Much ado has been made over the chip shortage that apparently affected Nvidia's Fermi architecture, but they're not the only ones dealing with a tight supply. According to DigiTimes, supply for AMD chipsets is also running tight. So much so, that AMD is letting foundry partner Taiwan Semiconductor Manufacturing Company (TSMC) tap into its 55nm capacity reserved for chipsets to help ease current GPU shortages.
You probably noticed that AMD's high-end HD 5870 and HD 5970 videocards were pretty tough to come by in the second half of 2009, which DigiTimes says was the result of low yields on TSMC's 40nm process. As a result, graphics card partners placed more orders for 55nm Radeon HD 4000 series parts than they otherwise would have.
TSMC's recent 40nm yields have been improving at a steady clip, which should help ease the tight supply of both 40nm and 55nm parts, but part of this is being offset by high demand, DigiTimes says. As it stands, AMD's chipset shortage will likely last throughout the third quarter, and possibly into the fourth.
Hynix today announced what it claims are the industry's first 2Gb (gigabit) GDDR5 chips using the 40nm manufacturing process. Boasting 7Gb/s of bandwidth and processing power of up to 28GB/s with a 32-bit I/O, these rank as the highest density graphics memory available.
But it's not all about sheer speed. Hynix says its new 2Gb chips also impress on the power consumption front. With an operation voltage of 1.35V, energy consumption drops down by 20 percent over previous parts built on 50nm technology, the company claims.
Hynix will begin mass producing the new chips in the second half of next year to coincide with increased demand for high-performance graphics memory.
Look for low-power DDR3 modules to hit retailers before the end of the year. That's because Elpida Memory today said it has finished development of its 40nm 2-gigabit (2Gb, with a lowercase 'b') DDR3 SDRAM and will ship samples next November. Mass production is slated to begin before the end of 2009.
On the manufacturing side, Elpida's smaller 40nm chips allows the company to achieve a 44 percent higher chip yield per wafer compared to 50nm, and a 100 percent yield for DDR3 products that operate at 1.6Gbps, the company said.
Elpida claims its 40nm 2Gb DDR3 chips use about two-thirds less current and support 1.2V to 1.35V operation, in addition to the DDR3 standard 1.5V. That's about a 45 percaent reduction in power consumption, which might not sound like much for a typical home user, but could add up in a server farm.
If you've waited this long to upgrade your graphics card, you might as well finish off the summer with whatever GPU you've been getting by with. That's because both AMD and Nvidia plan to release new videocards this fall..
According to news and rumor site DigiTimes, Nvidia's upcoming 40nm GeForce 210 (GT218) GPU-based cards will start shipping in October thanks to improved yields at foundry partner Taiwan Semiconductor Manufacturing company (TSMC).
Detailed specs remain light, but the GeForce 210 will come with either DDR2 or DDR3 memory and offer up support for DirectX 10.1 and Shader Model 4.1, sources say. Nvidia will follow up the GT218 launch with GT230 and GT300 parts in the fourth quarter of this year.
As for AMD, the CPU/GPU maker will finally launch its RV870 GPU this fall, possibly as early as September.
Back in February of this year, Samsung developed and validated its first 40nm DRAM chip, and now five months later, the chip maker announced it has begun mass producing 2Gb DDR3 using the smaller manufacturing process.
Samsung says the move to 40nm will provide around a 60 percent increase in production capacity over a 50nm process, and it won't all be relegated to the server market, according to news and rumor site DigiTimes. In addition to 16GB, 8GB, and 4GB RDIMMs for servers, Samsung will use the 40nm manufacturing process to build UDIMMs (unregistered in-line memory modules) for workstatios, desktops, and notebooks of up to 4GB.
The energy efficient chips support a data rate of up to 1.6Gbps at just 1.35V, double that of an 800Mbps 1Gb-based dual-die package.
Some recent reports have suggested that Nvidia is planning to launch their new 40nm GeForce GT 220 and GeForce G210 GPUs at the end of September.
Until now, Nvidia has had to delay the launch of their 40nm GPUs due to low yield rates from TSMC. But, recently the rate has improved a great deal, allowing Nvidia to schedule a launch before the end of the year and most importantly – in time for the holidays!
ATI today announced a pair of mobility chips -- ATI Mobility Radeon HD 4860 and HD 4830 -- built off of AMD's new 40nm manufacturing process, purportedly making them the first-ever 40nm notebook GPUs.
Both the HD 4860 and HD 4830 come with 640 stream processor, 826 million transistors, support for DirectX 10.1, dual integrated DisplayPort, HDMI with 7.1 surround sound, and are CrossFireX ready. The 128-bit HD 4860 boasts a 650MHz core clockspeed and GDDR5 memory clocked at up to 4Gbps. The HD 4830 (also 128-bit), meanwhile, ships with a core clockspeed ranging from 450MHz to 600MHz, and GDDR3 clocked between 800MHz to 900MHz.
