Then you look at the power consumption.
Desktop or server, IPC is still down.
That is why I dont understand why everyone says Bulldozer will be such a great server chip. I admit I dont know much about servers, but isnt power consumption of paramount importance in the server area?
Still dont see how a chip with such low IPC and high power consumption can be that great for servers either. Couldnt Intel just put more cores like AMD does, except each core is more efficient?
Nope. I'm also waiting to see some SPEC benchmarks but still nada. :hmm:Does anyone have benchies of a Bulldozer CPU in server workloads?
Nope. I'm also waiting to see some SPEC benchmarks but still nada. :hmm:
It's not speculation. Bulldozer does well with highly parallel code and can hang with an i7 2600K, albeit while consuming a good 100 watts more to do the same thing.The speculation that Bulldozer will be a better server part without any testing is driving me nuts. It's not like you have to have an Opteron to see how it scales with SQL queries and such.
It's not speculation. Bulldozer does well with highly parallel code and can hang with an i7 2600K, albeit while consuming a good 100 watts more to do the same thing.
yeah he's called JFAMD
im sorry we dont need an "offical" thread.
JFAMD will tell us what he can, and what he can't he wont.
As long as people dont bait him and call him names, he's usually very helpful in answering AMD questions.
Some really good questions being asked over at Hardforums, makes this forum look a little bad.
Some really good questions being asked over at Hardforums, makes this forum look a little bad.
What is really going to make or break that effort, is if the question will be answered honestly, or danced around like a politician around Zucotti Park. (Occupy Wall Street).
He's a marketer, not an engineer, not a technical guy. He repeats what the engineers and higher up marketers tell him. "IPC Increases," "50% performance increase" over Magny-cours, "35% performance increase" over Magny-cours. "It's better to have threads run on the same module and for BD to turbo." Honestly, what questions has he answered that are correct? So far none of these have been correct. "All benchmarks are not representative" even though obrovsky and many other pre-released benchmarks had it dead on. He stated the only accurate benchmarks would be from AMD, and honestly, those are the second to last benchmarks I would trust, slightly above Intel's benches of AMD products.
It's funny how his sig says all of his statements are his own, but every single one of them is towing the AMD line.
Interesting, pretty much every question is "when are you planning to release a good version of BD?"
My question would be:
"Why didn't you die-shrink Phenom when you realized BD 1.0 was a disaster?"
Imagine a die-shrunk 1090T BE using much less power, and costing enough less to make that it could be sold profitably for $99.
A six-core Thuban at 45nm is 346mm^2. With perfect scaling, 32nm should have made that roughly 175mm^2. Even without perfect scaling, it just seems like AMD could have made a true 8-core CPU, with the IPC increase realized by Llano, added a couple of features that made it into BD (like AES), and still made it under 250mm^2.
2 Llano cores, complete with 2MB L2$, are nearly identical in size to one BD module (also with 2MB L2$).
So, keeping everything that is uncore about Zambezi (power gating logic, L3$, etc), in theory they could have made an 8-core llano architecture zambezi that would have been the same die-size as the existing 8core BD.
What we don't know is how well it would have clocked, or how its power consumption would have turned out, considering where the A8's are clocking now and their power usage as mere quads.
Just as it took the Pentium 4 a number of years to "die", it's going to take some time before the world is void of Bulldozer-esque CPUs. AMD will try to re-spin it and from the sounds of it enhance it with their "Piledriver" product.Its all well and good keeping up with a 4C/8T chip in MT apps, but at what expense? 2 billion transistors and a massive power draw as you stated. What happens early next year when Ivy Bridge comes out?
Just looking at the upcoming IB QC chips at 77W TDP, it appears Intel could easily make a 6C/12T IB chip with a ~100W TDP... then what for AMD? A 3 billion transistor 12 core Piledriver at 200W TDP?!
2 Llano cores, complete with 2MB L2$, are nearly identical in size to one BD module (also with 2MB L2$).
So, keeping everything that is uncore about Zambezi (power gating logic, L3$, etc), in theory they could have made an 8-core llano architecture zambezi that would have been the same die-size as the existing 8core BD.
What we don't know is how well it would have clocked, or how its power consumption would have turned out, considering where the A8's are clocking now and their power usage as mere quads.