News of the day (January 4, 2010)

 GeForce 300Ms, NVIDIA renames again!
  Posted on 1262622625 by Marc

NVIDIA has launched no less than 7 new notebook GPUs, from the GeForce GTS 360M to the GeForce 305M and including the GT330M. Unfortunately, NVIDIA is simply hiding the GT215, 216 and 218 behind these new names, which were announced 6 months ago!

We therefore get the following:

- GeForce GTS 360M : GTS 260M + 4%
- GeForce GTS 350M : GTS 250M
- GeForce GT 335M
- GeForce GT 330M : GT 240M + 4%
- GeForce GT 325M
- GeForce 310M : G210M
- GeForce 305M

Some products are strictly the same as the previous 200M range, while others benefit from a slight increase in performance. The range has also been extended, with, in particular, the intermediary 335M with 72 CUDA Cores, which wasn’t in the previous range. Note that NVIDIA is only now highlighting DirectX 10 support and not DirectX 10.1 support, although these GPUs are identical to the previous ones and do therefore have DX10.1 support.

 1st tests of the Vertex 2 / SandForce SSD
  Posted on 1262619471 by Marc

AnandTech has been able to get its hands on a pre-version of the Vertex 2 Pro SSD, an early vintage of the collaboration between OCZ and SandForce announced last November. This SSD won’t be commercially available before March. The first surprise is that SandForce doesn’t use any external DRAM but just the cache built onto the controller. The SSD has a large 90mF capacitor designed to make sure no buffered data is lost should there be a power out.

Performances in practice are very high with the Pro version, using the highest end controller, reaching sequential reads of 265 MB/s and sequential writes of 252 MB/s. For random reads of 4 KB blocks we’re at 51.3 MB/s (as against 38 MB/s with Indilinx and 64 MB/s with Postville) and writes at 50.9 MB/s (compared to 13.6 MB/s with Indilinx and 39.1 MB/s with Intel). In PC Mark Vantage this Vertex 2 Pro is 11, 18 and 46% faster than the Postville 80 GB and 160 GB and the Indilinx 128 GB.

To achieve such results, SandForce has developed DuraWrite technology which will lower write amplification under 1x! Remember, the write amplification is the ratio between the volume of data written on the flash memory (flash write) and the volume of data to write requested by the system (host write). Intel has highlighted the fact that it kept write amplification on its SSDs down to 1.1x. SandForce has announced 0.5x and says, for example, that installation of Vista and Office requires a total of 25 GB (host writes), but that in practice only 11 GB are written in Flash (a write amplification of 0.44x).

How is this possible? Obviously this isn’t bit-for-bit data storage and SandForce must have implemented compression and deduplication algorithms within the controller. Of course this technology will work less well with compressed data and write speeds for compressed data will drop to 145.9 MB/s according to AnandTech’s tests, which is in any case not bad.

Reducing flash writes is of course beneficial both in terms of write speeds and reliability as the number of write cycles is critical in terms of how long flash memory lasts. So as to limit wear on the flash, SandForce also includes additional extra internal information similar to RAID5 (without going into more detail) and makes it a point of honour to have a lot of additional flash: the first SSDs will be the 50, 100, 200 and 400 GB versions but they will actually have 64, 128, 256 and 512 GB of flash memory. These are the enterprise versions, with the standard versions having less additional space.

All in all a very promising controller then, although we would have liked to see SATA 6 Gbits/s support.

 VIA launches USB 3.0
  Posted on 1262613107 by Benoit

VIA is the 1st to launch an USB 3 hub controller. Using a 80nm process, the VL810 allows a single USB port to be used by up to four high-bandwidth USB 3.0 devices simultaneously. It will be integrated in external hub as well as in hub integrated in screen for example. We should see a demo of the VL810 at CES.

 News of the FTC case and Intel compilers
  Posted on 1262612688 by Marc

The FTC’s proceedings against Intel in December were important for a number of reasons. Among them, the part concerning compilers is the least known by the general public.

In addition to processors, Intel also sells a range of compilers. They have a reputation for high performance and are also generally faster with Intel processors, so much so that they're sometimes preferred to compilers from third parties (Gnu, Microsoft) by those who prioritise rapidity above all else.

Like all other compilers, the Intel ones offer various compilation options. You can therefore target specific levels of instruction (LA32, SSE, SSE2, SSE3, etc.) but a more elegant option allows you to generate executables that integrate different code that is executed according to the processor’s features.

And here lies the rub. With this option on the Intel compilers, the executable will first check to see if the processor is an Intel one before executing the optimal code for the features of that particular processor. When it isn't an Intel processor, the default code is used whatever the instruction set supported by the CPU!

