AMD’s “fusion” approach has been pushed as strategy for the past few years, however, until now, the concept has been somewhat nebulous. That is, there wasn’t really anything ground-breaking new that would justify the “the future is fusion” motto, but of course the same line can also be interpreted to mean that fusion was really a thing of the future. Arguably, AMD has been trail-blazing some of the features on new CPUs that nowadays are essentially granted. For example, take the integrated memory controller, which in the case of Intel’s Timna never made it into production and which really fused the CPU proper with the North Bridge into a single monolithic design.

Needless to say, though, that the CPU architecture still followed the x86 concept with the addition of the floating point co-processor in a single, integrated design as it has prevailed since the days of the 80486 or the K5 processors, regardless of whether there were some difference in the ratio of integer vs. floating point units. In other words, there hasn’t been anything spectacular that would increase the efficiency of the CPU as we know it on the basis of architectural / conceptual changes.

In our last CrossfireX article we performed a mix and match experiment, pairing up two RADEON 5870 from two different manufacturers and also with two different memory configurations to see whether they would play along nicely. The outcome was that in most of the applications tested, no problems occurred, however, in some of the synthetic benchmarks like FutureMark’s 3DMark Vantage, some of the test suites scored more like a single card. More importantly, we saw similar results in some of the games we tested. Needless to say that the potential root causes could have been anything from a system or CPU limitation to differences in the BIOS string and last not least the differences in memory space allocations that could wreak havoc on the performance. The bottom line, regardless, was that we didn’t recommend extreme mix and match because there may be some sleeping dogs that wake up just at the most inconvenient time or game play.

To look into the issues encountered more closely, we took a pair of uneven twins, that is ASUS RADEON EAH5850 and ASUS RADEON EAH 5850TOP, the first one following AMD’s reference design and the second one ASUS home-brew concoction of the 5850 series and teamed them up in CrossfireX. The default frequencies and the port configuration as well as the layout and cooling solutions of the two cards are about as different as one could build them within the same design constraints posed by the components but we did not really expect to find any major problems.

Solid state drives are becoming increasingly popular. The trend started with the enthusiast and early adapter community, and arguably, there were a few bumps and glitches. Some of the primary issues were simply caused by trying to squeeze solid state drives into an environment custom tailored towards hard disk drives, some other issues were insufficiencies on the SSD controller, particularly with respect to the exact workings of NAND flash.

In several earlier articles, we already covered some of the important aspects of flash memory cells and the inherent limitations caused by "Fowler-Nordheim" quantum-mechanical tunneling as programming and erase mechanisms. We also covered some of the system-level challenges faced by SSDs along with countermeasures implemented in the latest versions of operating systems or drivers to keep SSDs at a steady state of performance by proactively performing TRIM and garbage collection.

The one thing we have not covered yet is the heart and soul of NAND flash memory, that is, the organization of NAND flash, as opposed to for example NOR flash or any similar non-volatile memory, which is the reason for its cost effectiveness but also has some functional consequences in the form of read disturb and write disturb. Disturbing? On the surface, yes, but not if you drill down below the surface of the buzzwords to understand what you are actually talking about.

Growing up in the last millennium and reading a lot of science fiction, not to mention going on the daily quests for UFOs, has the advantage that quite a few of the new names popping up in the tech world are, in fact, very old acquaintances. The Little Green Men from Mars, the Arecibo message and finally the Fermi paradox were things that any geek had to be familiar with. I mean, you didn’t even have to be a geek, hi-school alone already qualified.

So I have been looking at Fermi for the last 60 years and finally the little green men, this time on sabbatical in Santa Clara, came out with it. Not a sex toy this time, though undeniably sexy, it is still somewhat different from what I anticipated half a century ago – no I am lying, I am not that old yet.

To get back to the topic at hand, we are looking at nVidia’s Fermi graphics processor / general purpose graphics processing unit and, truth be told, we have been hearing about it almost as long as about the Fermi paradox. But it is finally here.

What happens if one takes two RADEON 5870 non-reference designs running at different core and memory frequencies, not to mention different memory configurations, and runs them in CrossfireX mode? The candidates in this case are the ASUS MATRIX 5870 Platinum Edition and the PowerColor PCS+ RADEON 5870. The MATRIX runs its core / memory at 900 / 1200 MHz and features the full-blown 2 GB local frame buffer supported by the design, whereas the PCS+ is clocking the core at 875 MHz but pushes the memory frequency up to 1250 MHz for a whopping 5000 Mbps. The reason we are asking is simply that the market often enough dictates the purchase and whatever is on sale may or may not be compatible with another card that is already in the system. Theoretically, there shouldn't be much of a problem but reality bites more often than not.

The next question is about scaling in crossfire mode. How much performance increase is delivered by the two cards in tandem and in which applications? We ran an entire gamut of synthetic and real world gaming applications to have the answers.