Some nice speculation on Hammer

[Martin]

Interesting, but keep in mind, only speculation.

 

[grant2]

It makes sense.... 50% higher IPC is pretty good return for 20% more stages!

 

[ProviaFan]

Wow, if he's right, that would be awesome. I can't wait for hammer. 😀

 

[bigboxes]

Interesting read. Thanks for the linky! 🙂

~box

 

[Rand]

It doesnt seem at all viable from my POV.

He compares this to Intel's double pumped ALU's but this seems a flawed comparison.
The P4's ALU's are comparatively simple whereas X86 decoders are awfully complex.
Only one of the P4's ALU's are complex ALU's capable of handling all X86 integer instructions, the other 2 ALU's are "simple" ALU's that can only handle roughly 60-70% of instructions and have limited instruction lengths. Intel was only able to double-pump the two "simple" ALU's.

Double-pumping a full X86 decoder would seem to be considerably more difficult then double-pumping the P4's single complex ALU- a task Intel was not able to accomplish. I can't imagine it's at all a realistic proposition.

Edit: Ace's Hardware's board has a thread with similar sentiments as my own view.

 

[zephyrprime]

I also think the article's argument is uncompelling.

The author seems have a screwed up conception of instruction level parallelism. He says, "the Athlon can theoretically decode and execute three instructions per clock cycle, but poor instruction parallelism keeps it closer to two instructions per clock cycle in most instances." If this is the case, why would having a more powerful instruction decoder help things at all? Poor instruction level parallelism is the fault of the way code is and the number of registers on the x86 architechture.

Also, a double pumped decoder would contribute nothing to higher clock speeds. I think it highly unlikely that AMD would not try to increase clock speed by means of longer pipelines when this has been the standard thing to do with processors ever since pipelining was invented.

I never understood why the Inquirer has such a bad reputation but now I'm beginning to.

 

[socketman]

Its a reasonable argument. And he made it real clear in the article it was speculation. Sadly, we will likely have to wait at least 3 more months before any kind of benchmarks.

Ive only seen one benchmark on hammer. But the the number was so outrageously good i have serious doubts it is accurate. So i will just wait like everyon else.