Phenom X3 reviews

Yeah, AMD's top-end 2.4Ghz X3 is going for $200, and so is Intel's 2.4Ghz Q6600. "Meh" is about the best I can muster, even if it didn't have any software issues.

I will say that their sample overclocked better than quite a few of the "fully working" Phenom quads, so maybe that's something the enthusiast crowd will get into.

 

I don't get it. $40 cheaper than the (4core) 9850 (which is clocked higher already)? Why bother?

 

I just don't see the point in buying one of these, nor AMD's logic in making a triple core version.

 

It would have made more sense if Phenoms were higher clocked.

 

My only thoughts are if it allows them to salvage a chip that otherwise comes off the line with a defect(s). Disable one core, the chip still runs, it gets sold instead of thrown away.

 

Salvaging otherwise unsellable quad cores is the only reason why Toliman exists.
It's not only defects, however.

Looking at the power draw on some of the examples, the X3s also allow AMD to sell quad cores that exceed TDP limits due to device variation.
If the chip is a little over a 95W TDP with 4 cores, it might still make it with 3.

 

That's a good point too. I think AMD was hoping for some increased market segmentation with the X3s, but then got bit on the ass by Phenom's poor frequency scaling. So now these planned segments are piling up on top of each other, price-wise.

 

Also maybe AMD is aiming for OEMs. Saving $50 on a packaged system with a marketing line -- 3 cores is probably a good thing for them.

When I built systems, different levels of machines needed different prices. 50 dollars here and 14 dollars there all added up on a total system.

But for our own stuff which we build, that definitely is not the case.

 

How come AMD so often stumbles on clock speed?

Ever since K7 they've had problems in this regard. Surely they ought to know by now how not to design to cause this behavior to happen.. You don't see this happen to Intel after all.

Or is it just that the engineering expertise and silicon process challenge is so high that only really Intel is the company which has the resources to overcome these problems? In the desktp space anyway; I know IBM has their super 5GHz power6 or whatsitscalled.

Peace.

 

It seems I misspoke. I do this often. >.<

I meant to say I don't see why they make a 3 core version when a 2 core is more logical.

Apparently, I suck at the English language.

 

As we all know, the trick is to keep IPC up at the same time as getting clock speeds to ramp up. Obviously Intel had little problem getting to the 4Ghz+ mark, but was complete shite when it came to IPC.

I don't really know where the barrier is, but if I had to guess, it would be the logic routes / traces / something else other than process lithography. If Intel could mass produce a 3.73Ghz CPU on 90nm, I'm doubting that AMD's 65nm node is actually behind that curve, so it has to be something in the layout or logic.

 

They dont make a three core version, they make a quad core version -- and when the quad core chip has thermal, power or defect issues that can be resolved by disabling one core, you now have a three core bastard child

Does that make more sense? They're selling otherwise "faulty" (for various possible reasons) quadcore that has been castrated.

 

I wonder hw much power that disabled core pulls even being disabled.

If there's tens of millions of inactive transistors sitting there and leaking pwoer..
Peace.

 

It's probably a combination of a lot of factors, including AMD's 65nm process.

The 10h pipeline is not significantly different from K7, a design that was created before the serious issues with leakage and signal integrity that popped up below the 130nm node.
Without a design that really took these new issues into account, the basic architecture would become a stumbling block.
Intel with the P4 was in some ways right. While the thermal problems were a nightmare, it was clear that Netburst was not timing or signaling limited.

Both Intel and IBM ramped up either totally new or significantly redesigned chips for 65nm, and they got respectable clocks for it, though IBM's different design constraints make it a difficult comparison.
AMD did not make such a redesign, and it suffered for it.
AMD's voltage scaling has been lackluster, and its power curve is extremely steep and early, even when compared to the paltry gains other manufacturers have made, due once again to problems with signal integrity and leakage.

The process isn't blameless, since it shouldn't have been such a barrier for K8's scaling at 65nm.
AMD's designs have not significantly changed the FO4 per pipeline stage, which means there is one less factor between final clocks and the quality of the process.
We can look at the gate oxide thickness, which regressed from 90nm to 65nm, for part of the problem with clocking high.
I ran across discussions that part of the reason for this regression was AMD's shift to 12 inch wafers, and device variation worsened parametric yields on the wider wafers. The thicker oxide kept things in line, but at the cost of cutting off the top speed bins.

AMD's design has not changed enough with a significantly more stringent manufacturing environment, and its process is not up to snuff.

A lot of this is the result of limited resources and serious missteps.

 

It's not up yet, but my review ends with just that comment, that perhaps for AMD's sake the OEMs will pick up on the X3s because I certainly don't see enthusiasts doing so.

 

I think the real problem right now is that pricing-wise you can put the whole AMD lineup under a hat. There's no room for real price differentiation.

Tho 95W will make it more attractive to some people. I recently purchased a new mobo/cpu/memory and ended up going with a Q9300. I looked hard at a "Black", as much for the future lifespan of the platform (i.e. midlife cpu upgrade 12 months out), but I decided I didn't want a 125W cpu in my case.

I'm not convinced that multi-threaded apps are the best way to measure the utility of 3 and more core cpus, tho I can see why that's easier to do. I think the more cores there are the more you're really talking about smooth multi-tasking and background and rogue apps not harshing your mellow on your foreground apps performance.

 

I agree 100%. I work from my home office with many applications open at the same time. My Q6600 makes my PC smother to work on. Occasionally I might have 2-4 CPUs working on different applications but very very rarely. It's the ability to work on the prime task without slowing down that is extremely attractive.

A few applications take advantage of the multi-cores but most that the average person uses do not.