Guess I'll get an Ivy next year... (The CPU not the saur)
What a disappointment. Guess skipping another generation or two again.
This is what boggles my mind.
4 core sandy bridge = 900 million transistors at 32nm (about 1/3 of that is the integrated graphics processor so 600m is cpu logic/caches/bus/etc.).
8 core bulldozer = 2 billion transistors all cpu logic/caches/bus/etc., no integrated graphics processor, 32nm frabrication
In other words bulldozer has 3 times as many transistors if you ignore the IGP. Yet sandy bridge blows it away, big time. Do the math, sandy bridge is about half an order of magnitude more efficient in terms of processing power vs. transistor count.
Also bulldozer was an attempt to focus on multiprocessing and clock rate yet sandy bridge still overclocks better despite having a shorter pipeline, less transistors, and smaller core logic.
Only shows that Intel is the only way to go.
When my 2600K have done it's share, it's a new Intel that's gonna be installed and nothing else.
And if I didn't have the 2600K I would have wait for the upcoming 2700K and take that over the Bulldozer.
(10-13-2011, 06:31 PM)Maverick Hunter X Wrote: [ -> ]Only shows that Intel is the only way to go.
When my 2600K have done it's share, it's a new Intel that's gonna be installed and nothing else.
And if I didn't have the 2600K I would have wait for the upcoming 2700K and take that over the Bulldozer.
Yeah, I should've done the same, really, looking back. You'd have thought the signs would've been all too clear even months ago, but I do like my AMD...
I'm going to get Z75 board with Ivy come spring, I believe.
(10-13-2011, 09:19 AM)NaturalViolence Wrote: [ -> ]In other words bulldozer has 3 times as many transistors if you ignore the IGP. Yet sandy bridge blows it away, big time. Do the math, sandy bridge is about half an order of magnitude more efficient in terms of processing power vs. transistor count.
The IGP takes up about
1/5 of the die. BD has 2x the cache (
L3$ takes about a 1/3 of the die instead of 1/5 on SNB so AMDs cache is less dense further increasing die size) where most of those transistors are going and 3 more HTT links that a single socket SNB doesn't need.
If the FX-6100 overclocks above 4.5 as easily as 8k and the price drops even further It's gonna be a bang for the buck. Performance when OC'd should be on par/higher than 2500k so im okay with this. Awaiting Dolphin SMG/LLE tests as well as PCSX2 SSE4.1/AVX.
Obviously FX-8150 being priced between 2500k and 2600k is not worth buying, looking forward to a price drop or lower clocked variants.
(10-13-2011, 09:19 AM)NaturalViolence Wrote: [ -> ]This is what boggles my mind.
4 core sandy bridge = 900 million transistors at 32nm (about 1/3 of that is the integrated graphics processor so 600m is cpu logic/caches/bus/etc.).
8 core bulldozer = 2 billion transistors all cpu logic/caches/bus/etc., no integrated graphics processor, 32nm frabrication
In other words bulldozer has 3 times as many transistors if you ignore the IGP. Yet sandy bridge blows it away, big time. Do the math, sandy bridge is about half an order of magnitude more efficient in terms of processing power vs. transistor count.
Hmmm, at this point this is a bit more technical then what I'd try to discuss ^^
But you made me too curious, there is nothing more that could influence this much on performance?
@lamedude
Is it enough to make this difference? I was expecting SandyBridge to beat BD specially on single threaded performance, since per core performance wasn't one of the goals, but I imagined something far more... sutil. And what about overclocking? I mean, what are these things dependent on, besides the obvious? something about the socket format/structure maybe? I dunno, as I said I'm very nooby about this
Quote:The IGP takes up about 1/5 of the die. BD has 2x the cache (L3$ takes about a 1/3 of the die instead of 1/5 on SNB so AMDs cache is less dense further increasing die size)
Yeah I was thinking of the dual core sandy bridge die. Still even with 1/5 that means 720million transistors if you remove the IGP. Even with 720m bulldozer still has about 3x the number of transistors if you exclude the IGP, so my point still stands. It's odd that bulldozers cache is so much less dense yet still has a much higher latency and much lower bandwidth than sandy bridge.
Quote:where most of those transistors are going and 3 more HTT links that a single socket SNB doesn't need.
But the die shot shows all of the HTT links together using only about 5% of the die space. It doesn't seem to use much space at all. Nearly all of the die space appears to be taken up by the modules and caches.
Quote:But you made me too curious, there is nothing more that could influence this much on performance?
Huh? I don't understand what your sentence is asking.
As I stated above bulldozer does beat sandy bridge in heavily multithreaded integer heavy applications but loses in every other area since it is inferior in every other way (it's only real advantage is having 8 integer cores). Sandy bridge overclocks better according to all of the reviews I've seen.
yeah I was asking why was this happening, now I got that
I didn't understand your post entirely, now I see what you mean