IDF 2008: Turbo Mode and the Nehalem   Posted on 20/08/2008 at 18:35 by Damien

	Intel has unveiled a little more on the Nehalem’s Turbo Mode. This technology consists of putting some of the cores at rest while increasing the frequency of active cores. In this way, if CPU use is high but concentrated on 1 or 2 cores when the application(s) is not adapted to multiple cores, performances will be boosted. In the beginning this was a technology that appeared in the mobile world in order to maximize performances for any given thermal envelop. For desktop PCs, its interest is less compared to an entire CPU’s higher frequency but if it is functional why not integrate it? Compared to current technology, the Nehalem should provide support with less intervention by the OS but we do not know anymore as Intel did not want to go too much into detail. Neither did they elaborate on the maximum authorized frequency for this "overclocking" even if on some samples that are in current circulation, it consists of increasing the multiplier by one unit in going from 4 to 2 active cores and then another unit by going to a single active core. For example, with a Nehalem running at 2.93 GHz with a multiplier of 22, each increase consists of a supplementary 133 MHz. Therefore a Nehalem at 2.93 GHz will go as high as 3.2 GHz in load on a single core. However despite everything, Pat Gelsinger gave us one other detail: Extreme versions of the Nehalem will still have an unblocked multiplier as much in base frequencies as in those in Turbo Mode. It will therefore be possible to specify maximum levels in the bios.

Nvidia also needs Transmeta   Posted on 11/08/2008 at 20:37 by Nicolas - source: Transmeta

	Transmeta has announced having sold a license to Nvidia for Long Run and LongRun2 as well as for other technology and intellectual property. The « package » should help the company with the chameleon logo to develop products that consume less energy even if we do not yet know exactly which ones. The cost of this transaction was a rather significant $25 million although compared to the Geforce’s $800 million in profits last year it may seem small. Note that the agreement was unveiled at a moment when Nvidia needs to set things straight after the GeForce 8M overheating problems and also has to rapidly acquire a maximum of knowledge and know-how if it wants to give a chance to the Tegra’s successors. You may recall that Transmeta made a name for itself at the start of the decade with Crusoe and Efficeon processors capable of functioning with x86 applications via Code Morphing Software which translates x86 into VLIW (Very Long Instruction Word) instructions; however, it had to throw in the towel due to stiff competition, especially from Intel. Moreover, the latter pays $20 million per year for the right to use some of its technology while AMD, which invested $7.5 million in the company, should also benefit in this domain. In short, the three giants seem to be equally armed – at least in this area...

Nehalem = Core i7   Posted on 08/08/2008 at 11:10 by Marc

	Thanks to Expreview, we now know the official name of processors based on Nehalem architecture: Core i7. There will be two associated logos, a black one that will go on Extreme versions and another blue one for standard processors. The official launch of these brands is planned for August 11th with processors arriving in the first quarter.

Performances of the 45nm Phenom   Posted on 07/08/2008 at 15:10 by Marc

	Hardspell has published a preview of the « Deneb » which you may recall is the 45nm version of the Phenom. Set at 2.3 GHz and equipped with an L3 cache of 6 MB, as planned the processor has lower power consumption. On the ATX12V, they obtained 32.4W at rest and 57.3W in load versus 39.4W at rest and 104.1W in load for the Phenom X4 9650. Compared to previous first tests of this kind, there is a much more significant gain in load which is a good thing as it comes closer to Intel 45nm solutions. In terms of performances, the gains due to the transition from 2 to 6 MB of L3 cache cannot be ignored: - Fritz Chess : 4.7% - Wprime : 5.6% - Pov Ray : 17.2% - MainConcept H.264 : 3.6% - 3DMark Vantage (CPU) : 6.4% - Quake Wars : 8.1% - Crysis : 9% So here we have some promising results. We are now impatiently awaiting the arrival of these processors in their definitive version which will also give the Santa Clara giant something to think about.

Atom vs. Nano: room for improvement   Posted on 30/07/2008 at 16:49 by Nicolas

	PC Watch has published a comparison of the VIA Nano L2100 (1.8 GHz-TDP of 25W) and Intel Atom 230’s (1.6 GHz-TDP of 4W) performances in addition to those of the reference chipsets from their respective manufacturers as other components were identical on the two test platforms. Overall, the Nano clearly has higher performances than its rival and this even in applications optimized for HyperThreading which enables taking advantage of the full potential of Intel’s CPU. However, things are more complex if we consider each solution and their power consumption separately. On the one hand, there is the Nano L2100+CN896 combo that needs 58W in full load while the Atom 230 combined with a 945GC is "content" with 43W. For the former, another model probably could have done better and this without sacrificing performances too much (we were also surprised by the fact that the VX800U and its 3.5W was not used). In the latter, a better adapted chipset (i.e. conceived specifically for the Atom and not the Pentium 4) would have enabled to go much lower. In short, the potential indeed seems to be there on both sides but a little more effort is needed to offer more coherent platforms... The VIA Nano platform, there is a PCI-E x16 port.