Initial Penryn-based C2Ds will NOT be native quad-core

[Android1]

Originally Posted by AnandTech

...the first quad-core Penryn chips will simply be two of these on a single package, although later on we may see a single-die solution.

Originally Posted by ExtremeTech

The quad-core version will be similar to the current Intel Core 2 quad series, and will consist of two dies in a multi-chip package.

Full ExtremeTech article - Intel Takes Wraps Off 45 nm Penryn

Full AnadTech article - Intel Demonstrates new 45nm Transistors and Conroe's Successor





Interesting news. I am a bit disappointed about Penryn not being a native quad-core solution, although I am not sure if that will mean anything or not. Discuss.

If this is old news, I apologize.

[cjl64]

Originally Posted by Android1

Full ExtremeTech article - Intel Takes Wraps Off 45 nm Penryn

Full AnadTech article - Intel Demonstrates new 45nm Transistors and Conroe's Successor





Interesting news. I am a bit disappointed about Penryn not being a native quad-core solution, although I am not sure if that will mean anything or not. Discuss.

If this is old news, I apologize.

very interesting! Thanks for the post.

[caveman-jim]

its not that old, it only hit the native quad core discussion in the AMD Processor forum this afternoon, maybe and hour ago.

I'm disappointed also, as previous sources I'd seen had stated penryn was native dual core... but that means AMD is going to be first to market with a native quad which should boost their server sales nicely, especially combined with (a) HTT 3.0 and (b) Intel's bandwidth problems with scaling multi socket quad cores (c) cooler cheaper DDR2 registered ram vs. hot slow expensive FB-DIMMs.

[Lupine]

Does Intel plan to migrate away from FB-DIMMS? Is this history (rambus) repeating itself?

[moshpit]

Originally Posted by Lupine

Does Intel plan to migrate away from FB-DIMMS? Is this history (rambus) repeating itself?

God I hope so. FB-DIMM must die and go the route of rambus, away...

[Lupine]

Originally Posted by moshpit

God I hope so. FB-DIMM must die and go the route of rambus, away...

Leaving scores of happy FB-DIMM users who bought into the "future" just like RAMBUS users did. Nice. If true, Intel deserves another kick in the nuts.

[moshpit]

To be honest, I feel badly for those "scores of users", but the moment Intel switched to FB-DIMM is the moment those customers should have been wise enough to switch to AMD based servers for that purchase instead, period. Anybody who's researched into what they're buying would find the TERRIBLE latencies that FB-DIMM brings with it and should have known better. I wasn't merciful on Rambus users and I'm not going to start rewarding lack of research now with FB-DIMM by wasting time feeling too sorry for those who screwed up and went for it.

I may like Core 2 Duo, but that doesn't mean I magically have started trusting Intel. You STILL have to research what your getting and not buy on brand loyalty and those who do buy by brand only don't get my sympathy at all.

[Lupine]

Originally Posted by moshpit

To be honest, I feel badly for those "scores of users", but the moment Intel switched to FB-DIMM is the moment those customers should have been wise enough to switch to AMD based servers for that purchase instead, period. Anybody who's researched into what they're buying would find the TERRIBLE latencies that FB-DIMM brings with it and should have known better. I wasn't merciful on Rambus users and I'm not going to start rewarding lack of research now with FB-DIMM by wasting time feeling too sorry for those who screwed up and went for it.

I may like Core 2 Duo, but that doesn't mean I magically have started trusting Intel. You STILL have to research what your getting and not buy on brand loyalty and those who do buy by brand only don't get my sympathy at all.

Absolutely agree. The nuts kick is deserved for pushing FB-DIMMS, not for getting smart and going mainstream. Intel, intentionally or not, screwed a lot of people w/ their short-lived commitment to RAMBUS.

Thing is, at the server level, Woodcrest is pretty much the only game in town right now, or has been until recently. And Woodcrest = FB-DIMM IIRC in server class mobos.

EDIT: clarification. Woodcrest server-class mobos.

[cjl64]

Originally Posted by moshpit

To be honest, I feel badly for those "scores of users", but the moment Intel switched to FB-DIMM is the moment those customers should have been wise enough to switch to AMD based servers for that purchase instead, period. Anybody who's researched into what they're buying would find the TERRIBLE latencies that FB-DIMM brings with it and should have known better. I wasn't merciful on Rambus users and I'm not going to start rewarding lack of research now with FB-DIMM by wasting time feeling too sorry for those who screwed up and went for it.

I may like Core 2 Duo, but that doesn't mean I magically have started trusting Intel. You STILL have to research what your getting and not buy on brand loyalty and those who do buy by brand only don't get my sympathy at all.

100% approve of this post Exactly Moshpit!!!

[caveman-jim]

Originally Posted by moshpit

To be honest, I feel badly for those "scores of users", but the moment Intel switched to FB-DIMM is the moment those customers should have been wise enough to switch to AMD based servers for that purchase instead, period. Anybody who's researched into what they're buying would find the TERRIBLE latencies that FB-DIMM brings with it and should have known better. I wasn't merciful on Rambus users and I'm not going to start rewarding lack of research now with FB-DIMM by wasting time feeling too sorry for those who screwed up and went for it.