"It's not just 40nm process technology that makes these chips so potent, they are based on the same award-winning TeraScale engine of our ATI Radeon™ HD 4800 desktop series," Rick Bergmen, senior VP and GM for AMD's graphics products group, wrote in a blog post. "Combining this gaming power with our ATI Avivo™ HD technology and Unified Video Decoder will keep all your HD content humming along at full 1080p resolution with bright colors and seamless playback on your HD display. We've also packed in our power-saving technologies like ATI PowerPlayTM, ATI PowerXpressTM, and ATI Switchable Graphics™ technologies so that you can keep gaming, watching and surfing a little longer."
Bergmen went on to say that 40nm desktop parts are "coming soon," with at least one site having already posted benchmarks of what's believed will be ATI's first 40nm-based desktop graphics release.
As for the 40nm Mobility parts, Asus is scheduled to ship laptops using the new processors in the second half of 2009.
Update 3/4/09 AMD has has posted more pictures of its new Mobility chips, along with a video showing the Mobility Radeon HD 4860 running a on a desktop system uing the MXM modules (no notebooks are currently shipping with the part).
Chinese news and review site ExPreview.com claims to have the skinny on Nvidia's upcoming GT212 GPU, which is being positioned to replace the company's GT200 series (GTX260/280). The site says Nvidia's 40nm GT212 will ship with 384 stream processors, up from 240 on the GT200. Texture mapping units (TMUs) will also be bumped up from 80 to 96 on the new part.
Interestingly, ExPreview says Nvidia will slash the memory bus interface from 512-bit to 256-bit, which the GPU maker plans to offset by using GDDR5 memory running at a higher frequency. The GT212 will also come with 1.8 billion transistors, compared to the 1.4 billion found on the GT200, ExPreview says. And with a die area measuring 300mm^2, the site expects power consumption will be "reduced greatly."
Stay tuned, as more information on Nvidia's upcoming flagship GPU will likely be forthcoming during this year's CES.
News and review site VR-Zone claims to have seen documents outlining Nvidia's plan to move to a 40nm process technology on its entire lineup of upcoming videcards starting in Q9 2009. On the high-end side, that includes the GT212 GPU, expected to debut in Q2 and replace the 55nm GT200, meaning a relatively short lifespan for the not yet released GTX285 and GTX295 videocards.
On the mainstream level, both VR-Zone and DigiTimes are reporting the launch of 40nm-based GT214 and GT216 GPUs in Q3 2009, which will replace the current G94 and G96 chips. Fleshing out the lineup will be four desktop SKUs for the GT216 and six desktop SKUs for the GT218, presumably representing different clockspeed and memory configurations.
Looking at the entry level, Nvidia's 40nm GT218 will supplant the company's G98 GPU, also in Q3 2009. Meanwhile, Nvidia's IGP line will see the launch of the iGT209, which is set to replace the GeForce 9300 and 9400 motherboard GPU series.
Is there no such thing as bad news or is no news good news?
From ATI’s point of view, it’s no news. Speaking Thursday at AMD’s analyst day, ATI’s graphics chief, Rick Bergman said it would continue with a steady as she goes path. That’s no surprise as ATI has recently seen a resurgence in market share, revenue and reception by power users.
Next year, Bergman said ATI would build on its “sweet spot” strategy. Instead of building huge monolithic graphics cards as Nvidia does, it would continue to concentrate on great $200 to $300 parts and combine them to take on Nvdia’s high-end parts.
With no firm hardware to reveal, Bergman took a few shots at Nvidia’s widely reported mobile GPU failures and lack of DX10.1 support which, he said, is the easiest way for developers to get to DX11 next year.
ATI also expects to be the first to get to 40nm with its graphics and chipset parts, Bergman said. He noted that ATI was the first to 65nm, 80nm and 90nm for graphics parts. What isn’t clear is where those 40nm parts would be fabbed. Although strongly hinted that ATI’s graphics chips and chipsets would made on AMD’s recently spin-off fabs, neither Bergman nor Doug Grose, the new senior VP of The Foundry Company, would commit to it. Grose did say that in 2009, The Foundry Company would have the capability to produce those 40nm parts.
Depite the lack of hardware news, Bergman said ATI would have an early present for its users; in December, the company expects to release a free GPU-based encoder that runs on modern Radeon cards. Based on the company’s new Stream API, the Avivo Video Encoder uses the parallel processing power of the GPU to transcode or encode video. Similar functionality has been available for Nvidia products under its CUDA API for months but the encoder is not free and published by Elemental Technologies.
Bergman also said that Avivo Video Encoder with a mid-range Radeon HD 4870 card will actually outperform Nvidia’s top-end GeForce GTX280 card in encoding tasks using Elemental’s Badaboom Media Converter. ATI’s Free Avivo Video Encoder will run on any 4000-series Radeon HD card.