Evolution Of Intel Graphics: i740 To Iris Pro

[IInuyasha74]

From i740 to the Iris Pro family, we look at the evolution of Intel's graphics technology over the last 17 years.

Evolution Of Intel Graphics: i740 To Iris Pro : Read more

 

[adamovera]

Archived comments are found here: http://www.tomshardware.com/forum/id-2839510/evolution-intel-graphics-i740-iris-pro.html

 

[LORD_ORION]

Iris pro linux tablets please.

 

[LucaFire]

Great read Been waiting for something like this; no one really covers the history of Intel's graphics/integrated solutions.

 

[bit_user]

Larrabee (2009)
Picture 11 of 20

...
This FPU was theoretically specified at over 10 times more throughput than comparable Nvidia GPUs from the same time period.

No, it was theoretically specified at 2 TFLOPS, while Nvidia's GTX 285 could manage 708 GFLOPS. That's still impressive, but not > 10x.

Real world performance of Larrabee was demonstrated at ~1 TFLOPS.

Sources:

• 
  ■ https://en.wikipedia.org/wiki/Larrabee_(microarchitecture)
  ■ https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_200_series

Xeon Phi (2012)
Picture 14 of 20

As a result of the high compute performance relative to power consumption, the Xeon Phi 31S1P was used in the construction of the Tianhe-2 supercomputer in 2013, which persists as the world's faster computer today.

Perhaps you meant to say "one of the world's faster computers, today"? According to top500.org, Sunway TaihuLight is #1, displacing Tianhe-2 (MilkyWay-2) to #2.

Oh, and it's really not a GPU, though I guess it's fair to include, here. Without Xeon Phi, Larrabee ends up looking like a complete dead end. Plus, the whole HD graphics story is pretty uninteresting, with no huge changes from one generation to the next (aside from the addition of optional eDRAM).

BTW, you could've added the Knights Landing generation (https://en.wikipedia.org/wiki/Xeon_Phi#Knights_Landing), again showing the progression of the architecture started by Larrabee. With OmniPath, HMC, and motherboard socket compatibility, this made Xeon Phi start to look very interesting.

 

[bit_user]

The consoles run big GPU's along side ARM processors...

Sorry, the only mainstream console to use ARM processors, so far, is Nintendo Switch. XBox 360, PS3, and the last 3 non-portable Nintendo's used variants of PowerPC, while XBox One and PS4 are using AMD (x86-64) APUs.

Though, it's definitely true that the bulk of consoles' compute power is contained in the GPU. Same is true for many phones, tablets, and even Intel's desktop CPUs.

Speaking of consoles, I've always wondered if the original Haswell-era GT3e was originally developed as a bid to win the XBox One design. Microsoft has always had a thing for a chunk of fast, on-chip memory. XBox 360 and XBox One both have it, while neither PS3 nor PS4 went that direction.

The fastest computer's will be an array of co-processors including GPU's

This is actually pretty close to what the Xeon Phi is. The latest generation (codename: Knights Landing) is sort of a hybrid between a CPU and GPU, where they took low-power general-purpose CPU cores from their Atom product line and made them much more GPU-like.

Nvidia has not had to whip out it's ... and show it off yet.

Check out their Tesla products: https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#Tesla

As you can see from that table, they've been going after server & cloud computing for about 10 years, now. Intel is trying to keep up, but you're also correct that Nvidia holds a substantial lead in both raw compute and energy-efficiency.

http://www.nvidia.com/object/tesla-supercomputing-solutions.html

 

[namaku0]

Think about it, Intel's GPU are actually the best for what it is intended for

 

[bit_user]

Think about it, Intel's GPU are actually the best for what it is intended for

I'm not sure about AMD's older APU offerings, but Ryzen-based APUs will definitely stomp all over them, in this area.