AMD - Bulldozer Chitchat

[scaryjim]

One of the tables posted earlier said these were all black edition, and if that's true then they'll all have an unlocked mulitplier for easy overclocking. But that said, they also don't have the silicion integration of Sandy Bridge / Fusion, so they should also be partial to base clock increases as well - that gives a lot more flexibility in finding the perfect balance between multiplier and base clock.

EDIT: and core in this instance means integer core, of which there are two per bulldozer module: so the top end chips will be 4 module, 8 core.

[Pob255]

Just having an unlocked multiplier doesn't mean they will be massively overclockable, that's been one of the big selling points for intel since the c2d the amount of overclocking headroom on thee chips has been large.

going back a bit a friend ask me to help overclock a 7750BE I wasn't too familar with amd overclocking and the best I could get without touching the voltages was x14.5 from x13.5 so a massive increase of +1 to the multiplier.
Granted AMD improved since then, my point is just because it has an unlocked multiplier doesn't mean it has lots of overclocking headroom.

[scaryjim]

True, but EJP was specifically asking about *easy* overclocking, not the amount of headroom

The 7750 is a bit of an unfair example for clocking, since it was based on Phenom I which was a notoriously bad chip for clock speeds. Phenom II was much better. That said, sometimes you can get higher CPU clocks from pushing the base clock up and keeping the multi down. My Athlon XP (hurrah for VIA chipsets that allow upward multiplier adjustments ) always ran more stably with the multi notched to 13x (from 12.5x) with a higher FSB, than trying to push the multi up with stock FSB.

Of course, with relatively high launch clocks it's anyone's guess as to how far BD will clock beyond that. Only AMDs engineers will know how high the theoretical peak clock is, and how close the shipping silicon will get to that. Don't forget that Prescott was intended to ramp to 10GHz stock, but heat & power issues meant it never even reached 4...

[Terbinator]

What do we reckon the max reasonable O/C will b out of these then? 4.5GHz, give or take, like Sandy?

OTOH the 95W 8120 might be worth a gander.

[Biscuit]

i wonder wether the hex/quad will be chips that possibly have faulty cores and can be unlocked?

[Hicks12]

i think nearer to 5ghz, maybe even 5ghz is possible with these... if turbo goes at 4.5ghz then id think its safe to assume 5-10% more than that with a bit more juice... lets hope for that yeah? 5ghz 8 core chip, interesting thoughts so lets pray AMD have matched(or close) to Sandy bridge clock for clock.

[shaithis]

and there's also the outside possibility of core unlocking

[EvanJackPenn]

Do the little asterisks here mean anything? Perhaps a special version? Or is it just pre-overclocked?

[Hicks12]

probably subject to change or an estimate.... i guess it depends entirely on how well AMD yields are and specs they are coming out to.

Time will tell , we know that clocks arent going to be bad though (above 3ghz is good imo) however the key part is knowing how good it is clock for clock, would be interesting if it really does beat intels current stuff and it would be a nice change .

[shaithis]

It's a bit of a leap but by x-referencing that table to what I see on the gigabyte bios updates, it may be that the * denotes models that will have variants......much like intel do with SB.

[EvanJackPenn]

I wonder whether things will adapt to using all of these available cores. I know that many games stick to the two...

[shaithis]

I wonder whether things will adapt to using all of these available cores. I know that many games stick to the two...

Think of it like "reverse hyperthreading".....

Instead of trying to split a core with HT, AMD make 2 work as 1 when there are less threads active.

[EvanJackPenn]

Think of it like "reverse hyperthreading".....

Instead of trying to split a core with HT, AMD make 2 work as 1 when there are less threads active.

Ah I see. So if something is only using 2 cores, BD will double it up and use 2 sets of 2 (2 modules?) in conjunction?

[scaryjim]

not exactly. A bulldozer module is basically a hybrid processor: it has 2 sets of integer cores, which is why it appears as 2 cores to the OS, but it only has one floating point operator (although it's a big fat one!). A properly written threading aware game will run equally well regardless of the number of cores in your computer - it simply won't gain any benefit from having more cores available. That's partly what turbo core is meant to address - if you're not using all the cores then it'll power gate some of them and speed up the others. Intel's Turbo Boost does this particularly well.

There are, unfortunately, one or two instances of games that were written on the assumption they'd only be run on one core. Neverwinter Nights is the worst example I know of - play a clean install on a high end quad core rig and it's practically unplayable. This is because the OS keeps rescheduling the thread between the cores and the game isn't written to handle that (or something similar, anyway) - it needs locking to a single core. But games written like that are few and far between, to be honest.

Some games are now being written to make better use of multicore system, but actually you need very specific workloads to get the best out of all those cores and a lot of games simply wouldn't benefit much from being heavily threaded. But as long as the OS can schedule threads appropriately, and turbo core / turbo boost do their job, there's no really disadvantage to paying for as many cores as you can afford...

[CAT-THE-FIFTH]

It looks like an Interlagos based Opteron is on sale:

http://www.smsassembly.com/AMD-Opter...KTGGGU-2.3GHz/

[scaryjim]

only 512K L2 cache per "core"? I thought BD was meant to have 2MB per module, giving 1MB per "core"...?