skip to main |
skip to sidebar
Since PCI-Express and it's remarkable flexibility made a debut in the computer world, much has changed among graphical subsystem solutions - one of the key words today is scalability.
With the introduction of the new CrossFire X technology AMD will be able to supply solutions based on 2, 3 and 4 GPUs running together, with the aim of supplying the fastest graphics subsystem to users and OEMs everywhere. This is what multi-GPU solutions are all about: providing the highest possible performance with each hardware generation.
While this is true for benchmarking because it does break records and helps manufacturers show their technological superiority, but what about actual benefits, without trade-offs?
The problem with embracing these solutions is that most people are actually also embracing the possibility of making their gaming experience worse. This happens by the means of the added processing overhead, associated with the increased processing power required to handle synchronization and load balancing(in case of split frame rendering), resulting in lower framerate in worst case scenarios.
Two good examples of this are the reviews to the CrossFireX technology from HardOCP and PCPerspective; take also the HD 3870X2 review from Xbitlabs.
They show a general increase in maximum framerate but decreases in the minimum framerate - which is something both AMD and Nvidia should be focusing in increasing, at least for the sake of customer satisfaction.
I am citing illustrative examples, some games perform better, some worse; some are worse with three GPUs, some with four. It's a driver issue and will probably be worse with a slower CPU, unable to balance or synchronize the graphics cards properly. While I admit that a very fast processor is the way to go before considering this approach to increase performance, at which point is the CPU able to cope with both the added overhead, increased focus on processing more realistic physics and artificial intelligence? A big chunk starts missing from the processing power you had before.
Drivers are the most important part of a multi-GPU solution and will provide good performance improvements if done right, the problem is that's not the case most of the time. Most of the drivers provided require specific optimizations for a given game or they will perform worse or the same.
We also have the problem known as micro-stutter, a consequence of all these problems. From research, these problems are inherited from the use of AFR(alternate frame rendering) instead of the use SFR, the far better approach to solving the problem, yet far more difficult to implement, which sometimes gets lost along the way.
Ever since their introduction I've actually praised Multi-GPU solutions because, in theory, they offer a good way to upgrade your graphics system without throwing your existing one "to the trash can". This was assuming that all caveats were dealt with, which is something that hasn't been happening for a long time.
At which point, after spending so much money to get the best of the best, do you have to start doing your own benchmarks to check if you're really getting your money is worth or if you would just been better off with one graphics card?
In situations where large monitors and resolutions come into play, a multi-GPU solution is probably worth the extra investment, at least to have decent framerates most of the time, but it is still questionable.
A place where multi-GPU solutions will shine is when GPU physics come into play. Not using two graphics cards to render will actually decrease the overhead and with the GPU helping out with physics calculations it will surely help in the most processing-bound situations, therefore delivering a better gaming experience.
Since we as users have already lost much of the control we had on our computers by relying too much on the providers AMD and Nvidia and their drivers, has the PC gaming industry lost it's biggest call? Today we are almost almost forced to buy a multi-GPU solution to get the best of the best. When will every PC gamer just get sick and tired of this whole game and move to greener, console, pastures? That day has been farther away.
After some user feedback regarding the previous post about HT3.0 issues with the A12 revision from AMD's 780G, I contacted some people to set the record straight.
The issue will only apply to the Phenom 9100e and 9150e processors, which will run with an HT3.0 link at 1.6GHz.
With these CPUs, AGESA version 3.1.3 and up will automatically change the link to HT2.0 1Ghz. From what was disclosed now, all other will be able to use an HT3.0 link up to 2.6GHz, depending on the maximum speed supported by the CPU. This won't change in the B3 revision Phenom 9150e and Nvidia's MCP78 won't have the same issues with these low-end processors.
Some reviewers actually checked for the performance difference with both HT2.0 and 3.0 an it isn't good. For maximum performance in both integrated and hybrid modes, don't use the Phenom 91x0e or a K8 based processor, like the AMD Athlon X2, since both will have these performance losses.
I won't say you won't have problems later with other processors, but for now the issue has only been reported with these models. AMD had it pretty bad with the TLB issue: at first it was just an issue with the higher-clocked models but, after a while, a performance degrading fix was used for each and every B2 Phenom.
As posted before, the A13 revision of the chipset will fix these issues.
AMD's latest and greatest integrated chipset, will end up faster than it currently is.
