Imaginary AMD Quad core

$hawn

Distinguished
Oct 28, 2009
854
1
19,060
The following post is highly speculative. I'm just trying to figure out what the best quad-core FX chip, AMD could bring out by year's end. The calculations below may not be all that technically sound. Reading this could possibly be a waste of your time. Peace :)

Given that a 3.6GHz FX 4100, which is actually a half disabled 8core chip, has a TDP of 95W, its highly likely that a native quad core module with only 6MB cache, if manufactured, would have a TDP of only around 80W.

Add in cyclos tech, and we could pull this down under 65W.

Now crank up the clock speed to accommodate a 125W budget (almost double the TDP), and a 4.9GHz/5GHz+ turbo quad core FX seems very much possible:)

Also, being a native 2 module, it would also be much cheaper to manufacture:) This unlocked chip could seriously put a huge dent in the ivy bridge core i3's, if priced around $145. 95W chips at 4.4GHz are also possible:)

What do you guys think?? :)
 

noob2222

Distinguished
Nov 19, 2007
2,722
0
20,860
not 100% accurate but sort of correct. As the module has 1 fpu, 2 alu and shared l2 cache, your only turning off 1 alu. Granted that 1 alu should be ~0 tdp, thats only 1/4 sections in the modular architecture instead of 1/2.

http://www.legitreviews.com/article/1931/2/

what did suprise me is when I saw that trinity had a 1.4 ghz idle voltage of 0.275V, thats just nuts compared to anything else we have seen.

2.7ghz at 0.8V and 3.2 ghz at 1.025V are also insanely lower, the 8120 stock 3.1 ghz is at 1.275V Thats a power reduction right there by ~25% at full load if it holds true for PD. 125W parts should be ~ 95W without changing settings. Thats not taking into account clock mesh vs cyclos, just voltage drop.

All this on 32nm. Intel didn't even come close by dropping to 22nm.

I still don't think for most people that the 43xx will be that viable of a solution over the 6 or 8 series. It will still be plagued by the performance hit on true cores > module cores.
 


At the same time they retain heat and must be cooled as well, it seems that would play into this a little bit.
 

$hawn

Distinguished
Oct 28, 2009
854
1
19,060


I was talking about a native 2module quad core :) not a 4 half-disabled module quad core :)
 
Not sure why people thin that 5GHz would be possible. There are still limits that cannot be bypassed with any new tech. 5GHz is great but I still doubt we will see a native stock clocked wuad core at 5GHz from Intel or AMD for a bit of time.

As the process tech goes down, the lanes shrink and more electron migration happens, which is why we needed HK/MG for 45nm (Intel) and 32nm (AMD) as it helps to stop leakage.

The higher the leakage, the more power consumption. Bulldozer (Zambezi/FX) has very high leakage, which is why when the CPU is overclocked power usage nearly doubles.

Add into this that the majority of people do not overclock and would perfer to have a cool and quiet system, and AMD/Intel develop their heatsinks around their TDP I doubt they will start packing water/high end (think Noctua/Zalman etc) air cooling with their CPUs as cost would rise.

Still 5GHz would be nice. But I doubt we will see it at least not for the next few gens.
 

Cazalan

Distinguished
Sep 4, 2011
2,672
0
20,810


That's most likely a bug. Only Intel's near threshold voltage Pentium chip is operating at that low a voltage, and it had to drop to 3Mhz to do it.