Rumour: Bulldozer 50% Faster than Core i7 and Phenom II.

[jvroig]

Absolutely, which is why it (panic) is reserved for enterprise reliability.

For "enthusiast" usage, it is acceptable. I OC my Thuban to 4GHz core / 2.8GHz uncore. I"m happy with it. I have no guarantee that it will never cause problems, but this isn't enterprise usage anyway - who cares if down the line (2 years maybe) I have to deal with an issue (lowering my clock, for example), I can do it. And if I experience an OC-related crash/error/corruption while playing Mount&Blade, no problem, I can deal with it.

In other words, there is never "0 risk" - as an enthusiast, I just accept the risks. And in enterprise use cases, rarely (I mean never, actually) will you find a boss who is willing to take even the most minor risks, because if it does materialize into an issue, it will end badly for him.

 

[LOL_Wut_Axel]

Absolutely, which is why it (panic) is reserved for enterprise reliability.

For "enthusiast" usage, it is acceptable. I OC my Thuban to 4GHz core / 2.8GHz uncore. I"m happy with it. I have no guarantee that it will never cause problems, but this isn't enterprise usage anyway - who cares if down the line (2 years maybe) I have to deal with an issue (lowering my clock, for example), I can do it. And if I experience an OC-related crash/error/corruption while playing Mount&Blade, no problem, I can deal with it.

In other words, there is never "0 risk" - as an enthusiast, I just accept the risks. And in enterprise use cases, rarely (I mean never, actually) will you find a boss who is willing to take even the most minor risks, because if it does materialize into an issue, it will end badly for him.

Here's the problem with your argument: there's a lot less "risk" of OCing a CPU 300MHz on stock voltage than from OCing a CPU 800MHz+ on a moderate over-volt.

No one here has even proved if, in fact, running something like a Core i7 920 or a Core i5 2500 at 300MHz higher than stock on stock voltage will result in measurably less reliability. When someone can prove it to me, I'll admit I'm wrong. I'm not gonna hold my breath on that, though.

 

[jvroig]

There's no "argument", because it is not my 'argument". As an enterprise person, I am merely telling you how it actually is from an enterprise perspective.

It doesn't even matter how much risk is introduced in the system - what matters is if this risk can be mitigated by a vendor or manufacturer warranty and support (or any third-party contracted warranty/support you may have), and if it can't.

So let's say that you are correct, and an OC of X MHz is in fact 100% safe. Where are the studies to prove this? What do I show my bosses who have to approve it? How does this pass IT procurement screening? And no vendor or supplier will support this, so who will cover the warranty? I can't tell them that "LOL_Wut_Axel just proved it, he's a brilliant poster", because they will fire my butt faster than 5pm, and then there's the issue of you not accepting any warranty for workstations or servers that fail or produce data corruption because of X MHz OC.

Additionally, your way of thinking ("No one here has even proved if...") is the opposite of how things get done in the enterprise (but fine for enthusiasts). Here in EnterpriseLand, we don't care if it has not been proven - we care if it has been proven stable and comes with a vendor/manufacturer warranty&support to enforce that "proven stability". Because when I get a call or an email telling me one of the ERP servers are down and business is grinding down to a halt because of it, I don't want to rattle a list in my head whether it's the OC I did (or running anything else out of spec), or that I sourced it from a cheapo supplier that gave me a hefty bribe, or anything else that is just outside of warranty. There are already many points of failures, and I don't want to add to them unnecessarily, especially if it might cost me my job, no matter how small the risk.

That's how it is in the enterprise, which is why you'll probably never see any OC on them.

As far as enthusiasts are concerned, your view is correct, and I agree. All I have been doing in the two posts before this is explaining to you that what is good enough for enthusiasts cannot fly in the enterprise, and why. It is not my argument. It's just the way it is, a far cry from the world of enthusiasts.

 

[LOL_Wut_Axel]

There's no "argument", because it is not my 'argument". As an enterprise person, I am merely telling you how it actually is from an enterprise perspective.

It doesn't even matter how much risk is introduced in the system - what matters is if this risk can be mitigated by a vendor or manufacturer warranty and support (or any third-party contracted warranty/support you may have), and if it can't.

So let's say that you are correct, and an OC of X MHz is in fact 100% safe. Where are the studies to prove this? What do I show my bosses who have to approve it? How does this pass IT procurement screening? And no vendor or supplier will support this, so who will cover the warranty? I can't tell them that "LOL_Wut_Axel just proved it, he's a brilliant poster", because they will fire my butt faster than 5pm, and then there's the issue of you not accepting any warranty for workstations or servers that fail or produce data corruption because of X MHz OC.

