BenetanegiaYeah, prior to reading reviews I had no idea the chip itself had 8 cores. It is a little disappointment that they didn't enable all of them. I guess yields are not good at all and they are saving them for Xeons. Something inside me still tells me it's a little trick, in order to have something else to release down the line and charge $1000+ again. It's not like they really need the 8 cores in order to leave both the competition and their own previous generation in the dust.

Maybe it's kind of better this way. While it would have been interesting to see a direct comparison between both 2 billion transistor behemoths (SB-E vs BD), it would have been a bloodbath.

So it does have 8 cores with two masked off? I was wondering about this. It explains the enormous memory bandwidth improvement made, doesn't it?

I must look up the architecture diagram of SB-E.

sneeky blue devils....

Unlockable later .......for a price or neutered?

qubitI must look up the architecture diagram of SB-E.

Bandwidth boost is there literally because this CPU is two 2600K's glued together, memory controllers included. Here ya go:




And a die shot:

Shame they didn't enable all 8 cores :/

questionable

They couldn't enable all 8 cores and keep the high frequencies at the current tdp.

HalfAHertzThey couldn't enable all 8 cores and keep the high frequencies at the current tdp.

That's true, but how much higher would it be?

Surely, at this already high enthusiast level of performance and price, people will pay for the high performance cooling required (and the large electricity bills...)

I have a triple 120 rad dedicated to just my cpu. Bring on the tdp, I say.

Wile EI have a triple 120 rad dedicated to just my cpu. Bring on the tdp, I say.

+1 :pimp:

Wile EI have a triple 120 rad dedicated to just my cpu. Bring on the tdp, I say.

Hell I used a quad fan rad for my i7 920 at 4.7 GHz :)

As for the topic of this thread, I knew before hand that performance core for core clock for clock was no improvement.

btarunr owes me a 4pack of Guinness in 3 months when IB comes out on 1155 and smokes 2011.

(If timelines are correct and IB-E wont be out for another year)

TBH, it was apparent, the few things that SB-E was supposed to have that were improvements all got canceled. So it was left with PCIE lanes and quad channel memory controller, neither of which are important at this juncture.

There is a actually a slight hit in percore/clcok production due to the overhead generated by the BUS.

Now when IB-E comes out, will it be an improvement over IB... possibly... too early to speculate at this time. But at this time, and over the last 6 months, IB will be better than SB-E. I have nto seen any 58x SB-E chips yet... but for SB that was a matter of binning. Not many people can afford to bin 200 $11000 chips, even ones that could afford to bin 200 $300 chips.

Bulldozer faster & more efficient than Sandy Bridge E per thread.

Bulldozer faster & more efficient than Sandy Bridge E per thread.

ipc_specialist Nov 15, 2011 at 5:54 pm #
No, Intel doesn’t have an IPC lead anymore, Bulldozer is the new champion, check this out:

SB-E beats BD by 30%, but they need:

33% more die space
50% more threads

So, that means that AMD is faster per thread, and if they made a chip as big as Intel’s, then they’d be 20% faster, because they are more efficient per thread.

Or at least that was the reason everyone declared Bulldozer to be a fail, but thankfully SB-E came along to out-do Bulldozer in every perceived shortcoming.

Um, trolling the wrong thread much?:laugh:

nt300Bulldozer faster & more efficient than Sandy Bridge E per thread.

I call bullshit. Faster per thread in what? A program coded by AMD?

Just disable HT if you want "per thread" performance... You get a bit lower overall performance but half the threads...

Per thread performance doesn't mean much, and overall performance is a much better indicator

You you make Bulldozer 33% bigger and more packed, the CPU would need it's own power supply and phase-change cooling :laugh:

n-sterYou you make Bulldozer 33% bigger and more packed, the CPU would need it's own power supply and phase-change cooling :laugh:

Also the SB-E silicon actually has 33% more cores (8 vs 6) and cache (20 vs 15) than what it has been released for now, so any comparison that involves die size (and consequently power consumption) is moot.

For a real comparison of efficiency, compare the 3960X to the FX-6100. Both have 2 cores and a comparable amount of cache disabled.

