Although we won’t see Intel’s Nehalem on the market until 2009, the guys over at AnandTech used their dark forces and got hold of not one, but two of Intel’s next-gen CPUs. As you might know by now, Nehalem will be featuring up to eight cores and will be capable of dealing with up to sixteen simultaneous threads, thanks to an improved version of the Hyperthreading architecture.

Unfortunately, Intel didn’t want Anandtech to fully assess the Nehalem CPU, sending some slightly-crippled components: the PCIe slot didn’t work, Intel have locked down the chips to less than their full performance, plus there were some other downgrades that affected the overall results. Anyway, the Anandtech team managed to run benchmarks on a 2.66GHz Nehalem CPU and it turned out that the system could perform pretty well, in spite of those downgrades.

The benchmarks results revealed total performance gains ranging from 20% all the way up to 50% over the current high-end Penryn CPU and that surprised the Anandtech teach: “We’ve been told to expect a 20 - 30% overall advantage over Penryn and it looks like Intel is on track to delivering just that in Q4. At 2.66GHz, Nehalem is already faster than the fastest 3.2GHz Penryns on the market today.”

The Anandtech team also measured the CPU’s power consumption, which turned out to be increased only by a small amount, but this could be further improved by launch time. Apparently, AMD is going to have a hard time once again.

Supercomputers prove to be very helpful in scientific experiments. A supercomputer is usually made of clusters upon clusters of smaller interconnected computers that work in parallel. If your current desktop PC has 2, 4 or 8 cores, supercomputers may combine more than 1,000 multi-core CPUs to provide enormous number-crunching power. However, there’s another possibility to constructing a super computer. Instead of using many CPUs, one may use the GPUs found in graphics cards. It is known that GPUs are superior to CPUs when it comes to complex calculations. That’s why Nvidia came up with CUDA, a technology that allows you to use up to 4 graphics cards that would provide roughly the same performance as a supercomputer cluster consisting of hundreds of PCs.

The ASTRA research group, part of the Vision Lab of the University of Antwerp, decided to extensively use the CUDA technology to develop new computational methods for tomography. In order to achieve this, they put together the FASTRA system containing 4 MSI 9800GX2 graphics cards (8 GPUs in total|) which are said to deliver the same performance as more than 300 Intel Core 2 Duo 2.4GHz processors. Of course you don’t see much improvement in games while using quad-SLI configurations, but that’s because the graphic drivers are not quite fully developed and the game engines are not optimized for such configurations. Otherwise, the CUDA does tomographic calculations very efficiently and consumes a lot less power than a supercomputer cluster.

Here is a list with the components used for the $4,000 FASTRA system:

• AMD Phenom 9850 processor + Scythe Infinity CPU cooler
• 4x MSI 9800GX2 graphics card
• 4x 2GB Corsair Twinx DDR2 PC6400 memory
• MSI K9A2 Platinum motherboard
• Samsung Spinpoint F1 750GB HDD
• ThermalTake Toughpower 1500W Modular PSU
• Lian-Li PC-P80 Armorsuit case
• Windows XP 64-bit

The biggest problem of the system is that these four dual-GPU graphics cards are cramped together and generate quite a lot of heat. The FASTRA uses aircooling and with the sidepanel removed the GPUs run at 55°C in idle, 86°C under full load and 100°C under full load with the shaders 20% overclocked. Hey guys, how about using some water cooling system if not liquid nitrogen? Also check this demo clip for more details:

A couple of days ago I mentioned the fact that Nvidia is planning to annihilate the CPU market. It’s not clear why they intend to do this, but it looks like the Santa Clara company got Intel and AMD figured out and the green guys scheme a nasty takeover in small steps. Why small steps? Because their first targets are handhelds and notebooks.

According to Engadget, Nvidia readies the Tegra APX 2500 for handhelds , while the future notebooks will meet the CSX 600/650 CPU.

Tegra APX 2500 is an ARM11 based system-on-a-chip. Presumably, Nvidia wants to obliterate Intel’s Atom with this one. The system-on-a-chip measures 144mm, and can easily manage 720p encoding/decoding tasks at 14MBps, thanks to a 500+ MHz clock. Although it supports most of the current-gen features in GeForce GPUs, the Tegra APX 2500 will actually perform like a shrunk-down GeForce 6600. Still, the GPU part will be able to deal with AA and AF settings and will support the latest OpenGL ES 2.0.

CSX 600 and 650 are a bit more complex. However, as these aren’t x86-based chips, they can only run Windows CE, so if you plan to play around with XP-only programs and games, you might have to wait for the next-gen. The chip itself has 256K of L2 cache and can be die stacked to keep the footprint small. The CSX chips run at 700-800MHz and are said to support 1080p at 24FPS. Atom can do all these plus it doesn’t have the FPS limit. Now I wonder if Nvidia has some other aces in its sleeve, because with these two chips Nvidia has pretty much nothing on Intel.

Nvidia is supposed to showcase these two at Computex this year and maybe we’ll hear some info on price and availability in the weeks to come.

Intel’ Core 2 Duo generation rapidly nears its end, but we can’t deny that this one, as with the previous Core Duo generation, really restored Intel’s hegemony on the CPU market. It’s clear that AMD has cheaper solutions as far as quad-cores are concerned, but Intel is trying to keep the balance by slashing prices once in a while. But according to some market studies, quad-cores have not yet been widely embraced, as the majority of PC users still rely on dual-cores. Speculating this aspect, Intel now announces that one of last models in the Core 2 Duo series is going to be a cheap but fast dual-core.

Known as Core 2 Duo E8600, this CPU will end up the fastest dual-core with its out-of-the-box 3.33GHz clock and 6MB cache. It will be available in Q3 for only $266 price, but that is the announced wholesale price.

Intel surely reached the cap with the 3.33GHz clock and since these are the last months of the Core 2 Duo generation, it’s fair to say that we won’t see any higher clocks until the advent of the new generation. However, we can still overclock these babies and a higher standard clock automatically means that the overclocking potential is increased, and breaking the 4GHz barrier now becomes a walk in the park.

Maybe Intel could push the clocks some more with the next-gen, but the truth is it really isn’t necessary as Phenom with its current clock speeds still lags behind.

As previously reported, AMD plans to launch the Swift CPU (aka Fusion) in the second half of 2008. Now we know for sure that Swift is going to be exclusively oriented towards notebooks and mobile devices. However, AMD does not want to limit the Fusion architecture to the mobile world and at some point in 2010, a Fusion CPU+GPU is scheduled to hit the desktop market as well.

As related to notebooks, Intel announced that their Nehalem CPUs for notebooks and desktops will be commercialized in early 2009, but mainstream products based on the Nehalem CPU won’t be available until Q4 2009. We assume Intel has some problems with the memory controller integrated directly on the dye and the 45 nm architecture and this could be causing the delay for the mainstream market.

AMD, on the other hand, seems to have a good plan introducing the Fusion chips for notebooks first and foremost. The mobile market will clearly benefit from such a chip as a CPU with an integrated GPU will be cheaper to produce, though the process is a bit more complicated.

To sustain this point, Dave Orton, ATI’s former CEO, claimed a while ago that the integration of two or more chips significantly cuts costs. Apparently, a current Northbridge chip coupled with a cheap discrete graphics card is more expensive than an all-in-one Fusion chip that integrates a Northbridge with a GPU. From the manufacturer point of view, that is.