whats so hard with disabling with a faulty core? if the tricores are priced reasonably i dont see any reason of it being unsuccessful. at least there a gap filler between x2's to x4's...its like a porsche cayman from the boxter to the 911... its just too bad that the enzos (penryns) are just so far ahead
The K10 is hard to make. In comparison to Q6600 - the Tri-Core is harder to arrive at.
The K10 is hard to make. In comparison to Q6600 - the Tri-Core is harder to arrive at.
By that analysis the k8 is also harder to make than the Q6600.
However, the tricore is certainly "easier" to make than the quadcore.
I think the right analysis is "efficiency". Making a monolithic quad core on 65nm is not very efficient at all. However, it is just as "easy" to make as it is any processor on AMDs 65nm node. The issue is with how "random" defects work, you have a *ton* of non-functional die as you get to a larger surface area. So the larger the die, not only do you get less per wafer, but you also get (exponentially?) less functional die per wafer. Not "harder", per se, just statistics biting you in the butt. Which is why -- presto -- you get the K10tri. Still the same number of die per wafer, but now you've got a better shot at having something that works. Too bad it seems as if AMD's defects aren't related to speed binning, but just overall die functionality. And not in the volumes I might have thought... otherwise why aren't they selling now?
And compared to intel who 1) have a process with a much lower defect density and 2) have a smaller unit transistor size, and 3) glue two smaller chips together to make a quad... well, there is no comparison in overall wafer-use efficiency. Intel wins, hands down.
By that analysis the k8 is also harder to make than the Q6600.
However, the tricore is certainly "easier" to make than the quadcore.
I think the right analysis is "efficiency". Making a monolithic quad core on 65nm is not very efficient at all. However, it is just as "easy" to make as it is any processor on AMDs 65nm node. The issue is with how "random" defects work, you have a *ton* of non-functional die as you get to a larger surface area. So the larger the die, not only do you get less per wafer, but you also get (exponentially?) less functional die per wafer. Not "harder", per se, just statistics biting you in the butt. Which is why -- presto -- you get the K10tri. Still the same number of die per wafer, but now you've got a better shot at having something that works. Too bad it seems as if AMD's defects aren't related to speed binning, but just overall die functionality. And not in the volumes I might have thought... otherwise why aren't they selling now?
And compared to intel who 1) have a process with a much lower defect density and 2) have a smaller unit transistor size, and 3) glue two smaller chips together to make a quad... well, there is no comparison in overall wafer-use efficiency. Intel wins, hands down.
The Tri-Core is harder to arrive at - it is made by making errors! You cannot predict precisely how many Tri-Cores you'll "make" in a given week. The Tri-Core is just as hard to make as the Quad, perhaps even harder since the efficiency of the Quads is estimated.
There is no sense in defending the Tri-Core. In reality, the best case scenario would be a Tri-Core made from ground up with 6MB of L3 cache.
The Tri-Core is harder to arrive at - it is made by making errors! You cannot predict precisely how many Tri-Cores you'll "make" in a given week. The Tri-Core is just as hard to make as the Quad, perhaps even harder since the efficiency of the Quads is estimated.
There is no sense in defending the Tri-Core. In reality, the best case scenario would be a Tri-Core made from ground up with 6MB of L3 cache.
The Tri-Core is harder to arrive at - it is made by making errors! You cannot predict precisely how many Tri-Cores you'll "make" in a given week. The Tri-Core is just as hard to make as the Quad, perhaps even harder since the efficiency of the Quads is estimated.
There is no sense in defending the Tri-Core. In reality, the best case scenario would be a Tri-Core made from ground up with 6MB of L3 cache.
Actually, you can predict/model how many tricore you are going to get out of a typical wafer. At least to the degree that you can predict how many fully funcitonal quad you are gonna get. Understanding defect densities and their effect on yields is at the heart of wafer budgeting and cap-ex expenditures. If they get this wrong, their fabs will either be overstrained or idle. (that being said, AMD proves my point by missing the boat completely on this one. =))
Don't get me wrong, I'm not defending the tricore -- I'm not sure it's to AMDs advantage to acutally sell the things, given that they aren't able to clock them significantly higher than the quads. I *do* disagree with the "ground up" tricore.. do the geometry on that -- it'll have a serious timing problem OR you will waste a lot of wafer real estate in the process.
