Fudzilla: Bulldozer performance figures are in

[jvroig]

they never said WATER cooling just LIQUID For example, many solutions with heat pipes count as "liquid" because the liquid evaporates on top of the CPU, then condenses at the fins, and is absorbed by the wick and brought back to right above the CPU where the cycle repeats itself.

A good point. I didn't put much credence to XBit's "rumor" (shit made up by Anton Shilov). I've never seen Anand (or his employees) covering the left side of the screen with rumors... Carly Fiorina would've been President of the US by now.

No, it's not a good point, because context-wise, it doesn't make sense. AMD already offers heat-piped coolers for 125W - 140W procs for so long This wouldn't be news or even rumor-worthy if that's what they meant. Also, nobody, especially enthusiast media, ever uses "liquid cooling" to mean anything other than water cooling (although "water" actually isn't water but some other coolant).

They unambiguously mean stuff like Corsair H50. No two ways about it. Whether it is true or not remains to be seen, but there's no ambiguity in what the rumor is about - self-contained water-cooling setups like the H50, whose performance isn't really on par with "real" water cooling, and are beat by the high-end air coolers.

 

[Idontcare]

Anyone else concerned that packaging liquid cooling would mean it can't manage temps at the clocks they want to market at on air?

This is my concern.

AMD already does this with their 45nm CPUs. A 3.3GHz thuban can be binned and clocked at the clockspeed because the max operating temp that AMD allows for it has been lowered quite a bit.

IC stability is a function of temperature, not power-consumption. That is why LN2 cooled setups hit such awesome high clockspeeds despite consuming even more power because of those elevated clockspeeds.

 

[lol123]

IC stability is a function of temperature, not power-consumption. That is why LN2 cooled setups hit such awesome high clockspeeds despite consuming even more power because of those elevated clockspeeds.

This is a bit off-topic, but don't current and voltage play a part too, maybe not with regard to run-time stability but when it comes to degradation due to electromigration etc?

 

[Idontcare]

This is a bit off-topic, but don't current and voltage play a part too, maybe not with regard to run-time stability but when it comes to degradation due to electromigration etc?

Yep. Voltage, current, and temperature all play a role in the rate of irreversible degradation of the active components of the IC (xtors, wires, insulators, etc).

There is no temperature (save for 0 K), no voltage, and no current value for which degredation is reduced to zero. Changing the voltage, temperature, and current merely changes the rate of degredation.

Lowering the temperature improves it across the board, dramatically.

A chip clocked at 4GHz and 1.4V operating at a temperature of 70C is going to have half (50&#37 of the expected lifetime of the same chip at the same clockspeed and same voltage but cooled to operate at 60C. (every 10C decrease in temps will double the expected lifetime)

Reliability is the most difficult part of process development. Yields are tough but they are actually the easiest of the two. Getting yields up without cratering the intrinsic reliability of the IC itself is the challenge. Getting the intrinsic reliability of the IC itself to improve is the major challenge of all advanced process node development.

If you can't get the intrinsic reliability up then you have no choice but to lower the maximum allowed operating temp. That is the easy device-physics way out of the science-end of the conundrum.

Edit: corrected a critical omission in my lifetime reduction statement above (added text is underlined), thanks Martimus for letting me know of the typo!

 

[Keysplayr]

Arg Clin: Thanks. I don't know how I managed to miss that.
Now can anyone define "water-cooling" in the way they plan on it?
You know it isn't going to be a radiator, pump and reservoir.
Vapor Chamber, or at the very least heat pipes is my bet. A little bet.

 

[Meghan54]

Also, nobody, especially enthusiast media, ever uses "liquid cooling" to mean anything other than water cooling (although "water" actually isn't water but some other coolant).

Actually, in the water cooling community, amongst those that care about temps, water means exactly that.....pure, distilled water. Practically no other coolant performs as well.

 

[soccerballtux]

No, it's not a good point, because context-wise, it doesn't make sense. AMD already offers heat-piped coolers for 125W - 140W procs for so long This wouldn't be news or even rumor-worthy if that's what they meant. Also, nobody, especially enthusiast media, ever uses "liquid cooling" to mean anything other than water cooling (although "water" actually isn't water but some other coolant).

They unambiguously mean stuff like Corsair H50. No two ways about it. Whether it is true or not remains to be seen, but there's no ambiguity in what the rumor is about - self-contained water-cooling setups like the H50, whose performance isn't really on par with "real" water cooling, and are beat by the high-end air coolers.

oh, if they're currently offering heatpipe coolers then nevermind.

 

[soccerballtux]

Yep. Voltage, current, and temperature all play a role in the rate of irreversible degradation of the active components of the IC (xtors, wires, insulators, etc).

There is no temperature (save for 0 K), no voltage, and no current value for which degredation is reduced to zero. Changing the voltage, temperature, and current merely changes the rate of degredation.

Lowering the temperature improves it across the board, dramatically.

A chip clocked at 4GHz and 1.4V operating at a temperature of 70C is going to have half (50&#37 of the expected lifetime of the same chip at the same clockspeed and same voltage but cooled to operate at 60C. (every 10C decrease in temps will double the expected lifetime)

Reliability is the most difficult part of process development. Yields are tough but they are actually the easiest of the two. Getting yields up without cratering the intrinsic reliability of the IC itself is the challenge. Getting the intrinsic reliability of the IC itself to improve is the major challenge of all advanced process node development.

If you can't get the intrinsic reliability up then you have no choice but to lower the maximum allowed operating temp. That is the easy device-physics way out of the science-end of the conundrum.

Edit: corrected a critical omission in my lifetime reduction statement above (added text is underlined), thanks Martimus for letting me know of the typo!

i presume a superconductor wouldn't have electromigration issues no???

