is doing too. You can't partition this into two separate 48KB caches, you can't have 1.5-way caches.
Hyperthreading adds less than 5% to the core for 10-25%. Thats essentially free.
Hyperthreading adds less than 5% to the core for 10-25%. Thats essentially free.
Not for the consumer though. They have to pay almost 50% more for the 4770K compared to the 4670K, despite there only being a 5% die area difference. That sure is easy money for Intel.
Not for the consumer though. They have to pay almost 50% more for the 4770K compared to the 4670K, despite there only being a 5% die area difference. That sure is easy money for Intel.
The dies are the same size for a 4670K & 4770K.
HT is just turned off in the 4670K.
So you are admitting they are artificially locking HT on what otherwise would be a full blown i7?
Yep. Bulldozer/Vishera's the same way, AFAIK. I think core unlocking is still around.^To be fair I think both AMD and Intel have been doing that for ages. Some i5's may not have properly-working caches or HT so they simply fuse them off and rebadge them as a lower product. Although I suspect that's only the case for a small fraction of the total chips.
AMD did the same with the market demand for their Phenom II chips... Remember the days of core and cache unlocking? Buy a Phenom II x2 and unlock it to a quad-core...
I have nothing to admit.
I am stating however it is my belief that they are the same dies, with HT fused off or whatever they do to prevent it being available to users.
Incidentally, it's not really about width in an absolute sense; Bonnell/Saltwell were quite narrow but benefited tremendously from SMT.
Not for the consumer though. They have to pay almost 50% more for the 4770K compared to the 4670K, despite there only being a 5% die area difference. That sure is easy money for Intel.
And the hassle to actually disable it when you happen to run software that runs worse when HT is on.
So you are admitting they are artificially locking HT on what otherwise would be a full blown i7?
They also fuse off 2mb of L3 cache
Similarly, AMD will further detail its high-end x86 core called Steamroller initially implemented in 28nm technology and also previously announced.
Same as when you have to disable CMT with AMD because it runs worse.
You shouldn't have to disable CMT (i.e. disable 1 core per module), just make sure that you have the scheduling fix for Win7 (comes with Win8 and above).
@NTMBK: Thanks for the link. Very interesting.
This article is about ISSCC in February 2014, by which time we should have final Kaveri silicon tested by reviewers. Hotchips was back in August, I think AMD only presented some details of the console APUs this year.Should be more information on Steamroller at Hotchips:
http://www.eetimes.com/document.asp?doc_id=1319988
I never said it fixed the scaling issue. I said it fixed the scheduling issue (as in the scheduler hotfix comes with Win8 already, it doesn't need to be applied.)
The scaling problem isn't something that can be fixed via software, it needs to be remedied via hardware, and that is Steamroller.
Do you understand that manufacturing cost =! Price, right?