Intel Settles Lawsuits With AMD

Today was the day that AMD's legal efforts finally paid off.

In a joint statement the two companies commented, "While the relationship between the two companies has been difficult in the past, this agreement ends the legal disputes and enables the companies to focus all of our efforts on product innovation and development."

"We really couldn't keep appealing forever, and this time they really had us. We will now stop wasting time with monopolistic practices involving Dell and every other big player and will invest more in actual products."

Under terms of the agreement, AMD and Intel obtain patent rights from a new 5-year cross license agreement, Intel and AMD will give up any claims of breach from the previous license agreement, and Intel will pay AMD $1.25 billion. Intel has also agreed to abide by a set of business practice provisions. As a result, AMD will drop all pending litigation including the case in U.S. District Court in Delaware and two cases pending in Japan. AMD will also withdraw all of its regulatory complaints worldwide. The agreement will be made public in filings with the Securities and Exchange Commission.

"We thought we had them when AMD spinned off the fabs into Global Foundries, as we were expecting to revoke the x86 license. It was bluff and we ended up paying them $1.25 billion and doing a new 5 year cross license agreement on their own terms, so strong were the evidences against us. Damn."

It will be good to see both companies enter 2011 in a leveled playing field, to which Intel only has technology in it's advantage right now. AMD's efforts finally paid off in a big way, helping the company obtain a much needed cash infusion it was deprived off for some many times due to more than unethical business practices. Official court fillings detailed Intel paying as much as $6 billion to Dell over the course of 5 years to coherce Dell to stay away from AMD products. That considered, $1.25 billion was a small sum.

Lucid Hydra Performance Preview

The Lucid Hydra is supposed to be revolutionary: the new chip helps create more efficient Multi-GPU solutions by splitting work between the GPUs at a more fundamental level than what current software based Multi-GPU load balancing can do, although it hasn't been detailed exactly how that works, yet.

Lucid has said that it may allow to mix cards from different vendors although full support won't be available at launch. The chip will also support mixing cards from different performance and it will balance the load out by analyzing how long each card takes to render a frame.

This was what Lucid avidly promoted but today we're seeing another picture:


Driver profiles, or ways to deactivate the multi-GPU setup, as far as I could discern, due to possible glitches. Since Hydra was supposed to fix all of these problems of glitches or bad performance scaling, the dream starts to fade quickly.
Performance must be like... totally awesome! Right? No.


It apparently has little to no benefits. I don't expect performance to change going forward but I surely hope so.
One thing is working as promised, vendor agnostic multi-GPU:


These are good news indeed but the only lonely advantage. You can also use cards from different generations to have a slight boost but that would imply you use cards with the same shader model support or it will revert to the shader model support of the older card.
The technology looked promising but the whole picture isn't very pretty right now. The chips will cost money, perhaps more than the SLI license and definitely more than the free CrossFire support.
I'm not waiting for miracles on this one anymore.

Source: PC Perspective

AMD's 2010-2011 Roadmaps

AMD is... "ahead of schedule".

What do you think when you see "Ahead of schedule" in all slides of a roadmap presentation?


I think they know they're pretty late on the CPU front.
The six-core processor, codenamed "Thuban", also known as Phenom II X6, is ahead of schedule? The graphics cards are, AMD is the first to market; but ahead of schedule on a desktop version of a server CPU launched 5 months ago? I call that late.
Sure, the hexa-core will still be ahead of schedule for most desktop applications and users - just not ahead of schedule against Intel, which also has a faster micro architecture. The Core i9 is supposed to come out ahead of the Phenom II X6, not ahead. While AMD managed to tackle Intel on the server side by releasing "Istanbul" well ahead of schedule, on the desktop/workstation market, it doesn't look so good: they're late and will be slower. In the servers, "Istanbul" managed to strike a balance: high compute benefited AMD, high bandwidth requirements favored "Nehalem" - you didn't have a clear winner, especially with AMD's aggressive pricing.

Pretty interesting is the fact that AMD will apparently be releasing "Bulldozer" micro architecture based CPUs for the AM3 socket:


While we still don't know which socket the first "Fusion" based processor but it features the much anticipated APU(advanced processing unit) that will both serve as GPU and accelerator. The APU is very late to the party and AMD has started to suffer when it comes to integrated graphics performance. AMD needs to move the graphics core closer to the CPU pronto - Intel has been playing catch up and "Clarkdale" will released it around early January, featuring an integrated GPU with plenty of bandwidth and performance. This could be the turning point in Intel integrated graphics.
Given the current desktop market, it would make more sense to aim the "Fusion" based processor at the notebook segment first - two sockets for the same market usually doesn't work well, especially when AMD has less than 20% market share. Remember socket 754?
AM3 compatibility are very good news but, unfortunately, I fear for the worse.


