AMD - Bulldozer Chitchat

[Biscuit]

gotta be BD!

[CAT-THE-FIFTH]

I wonder if this refers to Bulldozer too:

http://www.facebook.com/photo.php?fb...type=1&theater

[CAT-THE-FIFTH]

The announcement is also mentioned here:

http://amd-member.com/newsletters/De...nZone2011.html

[Terbinator]

My body is ready.

[kopite]

please be some stats and a release date!

[CAT-THE-FIFTH]

It seems Hexus has some new info:

http://www.hexus.net/content/item.php?item=31727

[DanceswithUnix]

lol, overclock to 8.4GHz.
http://www.anandtech.com/show/4770/a...ozer-processor
4.8GHz sounds nice though.

[kalniel]

It's the new celeron!!

[Hicks12]

very impressive, as bulldozer isnt meant to be a cheapy budget poor performance chip and is actually clocking higher than intels clock whore then well... either AMD have completely messed with the clock for clock performance and so its terrible or its still very good and just has a crap load of head room.

Interesting stuff, cant wait for release as ill finally be able to get something better than a dual core lol!

[CAT-THE-FIFTH]

Interview with one of the chaps who was involved in the overclocking attempt:

http://translate.google.com/translat...srekordet.html

"NordicHardware: "It says you reached Frequencies" well above 5 GHz, "with only air and sub-$ 100 water cooling solutions. Mind Going Into more detail about your ventures on air and water?"
Sami Mäkinen: "In terms of CPUZ MHz I've seen VIRTUALLY every CPU reach over 5GHz When Using a good air / water cooler. I believe the highest result I saw with aircooling was around 5.5GHz CPUZ. This CPU hit around 7.8GHz on LN2 '.
"

[EvanJackPenn]

I wouldn't be surprised if it's just lots of headroom. It's the first of a new architecture - you're not going to ship at 5GHz, just because you can. I'd be happy with a lower, more certain speed with the choice to OC.

My friend who seems to hate ANYTHING which isn't in his computer (i.e., not apple, intel or nvidia) just keeps saying "More cores than needed". Nice approach to progressing technology. Yes, there might not be much which will use 8 cores fully now but it might start software to lean that way. Hardware is nothing without software. But in real life terms, does an 8 core really have much advantage over a 4 core? I'm most interested in game performance.

[kalniel]

But in real life terms, does an 8 core really have much advantage over a 4 core? I'm most interested in game performance.

Think of it a little like Intel's hyperthreading. HT can, in the best circumstances, make it seem like a 4 core processor has 8 cores, for a very small increase in transistor count. AMD couldn't do hyperthreading, but they instead doubled the integer processing cores also for a relatively small increase in transistor count. OK, it costs more than HT, but for operations that don't need a whole module, the increase might be much more tangible than HT as well.

It's not easy to think of many integer heavy tasks that are very easily parallelised, but decision trees are one of them, so things like AI, path calculations etc. are going to be well suited to it. However the meat of game processing tends to be floating point intensive - in the future this is could be very well accelerated by AMD's APU integrated into the chip, but currently without this it's not quite as well balanced IMHO. On the other hand while games aren't usually very heavily threaded, they do respond well to clockspeed.

[watercooled]

Think of it a little like Intel's hyperthreading. HT can, in the best circumstances, make it seem like a 4 core processor has 8 cores, for a very small increase in transistor count. AMD couldn't do hyperthreading, but they instead doubled the integer processing cores also for a relatively small increase in transistor count. OK, it costs more than HT, but for operations that don't need a whole module, the increase might be much more tangible than HT as well.

It's a matter of design choice rather than what they could or couldn't do. SMT is nothing specific to Intel, for example the Sun UltraSPARC T2 can handle 8 threads per core. Hyperthreading is a double-edged sword however - while some applications will benefit from an almost linear improvement, it actually harms the performance of others, and the rest won't notice any difference in performance. On the other hand, multithreaded performance will always scale with extra integer cores.

It's not easy to think of many integer heavy tasks that are very easily parallelised, but decision trees are one of them, so things like AI, path calculations etc. are going to be well suited to it. However the meat of game processing tends to be floating point intensive - in the future this is could be very well accelerated by AMD's APU integrated into the chip, but currently without this it's not quite as well balanced IMHO. On the other hand while games aren't usually very heavily threaded, they do respond well to clockspeed.

More and more modern games (i.e. not using dated engines) scale very well with more cores - take Deus Ex as a recent example.

[CAT-THE-FIFTH]

It's a matter of design choice rather than what they could or couldn't do. SMT is nothing specific to Intel, for example the Sun UltraSPARC T2 can handle 8 threads per core. Hyperthreading is a double-edged sword however - while some applications will benefit from an almost linear improvement, it actually harms the performance of others, and the rest won't notice any difference in performance. On the other hand, multithreaded performance will always scale with extra integer cores.



More and more modern games (i.e. not using dated engines) scale very well with more cores - take Deus Ex as a recent example.

Supposedly, a Phenom II X6 will be faster than a desktop FX8150 in multi-threaded applications according to "experts" on internetz forums!!

If that is the case AMD would simply just launch Bulldozer next year freeing up 32NM capacity for both Interlagos and Llano. After all if the Phenom II is going to be faster you might as well use a mature 45NM process too. Even, the original Phenom had better IPC than the preceding K8.

[kalniel]

Supposedly, a Phenom II X6 will be faster than a desktop FX8150 in multi-threaded applications according to "experts" on internetz forums!!

That would be very unusual, but I think you won't find said 'experts' here at Hexus

[watercooled]

Again, it's probably some early engineering samples they've had their hands on. These samples are meant purely so hardware partners can sort out compatibility and such. This is very early, experimental silicon made on a very small scale on a brand new process - anyone who claims it relates to final production performance simply doesn't know what they're talking about.