AMD Phenom II X6 Processors to Get Dynamic Speed Boost Technology
According to the sources close to AMD revealed to Xbitlabs, AMD will implements a dymanic performance boost technology into its upcoming six-core processors known as Phenom II X6 “Thuban”.
The technology is now called “C-state performance boost”, and works in a similar way to Intel’s Turbo Boost technology. When single-thread performance is needed most, Thuban processors will automatically disable idle cores and overclock the remaining engines to the maximum possible level.
If what we heard is true, it would be something great for AMD to compete against Intel.
February 6th, 2010 at 9:48 pm
I’m an AMD customer, but seriously. These CPUs have only 6MB L3 and that is not good for hexa-core. How hard is it to understand that more cache and faster cache improves per GHz performance. I would rather want some X4 with 12MB L3. Will there be penta-cores with locked 6th core?
Why does this stupid AMD have to copy Intel. We don’t need Turbo Boost or Hyper Threading, in fact we need the exact opposite of hyper threading, technology that would make 6-core have 3 threads, so you could have the full performance in applications that do not support multicore. Imagine the FPS improvement in games.
February 7th, 2010 at 3:12 am
why do they bother with L3 at all, why not just make a hexacore with 64+64 L1/CORE 512 L2/core and no L3, it will use alot less die space, and memory bandwidth will be off the charts, skip phenom, give me an Athlon II X6 2.6GHZ for $150 bucks, that will sell!!!!!
February 7th, 2010 at 5:41 am
are you kidding? Phenom II with L3 is way faster than Athlon II, especially in gaming. 2,5GHz Phenom II X4 905e = 2,8GHz Athlon II X4 630.
February 8th, 2010 at 5:50 pm
@Expertview:
I doubt, that doubling the L3 size for a potential speed up of 10-15% would outperform the 50% added cores, which take about the same area. Because of the larger required die area, the L3 latencies could increase, lowering the positive effect somewhat. There are also diminishing returns, because working set sizes won’t adapt to cache sizes, but stay the same.
The penta core wouldn’t make sense (not as much difference as triple vs. dual or quad core). We’ll see quad core instead (Zosma).
A turbo mode might help more, because – as you said – less threads would leave cores idle (power gating can shut them off then). But using the existing cores to increase performance for those 3 threads you mentioned, would bring too much complexity and too many additional latencies to return any significant increase in performance. See Pollacks rule for an estimate. This thing has been discussed in the past (reverse Hyperthreading) but still won’t provide as much easily gained performance as switching off cores (or putting them to lower P-states) and clock the 100% utilized cores higher.
Some sites:
Pollacks rule depicted:
http://pc.watch.impress.co.jp/img/pcw/docs/346/902/html/05.jpg.html
For more reading my blog:
http://citavia.blog.de/
@xaira:
L3 eases the mem bandwidth pressure caused by many cores. The more cores, the more limiting a missing L3 would be.
February 11th, 2010 at 9:33 pm
Intel has more cache and better performance. Most apps still can’t fully take advantage of 4 cores, as even in apps that are multithreaded, the effeciency drops down anyway. CPU load in multithreaded apps: X2 ~ 95%, X3 ~85%, X4 ~80%. Even if u lock the quad to CnQ most apps can’t get over 85% out from the quad. Who buy those expensive desktop processors – gamers. In most games you won’t see much difference between X3 and X4, so when it comes to games, reverse Hyperthreading would bring significant performance increase with X6 or better. And if you look into future – they keep increasing performance by adding cores, would you really expect the apps in future to have 16 threads, 32, 64, 128, 256? Do games need to have 3200 threads to get the optimal performance out from 3200 shaders, that’s just bs. Turbo mode with some extra 10% clock doesn’t give a shit, many of us run our processors overclocked anyway.
Did they have X6 or X8 CPUs in the past? It wasn’t effective on some dualcore or maby even on quad, but the ammount of cores is growing faster than fully optimized multithreaded apps, so the reverse Hyperthreading is the future. And why you think it has to be so strict, it could just turn these extra threads on when it’s necessary, just like turbo.