AMD announced they will be updating the APU every year, on a similar fashion to Intel's "new architecture on one year, shrink in the other", also known as "tick-tock" strategy.
The plan seems promising and important if GPGPU will ever really take it's place in mainstream computing. This is the only non-"ahead of schedule" slide which actually seems well ahead of Intel's plans.

Full roadmap @ Anandtech

Creative Keeps Playing Vaporware With Zii

Creative's Zii Labs subsidiary releases the updated ZMS-08 chip, brings a GHz Cortex-A8 ARM core inside.

Specifications galore:

• Blu-ray Quality 1080p H.264 Video Decode
• 1080p H.264 Video Encode
• 720p H.264 Video Conferencing
• Accelerated OpenGL ES 1.1/2.0
• Xtreme Fidelity X-Fi Audio Technology
• Multi Format Media CODECs
• ARM Cortex-A8 at 1GHz
• Accelerated Graphics and Compositing
• Advanced Image Signal Processing
• Rich Peripheral Integration & Connectivity

The new ARM core is a new addition, but this is mostly marketing talk and don't get me started on the X-Fi capabilites...
As for the mysterious "Processing Elements", Zii Labs now unveils some more "details":

• High Compute Density SIMD Architecture
• 64 Processing Element Array
• 8 Clusters of 8 Processing Elements
• Utilisation of Array Scales with Workload
• Integer and IEEE 32-bit and 16-bit floats
• Offloads the CPU from Intensive Media Processing Tasks
• Fully Programmable to Support a Wide Range of Current and Future Standards
• Multi-Format Video Decode and Encode
  - Including; H.264, WMV, MPEG4, MPEG2, Adobe Flash, XVid, DivX and more...
• Image Encode and Decode
• OpenGL ES 1.1 and 2.0 3D graphics
• Xtreme Fidelity X-Fi Audio Processing
  - Crystalizer Intelligently Restores Audio Detail Lost During File Compression
  - CMSS-3D Adds Surround Sound Experience to Headphones and Stereo Speakers
  - Up-Mix Stereo to Sound Great on Surround Sound Speaker Systems
  - Smart Volume, Dialogue Enhancer and Graphic Equalizer
  - Acoustic Echo cancellation, Microphone beam former and Noise Reduction Modules
• Advanced Image and Display Processing performed on the media array
  - Scaling, Rotation, De-interlacing, Color-space conversion, Color and Gamma correction
  

No kidding? The utilization of an SIMD array varies with workloads? My god, I never thought so! What a revolution!
Jokes aside, perhaps they mean that the arrays can be turned off to save power? That would make some sense, although very short of revolutionary.

This is ridiculous, Zii Labs mentions an array of SIMD processors, 64 total, but mentions no peak throughput, no vector width, nothing. This looks a lot like Nvidia's GPUs, they're SIMD, but with some subtle differences, especially at the hardware level: there are no vector units but an array of small processors, each with an FPUs and two ALUs.
What it seems to me is that Creative has no revolution on it's hands but instead has eight clusters of eight very small, cacheless processors. Each of these arrays can run the same instruction on what would be a 256bit width array, hence SIMD. It's not quite spectacular nor easy to program, so this seems one reason to upgrade the ARM core.
Since the ZMS-05 has 48 PEs and is claimed to perform 8GFLOPS/s, that would put the ZMS-08 somewhere around 11GFLOPS/s. That means it will be trounced by Nvidia's Tegra 2, which is expected to be a CUDA compatible device, to which there is plenty of software already(comparatively) and more FP throughput.

Now lets take a look at this picture:


It doesn't tell much, does it? It tells us more about what we already know: the details of the ARM Cortex-A8.
That an array of small processors can do all those tasks described inside the "Flexible Media Processing Array", most of us already know: it's typical for modern heteregenous architectures.
Don't be fooled, the Zii Egg may end up an interesting product but it falls (very) short of all the hype.

PS: Creative has been asked to provide more documentation on the architecture. I don't expect them to show up anytime soon but it's worth a try.