Sign in with
Sign up | Sign in
Your question

AMD delays Phenom 2.4 GHz due to TLB errata - Page 2

Last response: in CPUs
Share
a b à CPUs
November 19, 2007 10:17:27 AM

Well if they OC easily to 3Ghz and are heaps cheaper then I'll consider one.

The bigger concern is the multithreaded performance ... which is frankly a real puzzle.

The QX6600 just seems to power along while the Phenom's performance is all over the place.

Still seems to me like there are multiple issues with the chip.

Plus the memory results are shocking.
November 19, 2007 12:03:43 PM

Quote:
Actually I believe it was 3.4GHz, but it's possible. Again the issue will be power consumption.

Mean to tell me native quad-core hits a power consumption wall at 3.4GHz while glued dual-core has managed 3.6GHz on air? Oh well, so far it seems the chips to reach 3.0GHz only come out once every couple wafers. They have many more process issues to iron out.

Quote:
If I remembered correctly, AMD puts L2 to L3 wiring directly above the die itself. This means, in order for the data to transfer from L2 to L3, it has to first go through about 10+ layers of transistors, to reach to the top of the die. During the process, it will be subjected to heat and capacitance interference. Then the wiring has to go through the top of the cores, which will be subjected to heat and electromagnetic interference. After that, the wiring has to go through another 10+ layers of transistors, and finally reached L3.

If the diagrams were drawn accurately to this description and publicized, it would be a travesty, too. AMD has been such a proponent of direct-connect, and now it seems we have what's practically a separate L3 chip glued with the processor package, only they're really the same die, so if one part fails, the other gets tossed.
November 21, 2007 3:51:04 PM

I wonder why the TLB issue did not surface until production efforts were well under way? It would seem the prototypes would have revealed this issue way before production.

This problem is either smaller that I originally thought, or represents a much larger issue that is not easily fixed. From the current vantage point, it appears to be the latter.
Related resources
November 21, 2007 4:47:33 PM

onestar said:
I wonder why the TLB issue did not surface until production efforts were well under way? It would seem the prototypes would have revealed this issue way before production.

This problem is either smaller that I originally thought, or represents a much larger issue that is not easily fixed. From the current vantage point, it appears to be the latter.


Here's my guess, and its only a guess. The TLB probelm had surfaced, and maybe that was why Phenom wasn't introduced earlier, like last winter or even spring. AMD kept working on the it, but was finally forced to release the chip as it was and deal with fixing it later.

This may have been part of the deal with the $622 million dollar investment they got from the Arabs. Of course, AMD may not have revealed the long known about problem at that time. Don't know at all about that. Of course, if that's true, it reveals a much larger problem within the company that will not be easily fixed, as you point out.
November 21, 2007 5:02:56 PM

I think it was a long know problem.
Some of the articles claimed that is why Faster Phenoms were releaed.

The problem simply existed at slow speeds than AMD had hoped for.
Personally, it would not leave me with much confidence in the 2.2 and 2.3 versions either.

It's just a question of degree.

It probably explains why reporters were not allowed to OC and Benchmark the CPUs. They could load benchmarking software on the systems given to them but they were not allowed to OC those systems.

The one system they were allowed to OC, they were not allowed to Benchmark software on. That is why I found THG comment that they were able to get their system stable at 3.0 Ghz. (So long as they did not do anything that just let Windows boot.................)
      • 1
      • 2 / 2
!