Tomorrow is the day that AMD will be officially releasing the new AMD 780G(RS780) chipset worldwide, after it was released only in China a while ago. Turns out that the Chinese got "broken" samples of the chipset, namely the A11 revision of the chipset; we're getting the A12 tomorrow, which also isn't quite ready, yet.
The first commercial A11 revision was available very soon for reviewers and motherboard manufacturers, but it lacks full UVD support, which offloads video processing tasks from the CPU to the graphics card; AMD branded it as the RS780C and sells them cheaper.
The "fixed" A12 revision of the chipset is already being sold in China, it features full UVD support and has an interesting performance, even more when overclocked.
The delay to launch it worldwide, earlier, was rumored to be because of the lack of driver Hybrid Crossfire support - which proved mostly true: the embargo on the technology was lifted today.
What AMD isn't telling it's potential customers is that there is a major issue with the current A12 silicon: it doesn't support HT 3.0. HT3.0 is marketed as a big feature of Phenom CPUs and the AM2+ socket because it provides support for a maximum speed of 2.6GHz, providing more bandwidth than the HT2.0 link which is limited to 1.0GHz, or 2000MT/s.
AMD guidance suggests that the HD3200 core embedded in the 780G chipset will benefit a great deal from the added bandwidth provided by HT3.0 - which currently clocks at 1.8GHz for current Phenom processors, an 80% increase. Since the memory controller of the system is in the processor itself, the 780G will rely on the the HT link to move data to and from the RAM, hence limiting performance in case of a slow link.
The lack of HT3.0 support will also affect overall system performance, although not by as much as it will for the integrated graphics core - users of dedicated graphics will not be as affected. The Hybrid Crossfire technology will certainly also improve both due to the added performance of the integrated core and the improvement in communication with the discrete card.
The reason why most testers aren't noticing this workaround is because AGESA, a part of the BIOS, is automatically reducing the link to 1GHz HT2.0 when it detects an A12 revision of the chipset. Nvidia's upcoming, competing, solutions will offer full HT3.0 compliance, with production starting this month.
Mass production of the new A13 silicon is only estimated to start in May. Until then the GeForce 8200 chipset from Nvidia will have a good chance of competing with the 780G for the title of top performing integrated graphics.
UPDATE:
The issue only applies to Phenom 9100e and 9150e processors: read the follow up
Tags: Asus Eee PC ssd flash shrinking problem
The first mass produced notebook with flash memory as a primary storage device takes its first steps.
It's been the fear of tech forums everywhere, solid state storage failure. Commercial grade flash memory chips typically have a threshold of 100,000 write/erase cycles per block, after which it could become unusable due to an increased error count while reading. This is a mean value, some fail after, some before. Reads also take a toll on that value, but they are almost negligible when compared to an erase cycle.
This has raised the question as to whether or not NAND is "ready" for high write operations. Is flash memory reliable enough to replace magnetic media?
The ASUS Eee PC is the first multi-purpose computing device available en masse that employs storage based solely on Flash memory, commonly known as a solid state drive (SSD). Being a pioneer, it will be an excellent test subject and a testimonial to the durability of flash based drives for use as an HDD replacement. Other flash based devic es don't have the same kind of usage patterns and therefore aren't good for this purpose.
Since the Eee PC was released in November, one would think it is still quite early to expect some relevant data, but turns out that isn't true. Some users have already reported unusual behavior from their drives; some claim the main volume sizes have shrunk by as much as 500MB. All cases revolve around Eee PCs that feature SLC Hynix HY27UG088G NAND chips.
Hynix advertises that its memory performs wear-leveling on-chip -- a technique used to even out memory cell usage so the flash memory does not write particular cells more than others. Typically, if bad blocks are detected in this wear leveling, they are removed from the address table, resulting in fewer blocks in the drive volume.
ASUS seems confident the isolated reports are not systemic. In conversation with Kristopher Kubicki, from Dailytech, ASUS spokesman Randy Chang told that the defect rate for the solid state storage devices on the Eee are lower than that of similar devices with traditional rotational media. In addition, Chang assures us that any defective SSD problems, including shrinking volume sizes, are covered by ASUS warranty.
Indeed, a quick search on the ASUS forums details only a handful of new posts about defective drives.
