NVIDIA launches another GeForce GT 440 Memory interface at 12.8 Gbps and even 20 Gbps! Intel bug: what SATA to use? Sandy Bridge chipset bug: details

Février 2011 L M M J V S D 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 You can read previous news using our archives. [ Print all news of this day ]

<< Previous day's news January 31, 2011

Next day of news >> February 2, 2011

NVIDIA launches another GeForce GT 440   Posted on 01/02/2011 at 16:14 by Damien

	After the launch of the GeForce GT 440 a little over three months ago limited to OEM customers, NVIDIA has just quietly unveiled a second GeForce GT 440 which will be available in retail. To save the poor consumer from any possible shock that the release of an unequivocal product in the entry-level jungle may engender, not one but two GeForce GT 440s are going to hit the stores… and they will have nothing in common with the previous version. While the GT 440 reserved for OEMs is based on a cut down GF106 GPU, with 144 processing units and a 192-bit memory bus, the new GT 440s make do with the GF108, the entry-level GPU of the family. On the menu, 96 processing units and a 128-bit memory bus. The GeForce GT440 is none other than an overclocked GeForce GT 430. To recap, the small GF108 has just 4 ROPs, the units charged with writing pixels to memory, in spite of the presence of a 128-bit bus, while the other GPUs from this generation have 8 ROPs per 64 bits, against 4 ROPs per 64 bits for the GPUs on the previous generation. This increase is designed to allow maximum exploitation of GDDR5 memory. With 4 ROPs per 128 bits, the GF108 is therefore a departure from this thinking. We imagine that here the GF108 was designed to feed a 64-bit memory bus using DDR3 and that this bus has been extended to 128 bits, not for performance reasons but so as to be able to offer cards with 1 and 2 GB of memory more easily. Overall, this isn’t a problem because Fermi architecture is rather limited upstream in terms of pixel throughput and antialiasing performance (in part linked to the number of ROPs) is low for this sort of entry-level GPU. It is however astonishing to see such a GPU with GDDR5 memory, as is the case for one of the new GeForce GT 440s, because apart from increasing its price, it's unlikely to be able to benefit from the GDDR5. Don’t go for the GeForce GT 440s with GDDR5, unless they cost the same as the DDR3 models. If necessary, you’ll be able to distinguish the OEM versions of the GT440 by looking at how much video memory there is. These will have 1.5 GB or 3 GB while the new cards will be equipped with 512 MB, 1 GB or 2 GB.

Memory interface at 12.8 Gbps and even 20 Gbps!   Posted on 01/02/2011 at 15:28 by Marc

	Although RAMBUS is well-known for its court actions, it is nevertheless still active in research. Rambus will be demoing several innovations in the framework of its ‘Terabyte Bandwidth’ project at DesignCon 2011 which takes place this week. ‘Terabyte Bandwidth’ aims at a bandwidth of 1 TB/s with a single memory controller. Differential signaling (XDR) can now be extended to 20 Gbps and single-ended memory signaling (DDR/GDDR) to 12.8 Gbps. For comparison, 1st generation XDR memory could extend to 7.2 Gbps up to now in best case, against 9.6 to 12.8 Gbps for XDR2. DDR3-1600, the most common, functions at 1.6 Gbps, while the fastest GDDR5 used on graphics cards is at 6 Gbps. This 20 Gbps throughput, attained via a 32X data rate, means that on a single XDR 2 DRAM 32 bit chip you’ll be able to get 80 GB/s. It remains to be seen if these technologies will be used in general consumer products.

