The sad state of AMD's processors

[CougTek]

Out of curiosity, I was looking at the relative performance of AMD's processors compared to Intel's at various price points and the global picture is quite depressing for the underdog. As general purpose processors, no AMD processor can hold a candle to Intel's offering in the same segment. Not even by a long shot. Only in very specific applications do they, sometimes, score a victory. In most other case, they are beaten, badly, by their rival's product. I don't even get the reason-to-be of their A4/A6/A8 processors. Their performances are simply shameful. Period. Regarding their value when you include their integrated graphics core, please. You either game or you don't. If the only market segment where you lead is the absolute lowest-end gamers who like to play with ugly graphics (interesting eye-candy effects turned off in order to get playable framerates), you're in deep shit. That's AMD's position right now too.

Jim Keller is better to pull out a miracle in order to put back their CPU division on track or the best move they'll have left will be to seel or shut it down entirely. I'm quite pissed about it too as the only reason why we won't have Ivy Bridge 8-core Xeon this Fall is that AMD can't deliver something competitive against Intel last-generation design...today. The PII-era pricing is about to make a comeback.

 

[MaxBurn]

It would be nice to have a viable #2 or even a new #1. Mostly what I am concerned about at the moment is motherboard prices, they seem to be getting simpler and less complex than previous generations with things built into the CPU and yet are going up in price.

 

[Chewy509]

I don't know what AMD engineers where thinking with the current Bulldozer design, since they had a good design previously. But alas, you can't keep throwing cores at a design (just ask Sun about the Sparc T3s) and expect performance, you need a good IPC (instructions per cycle) coupled with a very good cache design, and a smart OOE engine in the CPU for any super-scalar design. IIRC, I remember seeing an article that AMD relied on automated design tools for the Bulldozer rather than hand-crafting many modules within the CPU, (this is practice very common with GPUs, which allows GPU designers to scale out to 1000's of cores), but it seems that decision didn't pane out too well for AMD for use in a general purpose CPU...

But something that is not mentioned too often, is that designs we see coming to market today, started on the drawing board 2-3 *years* ago, and design choices are made at that time. Is AMD's APU strategy a good one? In some respects, I think it's a little too early, since it would appear that AMD's management and market researchers may have come to a conclusion 2-3 years ago that today, most people would be using tablets and/or lightweight ultrabooks/netbooks (where APUs make *a lot* of sense, and Intel didn't have anything rumoured to fill that need), but alas they predicted wrong, while the desktop market is shrinking in favour of tablets/ultrabooks, those new platforms don't have the share that I'm sure AMD predicted.

While I'm still using a Intel C2Q-9400 currently, I've seen little reason to upgrade the CPU, and I don't game much at all... So in that respect, an AMD based APU would make some sense for me due to equal peformance, but at a vastly lower power draw... AMD has some nice options for the budget builder, but Intel also have some very, very nice options that directly compete with AMDs solutions. If I was building today, would I look at AMD? Doubtful... (And I've been running AMD for years, prior to my current desktop setup).

@MB, current Motherboards are more expensive due to the tight tolerances needed by today's electronics that sit on the motherboard. eg, as you know a traditional PCI bus runs between 133MHz and 266MHz (depending on the variant), but current PCIe3 runs at 8GT/s (or roughly 10GHz), so the electrical requirements are far greater for modern setups than older setups from the AGP/PCI era. Since PCIe is a serial bus, and not a parallel bus like traditional PCI we can ignore clock drift/skew issues, but it still takes very good engineering to achieve 10GHz on copper vs 133/266MHz on copper... Not to mention power requirements for the CPU, memory all running at far higher speeds than before. Yes, it's crap, but it takes a lot of engineering to get to those speeds reliably. PS. PCIe v4.0 should run at 16GT/s (or roughly 20GHz signal speed).

 

[Bozo]

Intels new i5s are not perfect either. We have yet to find a 'nix operating system that will run properly on them. We have to install a video card.
But Windows 7 (circa 2009 ) runs fine. But you have to ask, is that an Intel problem or a 'nix community problem.

 

[Mercutio]

Intels new i5s are not perfect either. We have yet to find a 'nix operating system that will run properly on them. We have to install a video card.

So there's an issue with video drivers, not the CPU, right?

I'm a tremendous fan of the i3-2120. Support for four threads, good speed, under $100 (on sale, at least) and they work just fine on cheap H61 motherboards. Side-by-side with an AMD 1100T x6, there's an obvious subjective difference in performance for the Intel chip in some places where I would've thought the extra cores would help the AMD chip, like opening a folder in thumbnail view. I like that AMD has some quad-core options down around $70, but if I'm doing budget builds it turns out that I can get a competitive Pentium-based rig going for even less cash. That means there's no reason to look at AMD at all.

 

[Bozo]

Maybe my timing is off, but I thought Windows 7 and XP came out before the Video-on-CPU chips. Or maybe it's just that Intel and Microsoft have been in bed with each other for so long, that Intel screwed with the video. Which wouldn't surprise me at all.

 

[Chewy509]

But you have to ask, is that an Intel problem or a 'nix community problem.

Intel problem, since they are the ones writing the drivers for *nix.
AFAIK, there were some known issues with IVB based graphics, but have been fixed in recent Intel driver updates. However for Ivy Bridge based CPUs, you really do need the Intel 12.07 driver set. (Mesa 8.0, Intel 2.20.0, Linux 3.4+): http://intellinuxgraphics.org/2012.07.html
Unfortunately, the only way to stay with or get these drivers, is to run something like Gentoo, Arch or any other rolling release type distribution.

 

[timwhit]

OpenSuse also has a rolling release called Tumbleweed.

I've always heard that Intel's Linux drivers were quite good.

 

[Chewy509]

OpenSuse also has a rolling release called Tumbleweed.

I've always heard that Intel's Linux drivers were quite good.

From my personal experience, I've rarely had an issue with them on the older GPUs (like those found in the northbridge, eg 945G/GMA3100). I've haven't used them with anything newer than Core2 based chipsets (G43), so can't tell for newer CPU integrated GPUs.

 

[CougTek]

IIRC, I remember seeing an article that AMD relied on automated design tools for the Bulldozer rather than hand-crafting many modules within the CPU, (this is practice very common with GPUs, which allows GPU designers to scale out to 1000's of cores), but it seems that decision didn't pane out too well for AMD for use in a general purpose CPU...

I remember the ex-AMD-employee who told about this said they left approximately 20% efficiency right there (hand-craft versus automated design). That's huge.

They need to focus on IPC at the core level on their next CPU design. The frequency has been stuck at 3GHz-to-4GHz for the past ten years, so they have to squeeze the most instructions possible within that range.