dynamic super threading

[Delusional Phil]

Its a balancing act, many cores is good when you have a workload that can take advantage of it but not everything can, often you need to know the answer to A before you can figure out B in which case 1 more powerful core is beneficial. Granted with enough cores you could work out every permutation of B while you are working out A and use the correct answer but then you loose efficiency. Many cores is great for computation but often fewer faster cores are better for general desktop use.

Yup ... That was AMD's downfall ... 

[linuxfan66]

Its a balancing act, many cores is good when you have a workload that can take advantage of it but not everything can, often you need to know the answer to A before you can figure out B in which case 1 more powerful core is beneficial. Granted with enough cores you could work out every permutation of B while you are working out A and use the correct answer but then you loose efficiency. Many cores is great for computation but often fewer faster cores are better for general desktop use.

I thought that was easy to answer because Intel chips can already detect when cores aren't used and slow down/turn them off to make them more powerful. Why not take that one step further and merge the unused cores processing power doubling available transistors for used corr

[Grrizz]

I thought that was easy to answer because Intel chips can already detect when cores aren't used and slow down/turn them off to make them more powerful. Why not take that one step further and merge the unused cores processing power doubling available transistors for used corr

Because at the end of the day the architecture of each core is designed to compute basic math i.e. having twice as many calculators wont help you figure out 3 x 3 any quicker you will just have one redundant calculator, things like hyper-threading have more to do with taking advantage of the down time of parts of a core due to latencies than they do with throwing more transistors at the same problem.

[linuxfan66]

Because at the end of the day the architecture of each core is designed to compute basic math i.e. having twice as many calculators wont help you figure out 3 x 3 any quicker you will just have one redundant calculator, things like hyper-threading have more to do with taking advantage of the down time of parts of a core due to latencies than they do with throwing more transistors at the same problem.

so transistors per core dont translate into ipc. i t thats why fxes were awful at normal loads and i*s were awesome at normal laods

[Grrizz]

so transistors per core dont translate into ipc. i t thats why fxes were awful at normal loads and i*s were awesome at normal laods

Yea basically, more transistors can help but the transistors need to be organised in such a way to achieve a specific goal so you would have to be able to take one core that was designed to compute math and dynamically reorganise/reconnect those transistors to another core in a way that would aid in making its math calculations faster (without the latency required to do that impacting the result) which might be doable but would be an incredible feat of engineering not to mention likely cost far more than the benefit it would bring because you would have to add additional transistors just to handle the circuit switching adding to the heat output and die size which would in all likelihood out weigh the benefit of its implementation vs. just having fewer more powerful cores, I'd be hesitant to say it would be impossible (as with most things) but I probably say it would be highly impractical.