Intel bug: what SATA to use?   Posted on 01/02/2011 at 10:50 by Marc

	Following the Intel announcement of a bug on series 6 Intel chipsets that can cause failure of the SATA 3 Gbps ports, we advise you to use only the first two chipset ports. These are the ports which can also function as SATA 6 Gbps ports. Many motherboards also have additional SATA 6 Gbps ports managed by an additional Marvell chip. While they are also compatible with SSDs and HDD SATA 3 and 1.5 Gbps, optical drives don’t necessarily work on this controller. We advise you to connect your peripherals as follows: - main HDD/SSD on an Intel SATA 6G port - optical drive on an Intel SATA6G port - additional HDDs/SSDs on the Marvell SATA6G ports To find the ports, refer to your motherboard manual. Here are a few examples on some P67 boards: ASUS P8P67 LE - SATA6G_E1: Marvell SATA 6 Gbps - SATA6G_1 and 2: Intel P67 SATA 6 Gbps - SATA3G_1 to 4: Intel P67 SATA 3 Gbps Use the Intel SATA 6G ports (very light blue), then the Marvell SATA 6G port (dark blue). Avoid the Intel SATA 3G ports (blue, to the left of the chipset cooler). ASUS P8P67, P8P67 Pro/EVO/Deluxe - SATA6G_E1 and 2: Marvell SATA 6 Gbps - SATA6G_1 and 2: Intel P67 SATA 6 Gbps - SATA3G_3 to 6: Intel P67 SATA 3 Gbps Use the Intel SATA 6G ports (white) as a priority, then the Marvell SATA 6G ports (dark blue). Avoid the Intel SATA 3G ports (light blue) Gigabyte GA-P67-UD4 (valid on UD3P/UD3R) - SATA3_0 and 1: Intel P67 SATA 6 Gbps - SATA2_2 to 5: Intel P67 SATA 3 Gbps Use the Intel SATA 6G ports (white) as a priority. Avoid the Intel SATA 3G ports (black). On the UD3, the SATA3G ports are light blue. Gigabyte GA-P67-UD7 - SATA3_0 and 1: Intel P67 SATA 6 Gbps - SATA2_2 to 5: Intel P67 SATA 3 Gbps - GATA3_6 to 7: Marvell SATA 6 Gbps Use the Intel SATA 6G ports (white) as a priority, then the Marvell SATA 6G ports (black, to the right of the whites). Avoid the Intel SATA 3G ports (black, to the left of the whites). ASRock P67 Pro3 - SATA3_0 and 1: Intel P67 SATA 6 Gbps - SATA2_2 to 5: Intel P67 SATA 3 Gbps Use the Intel SATA 6G ports (white) as a priority. Avoid the Intel SATA 3G ports (dark blue). ASRock P67 Extreme 4 - SATA38M1 and 2: Marvell SATA 6 Gbps - SATA3_0 and 1: Intel P67 SATA 6 Gbps - SATA2_2 to 5: Intel P67 SATA 3 Gbps Use the Intel SATA 6G ports (white beside the dark blues) as a priority, then the Marvell SATA 6G ports (white furthest on the right). Avoid the Intel SATA 3G ports (dark blue). On the Extreme 6, follow the same principle with four Marvell SATA 6Gs instead of two. MSI P67A-GD55 (valid on the P67A-C45) - SATA_1 and 2: Intel P67 SATA 6 Gbps - SATA_3 to 6: Intel P67 SATA 3 Gbps Use the Intel SATA 6G ports (white) as a priority. Avoid the Intel SATA 3G ports (black). MSI P67A-GD65 - SATA_1 and 2: Intel P67 SATA 6 Gbps - SATA_3 to 6: Intel P67 SATA 3 Gbps - SATA_7 and 8: Marvell SATA 6G Use the Intel SATA 6G ports (white beside the blacks) as a priority, then the Marvell SATA6G ports (white to the left of the blacks). Avoid the Intel SATA 3G ports (black).

Sandy Bridge chipset bug: details   Posted on 01/02/2011 at 10:36 by Guillaume

	Through Anandtech Intel have given additional technical details on the bug affecting chipsets designed for Sandy Bridge processors. The fault is reported to come from a transistor in the PLL clocking tree which serves the 3 GB/s Serial ATA controller. The transistor in question is reported to have a particularly thin gate oxide, which means it can be turned on at very low voltages. However, here the transistor is reported to have been exposed to too high a voltage. The resultant leakages in current caused by turning the transistor on apparently increase over time and cause the deterioration previously described. To correct the problem, Intel says that it has simply deactivated the transistor in question (by cutting the power on one of its upper metallic layers). This transistor is apparently part of a block that’s no longer needed in the 6-series design and is a “remnant of an earlier design”. Chipsets are in fact based on independent blocks that are validated and reused from one stepping to the next. Removing the voltage to this transistor (and those around it?) shouldn’t have any impact on functionality. More surprisingly, while Intel explains that these blocks are constantly re-used, the problem is apparently only with the B-stepping of Cougar Point chipsets (the chips that you find on motherboards in stores are B2-stepping). A-stepping, samples for which were released last year by Intel and its partners, didn’t have this fault. These explanations raise a certain number of questions, beginning with the real utility of the faulty transistor, which does nevertheless seem to be part of a useful block (to recap, its’s part of the Serial ATA PLL circuit). Wasn’t it turned on in previous steppings, or is there a specific error on the B-stepping? Whatever the case may be, given that the returns rate is estimated by Intel to be between 5 and 15% over the life of the machine (estimated at 3 years), it’s surprising that the Intel validation procedures went so wrong upstream. Most resellers have suspended sale of the affected motherboards. We’re still waiting to hear from manufacturers about the procedure for replacement, which will depend on deliveries of the new Intel Cougar Point stepping. Best case scenario is that they will be shipped at the end of February. In the meantime as we explained yesterday to owners of the affected motherboards, we advise you to use the SATA 0 and 1 ports (cabled at 6 GB/s and not affected by the bug) on the motherboard and avoid ports 2, 3, 4 and 5 and either to use any additional ports available on your motherboard (handled by another chip) or a Serial ATA extension card.