Mhackintosh blog

It’s usual for Apple to castrate the GPU of the Graphic Card, on iMac and laptops…

Previous iMac with reduced frequency and low-cost graphic memory (VRAM), laptops with down-clocked GPU, premium price but low-cost parts. And performance lagging far-behind half-cost PC!

This time, it’s the Radeon HD 5670, that ATI specify for GDDR5 with 64GB/s bandwidth. Only GDDR5. Apple put GDDR3 on it to attain 32GB/s, diminishing performance on a 1.5X to 2X factor! That might be a 10$ or 20$ economy for Apple, 1% of the iMac, but a real performance loss for gamers and anyone using GPU, for OpenCL, CUDA, Photoshop CS5 Suite, …

Notice that the ATI Radeon HD 5750 of the Quad-Core 27″ iMac is in fact a HD 5850 Mobility, offering approximately same level of performance, but less bandwidth (as usual).

Steve, sometime you are really cheap!

PS: Preceding iMac 27″ Core i5 Quad-Core had ATI Radeon HD 4850 with GDDR3, but offering 64GB/s bandwidth, having 256bits bus instead the 128bits bus of the HD 5670.

I discovered that looking at iMac 21.5″ and 27″ with Core2 Duo 3.06Ghz, both equipped theorically with Radeon 4670HD, without another GPU or OpenCL processor…

In fact, they both use nVidia’s MCP79 chipset, that is described like that:

The MCP79 chipset is a single-chip solution for small form factor notebooks. There are six versions of the MCP79. All the versions include a DirectX 10 GeForce graphics core which supports Shader Model 4.0, NVIDIA’s VP3 video processor, Hybrid Power, Hybrid SLI, and Hybrid Performance.

All versions of MCP79 includes a DirectX 10 GeForce graphics core (GeForce 9400M or GeForce 9300M). So this is disabled by Mac OS X on the new iMac instead of using it for OpenCL, as GeForce 9400M is far faster in real-world OpenCL processing than Radeon 4670HD due to it’s archiecture (tehnically: support of unified read-write in GPU SM shared-memory).

I don’t understand why Apple is using a chipset that is great for OpenCL processing, with peak performance-level of 40 Gflops (Core2 Duo 3.06Ghz peak performance is 24.28 GFlops, 3.33Ghz version reach 26.64), and enable good use of OpenCL, aggregating the computing power of CPU + GPU to sometimes double the performance-level of CPU-alone.

This is truly disturbing, as the same chipset stay with GeForce 9400M active on a MacBook Pro when display is handled by GeForce 9600M GT, and both GPU are used by OpenCL!

Why Apple act so strangely with it’s hardware, paying for it, putting it on the computer, making the consumer pay for it finally… And disabling it!

Finally, MacWorld changed it’s benchmark suite, to focalise on CPU more than other parts of what makes a great computer: balance between CPU, GPU, Memory and Hard-drive!

Anyway, what is really interesting is that old 3.06Ghz iMac is faster than new 3.06Ghz iMac on MacWorld’s new SpeedMark 6 suite. So the refurbished iMac 24″ 3.06Ghz is a real bargain

And for performance/price balance, the new entry-level 21.5″ Cor2 Duo 2×3.06Ghz equipped with GeForce 9400M seems to be unbeatable, and probably the most interesting model of the new line, except the 27″ Core i5 for power hungry

On OpenCL benchmarks that will rely on real-world OpenCL or CUDA ported applications, I wonder that old-generation equipped with GT130 or GeForce 8800 will easily beat Radeon 4670 or even 4850 cards, even if they are slower on pure games!

Alas, no USB 3.0, no eSATA, no SSD, no Firewire 3200 …

Wanna copy files from 1 external hard-drive to another? 45MB/s speed limit, my MacBook Pro with it’s eSATA ExpressCard is between 80MB/s and 90MB/s, 2X faster!

Please Apple, put an ExpressCard/34 slot into your next-generation iMac for us to be able to upgrade to eSATA, USB 3.0, 2 Firewire 800 ports, FireWire 3200, …