Additionally, your way of thinking ("No one here has even proved if...") is the opposite of how things get done in the enterprise (but fine for enthusiasts). Here in EnterpriseLand, we don't care if it has not been proven - we care if it has been proven stable and comes with a vendor/manufacturer warranty&support to enforce that "proven stability". Because when I get a call or an email telling me one of the ERP servers are down and business is grinding down to a halt because of it, I don't want to rattle a list in my head whether it's the OC I did (or running anything else out of spec), or that I sourced it from a cheapo supplier that gave me a hefty bribe, or anything else that is just outside of warranty. There are already many points of failures, and I don't want to add to them unnecessarily, especially if it might cost me my job, no matter how small the risk.

That's how it is in the enterprise, which is why you'll probably never see any OC on them.

As far as enthusiasts are concerned, your view is correct, and I agree. All I have been doing in the two posts before this is explaining to you that what is good enough for enthusiasts cannot fly in the enterprise, and why. It is not my argument. It's just the way it is, a far cry from the world of enthusiasts.

You haven't proved it's a problem or that it may become one. Several stability tests will tell you otherwise. Where's one of them that proves all them wrong? To say it's bad to do a mild OC, first you have to have some proof of it, not just saying "it may cause a problem". Where's the proof of this problem? If there's no evidence, I'm not buying it. So far the only argument I can say is valid is the one of voiding warranty; I'll give you that one.

There's also the fact that a mild OC will raise efficiency, in terms of performance/watt.

If you're in enterprise in which the hardware is supplied to you, I can understand not wanting to do it. If it's a home business for which you bought the hardware and you know there will be no issues regarding reliability, I see no issue with it. If you think increasing a CPU by 300MHz on stock voltage will cause instability, unreliability, and even havok to your work, you're being paranoid.

 

[dawp]

Only a fool will overclock a mission critical computer, even if you believe it's 100% stable. When your livelihood is on the line, a crashed computer, for whatever reason, will cost you money.

Quite a few companies actually lease their systems and have a service contract.

 

[jvroig]

You haven't proved it's a problem or that it may become one. Several stability tests will tell you otherwise. Where's one of them that proves all them wrong? To say it's bad to do a mild OC, first you have to have some proof of it, not just saying "it may cause a problem". Where's the proof of this problem? If there's no evidence, I'm not buying it. So far the only argument I can say is valid is the one of voiding warranty; I'll give you that one.

There's also the fact that a mild OC will raise efficiency, in terms of performance/watt.

If you're in enterprise in which the hardware is supplied to you, I can understand not wanting to do it. If it's a home business for which you bought the hardware and you know there will be no issues regarding reliability, I see no issue with it. If you think increasing a CPU by 300MHz on stock voltage will cause instability, unreliability, and even havok to your work, you're being paranoid.

I think we should just let it go, because you are not getting (or refuse to get) how things work in the enterprise, and even though I just told you what matters (guarantee, warranty, and support), you still insist in applying your way of thinking (which is great for enthusiasts) towards enterprise deployments.

I don't know what else to say to make it any clearer, so let's drop this line, it's a useless derail anyway.

So, how about interesting Bulldozer dirt, found anything recently we can talk about? Last I heard it was like the Chuck Norris of CPUs and travelled through time and all that, which explains why it's late. Or something. Also, I heard IPC decreased, dang!

 

[LOL_Wut_Axel]

I think we should just let it go, because you are not getting (or refuse to get) how things work in the enterprise, and even though I just told you what matters (guarantee, warranty, and support), you still insist in applying your way of thinking (which is great for enthusiasts) towards enterprise deployments.

I don't know what else to say to make it any clearer, so let's drop this line, it's a useless derail anyway.

So, how about interesting Bulldozer dirt, found anything recently we can talk about? Last I heard it was like the Chuck Norris of CPUs and travelled through time and all that, which explains why it's late. Or something. Also, I heard IPC decreased, dang!

Well, I already told you. I can understand why enterprise wouldn't do it, but not why a home or home business user wouldn't.

If IPC decreased in Bulldozer it's a complete failure. The fact that it's being delayed due to a bad stepping doesn't inspire confidence to me, either. If the problem was that they needed to get higher clock speeds (3GHz or so) to be competitive, I see no problem. If it's because the older steppings were too low performing in terms of IPC (they only got measly 10% improvements overall), then AMD is doomed in the Performance market and will be relegated to just Mainstream and Essential. There's no way you're gonna get 10% more performance with a new stepping. None.

 

[jvroig]

Right. I just want to say I was kidding with the "IPC decreased" thing before JFAMD starts frowning at me again. I was just referencing some of the hilarious stuff that's been thrown around and made into jokes, like the whole Chuck Norris and terrence125(?) "IPC decreased" thing.

I have forgotten, when is BD Desktop supposed to arrive now? Is it before or after the Q3 Server Launch?