Where'd you hear the 10C half-life? Wasn't aware of this!!!

 

[Idontcare]

i presume a superconductor wouldn't have electromigration issues no???

Where'd you hear the 10C half-life? Wasn't aware of this!!!

Electromigration is simply the result of electric field gradient assisted diffusion.

Superconductivity doesn't prevent diffusion.

"10C half-life" as you put it is fundamental kinetics at work, the same old Arrhenius equation that you learned about in your pchem courses.

 

[ShadowVVL]

I run my stock clock C2Q for about a few hours every other day, it sees around 10 hours a week. Ive had it for 3 years its hit 72c once during a benchmark but other then that its @ 28c idle 52c max. How long you think he'll last doc?

 

[nonameo]

Electromigration is simply the result of electric field gradient assisted diffusion.

Superconductivity doesn't prevent diffusion.

"10C half-life" as you put it is fundamental kinetics at work, the same old Arrhenius equation that you learned about in your pchem courses.

:'(

I only took chem 1 and 2

 

[drizek]

same here... although I did use an erroneous equation on one of my midterms once.

 

[Abwx]

Some food for the trolls...

Some MSI insider info again.

He tells that he can't share all details, but that he is worried about Bulldozer. Forecasts were given (to MSI -and I guess to all strategic partners- to 'predict' how much capacity to devote to certain mainboard lines) that AMD would supply the AM3+ segment with 'huge' numbers of Bulldozer CPU's. Last week(s) word was received that AMD only will supply the AM3+ segment with an 'alarming' small number of CPU's. Seems to point at yield/production problems.

Fact seems to be that:

AMD has drastically lowered forecasted available bulldozer CPU's

http://www.amdzone.com/phpbb3/view [...] 00#p209349

 

[dma0991]

It could be due to the fact that AM3+ is going to get replaced next year and keeping a large inventory of AM3+ motherboards might be wasteful to MSI when demand drops as AMD releases FMx sockets for Bulldozer.

I spoke to a local reseller here and he claimed that he would be selling Bulldozer CPUs by 19th September or a week earlier as supplies arrive early. :ninja:

 

[soccerballtux]

erroneous equation

I see what you did there.

 

[AtenRa]

Anyone else concerned that packaging liquid cooling would mean it can't manage temps at the clocks they want to market at on air?

This is my concern.

If High End 8 core BD is 125W TDP you don’t have to concern, that means it will have the same cooling requirements as the previous Processors at 125W.

At 125W TDP even AMD Thubans(heat pipe) Cooler will be adequate for BD and it only shows that Liquid Cooling (Water or any other liquid medium) is a form of marketing and not a necessity.

 

[RussianSensation]

Another reason why you may want superior cooling for your top end chips is because stock TDP is meaningless once you start overclocking. However, enthusiasts who overclock may already have superior cooling or may want something better than say what Antec 620/H50 offer. So I still think this may or may not be a wise move because bundling Antec 920 / Corsair H70/80/100 will be too expensive while bundling lesser coolers is just not good enough.

 

[LOL_Wut_Axel]

Another reason why you may want superior cooling for your top end chips is because stock TDP is meaningless once you start overclocking. However, enthusiasts who overclock may already have superior cooling or may want something better than say what Antec 620/H50 offer. So I still think this may or may not be a wise move because bundling Antec 920 / Corsair H70/80/100 will be too expensive while bundling lesser coolers is just not good enough.

For moderate over-volting they're fine. I wouldn't run CPUs at high voltage anyway because I want them to last 3+ years.

I'll probably be going Sandy Bridge anyway because of the upgrade path, so meh.

 

[lau808]

someone please remind me, is bd goin to use 4 modules before a 5th core or is it using up the full modules 1st so 2 modules before using a 5th core?

 

[Tuna-Fish]

someone please remind me, is bd goin to use 4 modules before a 5th core or is it using up the full modules 1st so 2 modules before using a 5th core?

This is something decided by the OS, not the chip. If you mean under Windows, well, given that the way Windows avoids putting two threads on one physical core with HT is tossing the work threads around aimlessly on free cores (neatly invalidating caches as it goes), I'd expect similar insanity on BD.

As for Linux, well, it depends on which way you get better performance -- having them all on separate modules means they get private L2 caches and FPU's, but piling them tight means you get to powergate the rest and turbo higher. Who knows?

 

[lau808]

oh ok, isnt turbo going to have to be turned off when ocing? and i think id prefer 1 thread/module as u get 400% single core performance vs 2 module 4 threads = 360% single core performance

idk, just a thought

 

[AtenRa]

someone please remind me, is bd goin to use 4 modules before a 5th core or is it using up the full modules 1st so 2 modules before using a 5th core?

BD will use less Modules as possible.

 

[podspi]

This is something decided by the OS, not the chip. If you mean under Windows, well, given that the way Windows avoids putting two threads on one physical core with HT is tossing the work threads around aimlessly on free cores (neatly invalidating caches as it goes), I'd expect similar insanity on BD.

As for Linux, well, it depends on which way you get better performance -- having them all on separate modules means they get private L2 caches and FPU's, but piling them tight means you get to powergate the rest and turbo higher. Who knows?

Hasn't the windows scheduler at least gotten better since XP? Hopefully Win 8 will introduce !retarded thread scheduling...

 

[Abwx]

 

[-Slacker-]

AMD needs to stop doing this. They're a chip making company, not the USSR's propaganda department. I don't see how people in general can be impressed by these little shenanigans, and fans will only feel insulted, including my self.

Yes, Intel corp. is evil etc, but AMD needs to show a little more class with their marketing - I know this is a lot to ask from them, since, historically, their marketing has been complete ASS, but this is just embarrassing, for them, and for me.