First,
Bottom line, nt300:

SB-E is only a little better than SB in terms of performance overall (considering most people's usage here).
BUT, it still blows away BD.

That being said, there is no BD chip comparable to the SB-E, and even if there was, per thread performance would not be better, ever. SB tech is 50-75% faster, single threaded, which makes everything faster.

Second,
An SB-E is not comparable to an fx-6100; an fx-6100 is, from an operational standpoint, a triple core with hyperthreading. You'd have to glue together two of them to come to some kind of comparison win a 6/12 SB-E.

InceptorSecond,
An SB-E is not comparable to an fx-6100; an fx-6100 is, from an operational standpoint, a triple core with hyperthreading. You'd have to glue together two of them to come to some kind of comparison win a 6/12 SB-E.

EDIT: Notice that I didn't compare SB-E to FX-6100, that'd be wrong, but only just as wrong as comparing a full Zambezi core to a 3960X. In both cases that's comparing a full core against a crippled core. I compared 2 crippled cores against each other.

I know you're essentially right and you know I've been saying the same thing since BD architecture was revealed. But as to your post, not really, since we are talking about competitiveness. First of all ask AMD about how many cores are there. Second, compare silicon versus silicon. Both SB-E and Zambezi have 2 billion transistors (which is key to comparing efficiency) and 8 cores (again ask AMD). Both 3960X and FX-6100 have 2 cores disabled, simple. SB-E has 4 MB more cache, which makes up for the small transistor count difference and specially die size difference.

I understand what you're saying, but comparing AMD's marketing-speak about 'cores' to the actual cores of SB-E doesn't make sense to me.
SB-E is definitely a more efficient design, that's crystal clear.

InceptorI understand what you're saying, but comparing AMD's marketing-speak about 'cores' to the actual cores of SB-E doesn't make sense to me.
SB-E is definitely a more efficient design, that's crystal clear.

Maybe he wouldn't do that, if AMD themselves hadn't been doing it. They market their Octo-Core CPU's as having 8 cores when they really are just as bastardized as Intel's offerings. AMD made a whole marketting campaign of criticizing Intel for not offering TRUE cores. People in glass houses.

xenocideMaybe he wouldn't do that, if AMD themselves hadn't been doing it. They market their Octo-Core CPU's as having 8 cores when they really are just as bastardized as Intel's offerings. AMD made a whole marketting campaign of criticizing Intel for not offering TRUE cores. People in glass houses.

I have to admit that it it sooooo hypocritical of them to call these true octo-cores when they and their fanboys would criticize Intel's HT and "Real men use real cores" and whatnot

InceptorI understand what you're saying, but comparing AMD's marketing-speak about 'cores' to the actual cores of SB-E doesn't make sense to me.
SB-E is definitely a more efficient design, that's crystal clear.

It makes absolute sense in this context. nt300 was speaking about efficiency, i.e. performance per core/thread and performance per mm^2. So how many cores BD trully has becomes irrelevant. AMD claims 8 cores so that's what I used for the comparison, but I could use 4c/8t and it's not going to look any better. Going with 8c is not the perfect choice, but neither is to compare 1 BD thread to 1 SB thread. They are completely different and have a completely different influence in die size. Intel's approach is almost free, a very small increase in die size and core complexity yields a small (yet bigger than die size increase) gain in performance. AMD increased die size by a lot and obtained a similar increase in performance. The chip is here, and the 20% die increase for 80% performance claim was BS.

Anyway, it doesn't matter which approach you take AMD's design is not more efficient and that's what I was saying, simply.

a) Comparing transistor count/die size. Zambezi is 8 "cores". SB-E is 8 cores/16 threads. Both have a similar transistor budget, if we are taking the approach of comparing architectures by this measure, BD and SB-E are indeed the chips to compare. Now performance wise, with 2 cores disabled SB-E is 40-50% faster compared to 8 "core" BD and twice as fast as the 6 "core" BD.

b) We go by real cores. Then Zambezi is 4c/8t. The only chip we can compare it to is SB 4c/8t. In this case transistor budget is half of that in Zambezi and it even has integrated GPU. Performance is similar, making performance per thread equal when all 8 threads are in use. When only up to 4 are used SB is a lot faster, and remember it's half the size.