PS: Exactly how hard/easy is it to fix this TLB thing any way?
It could possibly be a core design flaw. Which if that were the case it could be very difficult to figure out whats actually causing it. They would almost need to completely redesign the cores if that is the case. Maybe it would just be easier to tie the L2 and L3 into one large L2 cache.
AMD are just having their bad time at the moment, like Intel had with their P4. Good for Intel though because it means I'm buying my first intel chip since the Pentium 100!
I've had the following chips so far:
Penium 100
K6II 450 (Was waaay cheaper then a P3 / P2)
Athlon 1400 (Much better then intels fastest processor back then, Pentium 4 1.7 Ghz)
Athlon 3000+ S939 - My first overclocker hit 2.6 Ghz out of the box!
Athlon 4400+ X2 - Upgrade for the 3000+, it hits 2.64 Ghz
Intel C2D 8400 - Still on preorder, I'm waiting impatiently for some OC'ing fun XD
My graphics cards were as follows:
ATI 1 meg card (cant remember the chip)
ATI Rage Fury Pro 32 Mb
Geforce 2 MX 32 Mb
Geforce 3 64 Mb
Radeon 9800 XT 256 Mb
2x 6800's in SLI ... What a waste of money X1800 XT
X1900 XTX (Cost me +£40 from selling the X1800)
My next card will be this in a couple of months time:
Tri-Core is a DIE with a defective CORE that will continue to consume power no-matter what. AMD can't totally shut it off...
It's a bad DIE and i would not put it on my board.
I put my money on a better CPU.
The current QUAD core from AMD has a bug. BIOS work around and slower clock is implemented to mask it. The defect density on the current design is so high that AMD can't get a good yield from it.
I would wait for the next generation QUAD Core from AMD.
Tri-Core is a DIE with a defective CORE that will continue to consume power no-matter what. AMD can't totally shut it off...
Care to back that up with some kind of evidence or personal experience?
As far as I'm aware, they would use a laser to permanently sever all the connections to the faulty core ... Anything that would be left would have practically no resistance so no power consumption/heat generation.
You might be having stability issues because your +12V rail is running a less then 9?
I hope that I read that right but if you look at the first pic, the +12V is way low.
Hope that helps and I didn't just make a fool outa myself!
Zeph
Nope i've got a weak rail, and it reared its ugly head after I installed the hd3870. I've worked around it a bit though, but changing from case fans that use 4pin molex to fans that use 3pin mainboard headers. So at least Im not losing video and crashing in some games now.
---------------
AMD Phenom X4 9850 Black Edition, ZeroTherm Nirvana 120 Premium CPU Cooler, MSI K9a2 Platinum bios 1.1b3 or P.0J, 4GB (2x2) Mushkin DDR2 1066 (pc8500) 5-5-5-15 2.05v RAM, Sapphire Toxic HD3870, Raidmax RX-700SS PSU, Seagate Barracuda 7200.10 320gb SATA2 X
And nope I was getting a completely different machine stop error. This one specifically list Power supply, CPU, Memory, or Mainboard as possible culprits. First time it happened I was running 3dmark05 after installing the new video card, got a bsod talking about a machine halt error. Second time was while trying to play Civ4 Beyond the sword at 1680x1050. Third time was again with Civ4, and this time instead of a bsod, my video just went black and monitor went into standby no signal, even though the computer was still going. Since I moved some of the fans off the molex 12v connectors it hasn't been doing that, but i'll get artifacting every once in a while in civ4. Haven't had a problem with any other games.
And nope I was getting a completely different machine stop error. This one specifically list Power supply, CPU, Memory, or Mainboard as possible culprits. First time it happened I was running 3dmark05 after installing the new video card, got a bsod talking about a machine halt error. Second time was while trying to play Civ4 Beyond the sword at 1680x1050. Third time was again with Civ4, and this time instead of a bsod, my video just went black and monitor went into standby no signal, even though the computer was still going. Since I moved some of the fans off the molex 12v connectors it hasn't been doing that, but i'll get artifacting every once in a while in civ4. Haven't had a problem with any other games.
So basically your PSU is saying "I can't do it Captan!!! I just don't have the power!!!!! (in a Scottish accent)?
