AMD Bulldozer Core Patent Diagrams

Chips do get shipped ahead of launch to larger partners.
If the news stories about much of the Interlagos processors going to HPC are correct, it can also be a case of contractual obligations being met even if the launch date has moved.

Going with Interlagos first is consistent with the desire to meet demand of a high margin market, or contractually obligated deliveries.
Multisocket servers tends to fall in the former, while price/perf sensitive HPC can be the latter.

What neither case emphasizes with Interlagos is the kind of volume or pressure on the manufacturing process that a $300 (edit: and down) competitor to an i7 would entail.
 
Trying to do a small write-up for work on power loads, energy consumption and efficiency between Westmere and Bulldozer. Comparing Westmere and Magny Cours it seems AMD have a big advantage in energy consumption and Bulldozer will extend this further.
Bulldozer may not be such a failure for AMD after all, especially for its intended market.
 
Last edited by a moderator:
Rootax said:
I think Anand said on tweeter to not trust early and leaked performances info.
Click to expand...
That has been said since pretty much forever. I'm not saying those numbers are correct, just that they might be. There have been plenty of examples where prelimenary leaks have been spot-on in terms of final performance. Core2 is probably one of the best known where the final numbers leaked out months before official ones.
 
Trying to make some sense of the benchmark and it looks like Cryptography score is good if not excellent and the multi-core efficiency benchmark is either not suited to Bulldozer or vice versa.
 
As always, the biggest issue for AMD will be the cache/memory performance and design choices. There was a discussion by some Linux kernel developers (the site is down for now), about strange woes with Bulldozer's L1 instruction cache behaviour -- cache line conflicts trashed the overall performance to unseen lows in some cases.
 
Extremely impressive. I wasn't aware clock speeds had been pushed up past 8GHz already...

Then again, vcore was set to 2.016 during the run, so not exactly viable for long-term use. :oops:

I wonder how much power that thing dissipates at nearly 8.5GHz clock speed, even with just 1 bulldozer module active.
 
I would have liked more info on the 5 GHz+ clocks on more realistic setups, and whether any of them were on non-suicide voltages and at what performance and stability.

The LHe run, while a fun record book blurb, is no more relevant than the neighboring entry on largest beard of bees.
 
3dilettante said:
I would have liked more info on the 5 GHz+ clocks on more realistic setups, and whether any of them were on non-suicide voltages and at what performance and stability.

The LHe run, while a fun record book blurb, is no more relevant than the neighboring entry on largest beard of bees.
Click to expand...

Apparently 5-5.5GHz on high end AIR / cheap water with all 4 modules active and stable/usable around 1.5V. Seems for now GF has quite a bit of variation in process so samples spread-out is large.
Just to be on a safe side I think 4.8GHz should be doable in a normal enclosed system with full stability on regular basis. Besides historically AMD gains 200-400MHz over first 6 months of manufacturing so next year around IvyB they should be over 5GHz for average home OC.

Leaked pricing seems to indicate they some way behind SandyB per clock which means fully tweaked system won't match Intel for low thread tasks, but MT should be at least competitive. Can't wait for more info, already sold my X6 which I will part way with next week.
 
I just want to know if they can trade blows with a 2600k. Not in clocks, but real performance.
 
You must log in or register to reply here.
Back
Top