 

[LOL_Wut_Axel]

Right. I just want to say I was kidding with the "IPC decreased" thing before JFAMD starts frowning at me again. I was just referencing some of the hilarious stuff that's been thrown around and made into jokes, like the whole Chuck Norris and terrence125(?) "IPC decreased" thing.

I have forgotten, when is BD Desktop supposed to arrive now? Is it before or after the Q3 Server Launch?

AMD said in Computex there would be retail availability "in 60 to 90 days". That means an August or September launch. Server launch has not been affected by this; they say it's on schedule.

I was following JF-AMD on OCN and he says AMD will not give any information regarding performance until launch, which sucks horribly.

IPC has obviouly not decreased, but I still have my fears and doubts as to how much IPC has actually been improved. If it's only 10%, AMD is doomed in everything but applications that can take advantage of the eight cores. If it's 20%, then that's Nehalem territory, which is great since you get somewhat lower single-threaded performance than Sandy Bridge but much higher multi-threaded. At least in theory, anyway...

 

[jvroig]

Ah, that means also Q3, perhaps they will launch it along with the server line, hopefully not later since they promised it well before.

Perhaps it is just clockspeeds, after all. This is just me throwing stuff around, but if they continue the pattern of their server line in the Thuban era (well, that's just now, I hope I can use "era"), their server clockspeeds are far lower than desktop (I would go so far as to say "all server parts have lower clockspeeds than desktop parts", but I don't want to go to the trouble of verifying this right now, even though that is the general idea). MC was launched first if I am not mistaken, I think at a lowly 2.2 - 2.6 GHz. Then Thuban was launched, and it was almost a good 1GHz higher. On the absolute "positive-thinking" scenario, we could say AMD is now just "fixing" the clocks for the desktop to make it a hefty GHz better than the server part, so there are no "real" problems to speak of aside from that clock difference, which is why servers are not delayed and the desktop parts need a new stepping.

Of course, no substitute for a good bench. Too bad we have to wait for them, but it's a business decision I understand.

 

[bridito]

AMD said in Computex there would be retail availability "in 60 to 90 days". That means an August or September launch. Server launch has not been affected by this; they say it's on schedule.

http://tech.uk.msn.com/Computers-and-sat-navs/articles.aspx?cp-documentid=158441846

"Bulldozer remains a couple of months away from launch at best. It's likely to hit the shelves around early September..."

But what does MSNBC know anyway? They fired Mark Halperin for making an accurate observation of the President... (ducking and running to avoid the potshots from Democratic supporters...)

 

[LOL_Wut_Axel]

http://tech.uk.msn.com/Computers-and-sat-navs/articles.aspx?cp-documentid=158441846

"Bulldozer remains a couple of months away from launch at best. It's likely to hit the shelves around early September..."

But what does MSNBC know anyway? They fired Mark Halperin for making an accurate observation of the President... (ducking and running to avoid the potshots from Democratic supporters...)

Which makes sense, really. What we do know, though, is that it's probably gonna be a hard launch with immediate availability since AMD doesn't want to give performance information before then.

Clock speeds could well be the reason why the server CPUs continue on schedule since they're typically from 1.8GHz-2.6GHz or so.

The fact that AMD is hiding so much information regarding performance, coupled with the delays, makes me feel insecure about Bulldozer. But we'll see. I don't know if I can wait until then for what was supposed to be my summer build, though.

 

[formulav8]

Then why don't AMD/Intel just sell the chips at those frequencies?

yields, thermals, and power consumption to name a few

 

[bridito]

The fact that AMD is hiding so much information regarding performance, coupled with the delays, makes me feel insecure about Bulldozer. But we'll see. I don't know if I can wait until then for what was supposed to be my summer build, though.

I'm in the same boat. I want to replace my trusty i7 940 which will be reaching its 3rd birthday in the not too distant future and I would have thought a few months ago that by July I could have expected at least BD and possibly even SB-E to be available. And I too am losing faith in the earthshaking, earthmoving qualities of Bulldozer by the day. If this is just another midrange chip, it will be so damn disappointing as I would have loved to have a spheroids-to-the-wall AMD speed demon in my next system!

 

[Gundark]

IPC has obviouly not decreased, but I still have my fears and doubts as to how much IPC has actually been improved. If it's only 10%, AMD is doomed in everything but applications that can take advantage of the eight cores. If it's 20%, then that's Nehalem territory, which is great since you get somewhat lower single-threaded performance than Sandy Bridge but much higher multi-threaded. At least in theory, anyway...

