With the clichéd Phenom-phenomenal word play out of the way, let’s get stuck into the ins, outs and all abouts of the newly released quad core from AMD.
This quad core processor marks the end of a very, very, very long wait. Following delay after delay we appear to now have a functioning quad core processor from AMD; or do we? The first bit of info to be passed to the pages of black and green is that there is a bug within the L3 cache in the Translation Lookaside Buffer (TLB). This was found in the initially launched Phenom processors, hence the delay on the flagship Phenom X4 9700 hitting the retail market.
This TLB erratum has been the cause for much concern; in rare circumstances when all four cores are being utilised and large amounts of memory transactions are made the system can lockup. This issue is currently rectified through the use of a BIOS-level workaround for the TLB erratum and there are patches to achieve the desired effect – complete system stability. Unfortunately, this patch incurs a roughly 20 per cent performance hit compared to the non-patched performance. This issue is said to be rectified in a silicon level fix in the B3 stepping processors earmarked for release in Q1 this year. I guess it’s too bad we’ve got the yet to be released Phenom X2 9900 and it’s a B2 stepping sample – oh well, on with the testing.
We chose not to install the BIOS-level workaround, because we wanted to see the true impact of this rare issue and whether it would hinder our testing. It didn’t. That’s not to say that the fix isn’t necessary as it’s possible that the issue may raise its ugly head in other, more long-term circumstances; but at least within our experience we did not come across the problem.
The Phenom line of processors is obviously the first line of quad core processors from AMD. What makes these new chips unique is not just that they’re the first of their breed, but rather that the design is the first to house four cores on a single silicon die. Intel quad cores utilise a multi-chip module (MCM) solution, placing two dual core dies within a single package, thus constructing a quad core processor. Although this approach is an advance for micro-processor design and technology, it inherently causes issues in areas such as yield quantities. This is where the Phenom triple-core concept is thought to have been born; that said, neither enthusiasts nor media have had a confirmation of the Phenom triple-core variety from AMD as merely being Phenom quad-core chips with failed cores disabled.
Anyway, what can this puppy do? Being a stock 2.6GHz part, the chip bares a 13x multiplier at stock on a 200MHz HTT bus (FSB for the Intel fans). The Phenom X4 9900 we used had an unlocked multiplier with 0.5x increments available for scaling. As such, we decided it was important to test the two sides of the overclocking coin. We wanted to know what max overclocks were like as well as performance by utilising both HTT overclocking and multiplier overclocking to achieve the biggest possible numbers.
The end result of the two schools of overclocking going head to head was the HTT maxing out at 235MHz and using the multiplier we maxed out the overclock at 15.5x the stock HTT of 200MHz. The 16x multiplier would begin to boot but would then fail upon loading Windows as the heat started to skyrocket. These chips run hot -- 140W TDP at stock -- and overclocking just makes things much, much worse.
Combining the power of HTT and multiplier overclocking we maxed out at 3.105GHz with settings of 230x13.5. It’s important to mention that due to the isolated memory controller of the Phenom, which is not influenced by the HTT frequency, we obtained the best overclocking results by simply utilising the multiplier.
All in all we found this chip to be hotter and slower than the 65nm Core 2 Quad processors, let alone the 45nm Core 2 Quads, with bonus restrictive memory frequencies. And to top things off we aren’t likely to see the B3 chip revision until Q1 ’08.