Current 8GB Eee models use Mini PCI-Express SSD cards instead of onboard chips. These can be removed and replaced by ASUS staff quickly if the drive fails. ASUS would not comment about why it chose to not solder its storage directly onto the motherboard, saving precious space, but a clever forum poster points out that its easier to replace defective solid-state memory when its in in a PCIe adaptor.
One thing is clear though, with ASUS pledging to sell nearly four million Eees this year, the company is fairly confident in the ability of NAND memory and high write scenarios. Apple and Lenovo both pledge to sell millions more NAND-based notebooks this year too: the Airbook and X300.
Tags: Intel Atom Centrino Atom platorm diamondvilleDiamondville is Silverthorne based.
Intel as announced that the processor previously know as Silverthorne will be sold under the Atom brand of microprocessors, targeted at mobile internet devices(MIDs). In the case that only Intel chips are used for the MID platform of a product, the manufacturer will be granted the right to also use Centrino Atom brand; a similar strategy to the one used with the Centrino brand for laptops, where the manufacturer must purchase the processor, chipset and wireless controller from Intel in order to be able to put a Centrino sticker on it.
As for details, the previously speculated 2GHz clockspeed didn't come true and is instead of 1.8GHz with a TDP varying from 0.6 to 2.5W.
The new chip is composed of 47 million transistors and is built with Intel's current 45nm process, delivering a very small chip measuring less than 25mm2. The previously announced support of HyperThreading technology has also been confirmed.
The Atom brand will also engulf the Diamondville processor, the codename for a platform targeted at small form factor systems.
The architecture used in Diamondville wasn't know before but it is now clear that it is also based on the same architecture as Silverthorne, yet to be named. Intel guidance suggests that the Diamondville is a 1.6GHz processor with 512KB of L2 cache and HyperThreading support running on a 533MHz FSB, very similar specifications to the Silverthorne. The TDP of Diamondville processors is of 4 to 8W, depending on versions of the processor, probably for the single and dual core versions, respectively. This can be attributed to a more relaxed binning of the same dies; the ones who leak more current can be sold as Diamondville cores and the better ones, with a lower TDP, as Silverthorne cores.
More details of platforms based on the Atom processors are expected this week, at CeBit.
Source: Intel
New, high performance kit.
While Coolermaster has never been famous for it's incursions in the watercooling market, this new design seems to promise competitive performance.
Coolermaster claims that the system can handle 600W but that isn't something you should be amazed at, at least for practical applications. If it can handle a 200W processor well, they will have a rather good design.
The whole system looks promising; it uses a rather good radiator with two 120mm fans and a pump that is good for 450L/h and the more significant 2m of maximum pumping height. If Coolermaster isn't overly exaggerating, the system features a good pump indeed, which will certainly boost overall cooling performance.
The system also uses the 3/8" inner diameter tubing, a good choice instead of the regular 6 or 8mm ID tubing.
Not know yet is the design and efficiency of the CPU waterblock, but it will support sockets 775, 754, 939, AM2, 940 and F.
No information on pricing is available at publishing time
Tags: Intel DX48BT2 bonetrail 2 specs release dateEarly Q2 release.
Intel will soon be updating it's current desktop high-end motherboard, known as the DX38BT "Bonetrail", with none other than the already widely known X48 chipset.
Intel guidance points to an early Q2 release, possibly as early as April.
Being planned to enter the market at the same time as the other competing X48 solutions, the the DX48BT2 will feature the well known ICH9R and not the upcoming ICH10R. The ICH10 series will be reserved for motherboards like the DFI LanParty UT X48-T3R, expected to be available latter.
The X48 is a carefully selected X38 variant for higher FSB, pin-to-pin compatible with the X38, so this motherboard wasn't redesigned and uses the same PCB as the DX38BT.
The motherboard will feature support for dual x16 PCI-Express 2.0 slots and possibly also allow to split one of the x16 slots in half to provide 8 lanes to the third PCI-Express slot, making available a 16-8-8 lane configuration, although that is a detail still not fully known yet.
Other known features are the support for 8 SATA ports, FSB 1600MHz and official support for DDR3-1600.
Intel has built some interesting boards lately, which although trailing the likes of Asus and DFI, have proven to be more successful than what was usual from in-house designs. More concretely, the "BadAxe" 2 was a moderate success among overclockers - something not usual with Intel branded boards - due to it's good build quality and refined BIOS options for overclocking. The new motherboard may prove a good purchase if Intel manages to follow a similar philosophy with the "Bonetrail" 2.