Considering IPC, my understanding is that BD will use branch prediction very similar to Core arc. (it's already discussed on this thread). I can assume that it's one of the most signifficant factors of high IPC considering preCore CPUs. We can speculate that difference in IPC from Core and preCore is X, and hope that it will be simmilar to difference from Stars to BD. In theory, if that is the case, add to that that in some way Llano have managed to gain 6% more IPC and not from branch predictor (i don't know where i read it, so it might be false), and we can hope that it will be in 20% IPC territory. Too bad that things are more complicated that this.

 

[VirtualLarry]

No one here has even proved if, in fact, running something like a Core i7 920 or a Core i5 2500 at 300MHz higher than stock on stock voltage will result in measurably less reliability. When someone can prove it to me, I'll admit I'm wrong. I'm not gonna hold my breath on that, though.

http://forums.anandtech.com/showthread.php?t=2174024

What about my example? I had two strange app-crashes, in a weeks time, on a rig that was overclocked 500Mhz at stock vcore, passed OCCT 1-hour 64-bit linpack test at 90% (of 8GB) RAM coverage. Also passed Prime95 and Memtest86+.

I'm still scratching my head why too, even though an OS re-install had no problems (if rig was really unstable, should have errored during file decompression).

 

[LOL_Wut_Axel]

http://forums.anandtech.com/showthread.php?t=2174024

What about my example? I had two strange app-crashes, in a weeks time, on a rig that was overclocked 500Mhz at stock vcore, passed OCCT 1-hour 64-bit linpack test at 90% (of 8GB) RAM coverage. Also passed Prime95 and Memtest86+.

I'm still scratching my head why too, even though an OS re-install had no problems (if rig was really unstable, should have errored during file decompression).

I said 300MHz, a mild OC. 500MHz is a moderate OC, and may require a bit additional voltage.

 

[Phynaz]

I said 300MHz, a mild OC. 500MHz is a moderate OC, and may require a bit additional voltage.

I'm responsible for 33,000 desktops and laptops.

Are you willing to bet your job every one of those can take a 300Mhz overclock 100% of the time?

 

[LOL_Wut_Axel]

I'm responsible for 33,000 desktops and laptops.

Are you willing to bet your job every one of those can take a 300Mhz overclock 100% of the time?

As long as they use recent CPUs (from Core 2 onwards; first gen Phenom and 65nm Athlons not counting), yes. Chip degradation is caused by extra voltage, and 300MHz is something that can be achieved on stock voltage stably on any recent CPU. If they're computers owned by an enterprise, I understand you wouldn't want to do it because of warranty/support. My argument was for home and home business users and I pointed this out, so I don't know why you feel the need to mention you're in charge of 33K desktops. Perhaps you just want to start arguments for nothing, just like you always seem to do.

Anyway, you should try to stop getting this thread derailed further, especially since what you're arguing over I've already addressed in the thread.

 

[videoclone]

AMD needs bulldoser to be faster then PhenomII !!! it will be a joke if its slower.

 

[Cattykit]

I always overclocked home and work PC and never had a single problem. If one's afraid of OC failing, he shouldn't even use Windows or anything. Really, the problem is within OS, programs, and users rather than mild OC.

 

[Rifter]

As long as they use recent CPUs (from Core 2 onwards; first gen Phenom and 65nm Athlons not counting), yes. Chip degradation is caused by extra voltage, and 300MHz is something that can be achieved on stock voltage stably on any recent CPU.

False, chip degradation is also effected by temp as well as voltage, and that 300Mhz overclock is going to increase temps. The bottom line is overclocking ANY production machine(home office or large commercial/industrial office) is not a good idea. Stick to overclocking your gaming/personal use machines only IMO.

http://www.anandtech.com/show/2468/6

 

[LOL_Wut_Axel]

False, chip degradation is also effected by temp as well as voltage, and that 300Mhz overclock is going to increase temps. The bottom line is overclocking ANY production machine(home office or large commercial/industrial office) is not a good idea. Stick to overclocking your gaming/personal use machines only IMO.

http://www.anandtech.com/show/2468/6

Right, so you want to prove that overclocking does damage by linking to an article demonstrating how over-volting does damage? You don't need to over-volt to get a 300MHz increase, and a 1-2C increase in CPU temps won't degrade a chip. At this point, if you're gonna argue about temps, room temperature will make more of a difference than the OC. A mild OC on stock voltage won't do anything to a chip.

If it's machines in a home office or home business you're gonna support yourself instead of an enterprise, there's no problem with a mild OC.

But leave it to you to spread paranoia and BS.

 

[Lonbjerg]

But leave it to you to spread paranoia and BS.

wut?

 

[LOL_Wut_Axel]

wut?

What it says. He's always spreading BS. If you're paranoid over a chip increasing in temps by 1-2C and being degraded in any significant matter by a 300MHz OC, you're a moron. The amount of degradation over-volting with OCing causes is an order of a magnitude higher than what a mild OC with no over-volting will.