Today the default code is SSE/SSE2 code, but a few years ago it was non-vector code (pure x86) that was therefore executed on the Athlon 64s! The current situation is less problematic than it was in the past then but still, a programme compiled with this option won’t, for example, execute SSE3 code on a Phenom II.

Of course, there are alternatives, the first of course being not to use Intel compilers. Where this isn’t possible, you can compile your exectuable in several versions and have a programme that launches the right version upstream. Better still, you can replace the Intel CPU with another that works whatever the CPU manufacturer (page 125 onwards of this document) !

As part of its enquiry into Intel’s practices, the FTC, alerted to the issue, has come out against Intel compilers as uncompetitive behaviour, qualifying the compiler itself as defective ("defective compiler", page 21 of this document). As a result it has requested that Intel put a compiler from which this code is excluded at the disposition of developers and take responsibility for any recompilation of programmes compiled with the “defective” versions.

Of course, if the FTC’s request was seen through, it wouldn’t transform an Athlon II into a Core i7, but in terms of programmes compiled with Intel software, there could be a gain on rival CPUs, which is still a bonus.

 Is EVGA relaunching Skultraill?
  Posted on 1262610379 by Benoit

EVGA has allowed the photo of a quite particular motherboard to surface. Other than the brand, you can clearly see two 1366 sockets, each accompanied by 4 memory slots. This mystery card is likely to be presented at CES.

The concept isn’t a new one and is reminiscent of the shortlived Skulltrail platform launched at the end of 2007 based on two Xeon E5300s and a 5000X chipset. The question that arises (other than the utility of this type of configuration for the general consumer) is therefore to do with this card’s architecture. Unless EVGA is presenting an entirely new solution, it will probably have to use Xeons, the only socket 1366 processors that are officially compatible with a bi-CPU solution. That said, if the Core i7 9xxs have two active QPI links instead of one, this card should function with these processors.

In any case, the card will be expensive but ought to break new records for the various existing benches as we’re accustomed to seeing overclocking orientated cards from EVGA.

 The Radeon Mobility 5xxxs are on their way
  Posted on 1262604641 by Benoit

ATI was the first to offer a DirectX 11 range for desktops with the Radeon 5xxx range. With good performance and lower energy consumption, they are particularly appreciated as NVIDIA has, as yet, no equivalent to offer. Nevertheless, when it comes to laptops, neither firm currently has anything more than DirectX 10 and 10.1 chips.

This is all about to change and we should soon see the arrival of the first DirectX11 solutions for notebooks. Asus has just announced the G73JH-A1, a gamer laptop that is available for pre-release order on the other side of the Atlantic. Other than its Core i7, it also has a Radeon Mobility 5870 with 1 GB of GDDR5. While we don’t yet know the exact spec of this card, it should surpass the current 4870 Mobility, while keeping energy consumption reasonably low. At the same time, there is also word of the launch of the 5850 and 5830 Mobility at the same time, as well as lower end GPUs.

In the absence of official specs, this news should be taken with a pinch of salt. We should get more information on this shortly at CES.

 Core i3/i5, also on laptops!
  Posted on 1262596681 by Marc

The Core i5 and i3 launched for desktops today, have also been launched for laptops at the same time. Known under the name Arrandale, this series is made up of no less than 11 distinct references, to be added to the 7 for desktop:

Once again, Intel has managed to complicate the naming process. Thus in the laptop range we find the following:

- Core i7-9xx, 8xx, 7xx: quad 45nm
- Core i7-6xx: dual 32nm, 4 MB of L3
- Core i5: dual 32nm, 3 MB of L3
- Core i3: dual 32nm, 3 MB of L3, without Turbo

To give a minimum of coherence within its range as well as between desktop and laptop, it would have been much more logical not to add the Core i7 label to any of the Arrandales. Unfortunately, once again, marketing has won the day over reason.

TDPs vary between 35W (versions M), 25W (LM) and 18W (UM) and you can see that the accent has been placed on Turbo mode, particularly on the 18W versions. Of course, we’ll have to wait and see what gains it gives, given that it will be relatively limited due to the lowish TDPs.

With laptops, the Arrandale will benefit a good deal from the integration of the northbridge on the same packaging as the processor because this will allow it to reduce the space required on the motherboard. Arrandale offers two features in addition to what you find on Clarkdale:

- Intel Switchable Graphics, that allows you to switch from the IGP to an an additional GPU on the fly, in 7 or Vista
- Intelligent Power Sharing, a Turbo mode with support via a pilot that shares the increase in clock between the CPU and the GPU depending on load

Given the excellent results in terms of the performance / energy consumption ratio offered by Clarkdale in our test, there can be little doubt as to the success of Arrandale for laptops.

Copyright © 1997-2014 BeHardware. All rights reserved..