I may like Core 2 Duo, but that doesn't mean I magically have started trusting Intel. You STILL have to research what your getting and not buy on brand loyalty and those who do buy by brand only don't get my sympathy at all.

100% disagree.

Here's why. The majority of servers sold are not self build. They are complete items, often in bulk, bought from major OEM's - Dell, HP, IBM, SUN etc. And if they all bring the product to you, and have then benchmarks to show that the technology they offer is superior to another architecture for the environment you operate, what are you going to do? Obvious.

It's all well and good to say 'research' but the latency in FB-DIMMs is only recently coming to light. Dual 1333 FSB's with quad channel RAM offering 27Gb'sec b/w are hard numbers to argue with; especially when your alternative offers half that.

For my own project, the woodcrest servers I have are sufficient, the performance is excellent. A single 2.66Ghz woodcrest outclassing a dual 3.8 paxville is impressive in anyone's book.

So, those who have invested in intel woodcrest architecture with FB-DIMM's, think twice about dropping in that bios flash quad-core. You will get a performance increase; but it will be bottle necked. This seems familiar, doesn't it?

As for FB-DIMMs disappearing, I think they have at least a jump to 800mhz coming, possibly 1066. Due to their nature the more bus speed the better, so it makes sense to scale them. Quad cores, native or not, on 800 or 1066 fed FB-DIMMs are going to shine. FB-DIMMs are always going to be expensive, hot, and power consuming. All the BS about woodcrest power efficiency is wiped out when you run 4 or more sticks of RAM. This is another reason why AMD is still selling Socket F opteron's - they work. In fact, they are competitive at all real world applications; synthetic benchmarks show that woodcrest has full throttle speed advantage, but where both AMD and Intel systems are runing a 50-75% load and returning the same results, AMD is doing so cooler and cheaper. This better matches actual deployment in the enterprise - who sizes a server to run 100% capacity? No-one.

So, while I agree that FB-DIMMs are not an optimal solution, I don't think that the consumer should be blamed for buying what all the major OEM's are selling. If the consumer is fooled, then the OEMs are also - because they're standing their reputation and support behind it.

[moshpit]

Originally Posted by cavemanjim

Here's why. The majority of servers sold are not self build. They are complete items, often in bulk, bought from major OEM's - Dell, HP, IBM, SUN etc. And if they all bring the product to you, and have then benchmarks to show that the technology they offer is superior to another architecture for the environment you operate, what are you going to do? Obvious.

Yes, obviously you don't even BOTHER to read the performance data THEY give you, they WANT you to buy it. You do independant research into the technologies your buying for your IT department IN FULL. That means you KNOW what's going to be in those servers exactly, part for part or you don't buy them, anything else is irresponsible. You DON'T trust the fleet salesman to just give you what HE thinks you need for your company when buying new trucks, you evaluate your needs, find the EXACT specs you'll need to accomplish that by doing extensive research into what's available in that market segment, and then you TELL that fleet salesman what trucks you NEED and if his brand falls short on one of your needed specs, you go to ANOTHER company until somebody meets YOUR needs.

The OEMs can stand on their reputations all they want, use as stools is about as much as those reps are worth to a GOOD purchasing manager in IT. If Dell only will build a server for you with FB-DIMMs but you can buy the same level of performance from Sun without the loss of latency, it's your DUTY as IT purchaser to switch vendors.

[caveman-jim]

Originally Posted by moshpit

Yes, obviously you don't even BOTHER to read the performance data THEY give you, they WANT you to buy it. You do independant research into the technologies your buying for your IT department IN FULL. That means you KNOW what's going to be in those servers exactly, part for part or you don't buy them, anything else is irresponsible. You DON'T trust the fleet salesman to just give you what HE thinks you need for your company when buying new trucks, you evaluate your needs, find the EXACT specs you'll need to accomplish that by doing extensive research into what's available in that market segment, and then you TELL that fleet salesman what trucks you NEED and if his brand falls short on one of your needed specs, you go to ANOTHER company until somebody meets YOUR needs.

The OEMs can stand on their reputations all they want, use as stools is about as much as those reps are worth to a GOOD purchasing manager in IT. If Dell only will build a server for you with FB-DIMMs but you can buy the same level of performance from Sun without the loss of latency, it's your DUTY as IT purchaser to switch vendors.

Apart from the fact that you will get the same benchmarks and figures from each vendor, with the difference being ~5% between 1 server and another.

Show me a dual or more socket 771 motherboard that doesn't use fb-dimms.

And there are a number of organizations out there that only have a short list of vendors they can quote from - for security, local, state, and federal guidelines prohibit buying from non-sanctions/bidded/approved vendors.

In your ideal world of the consumer is king, your approach is ideal. In the real world of large corporations and enterprise agreements, the 'IT purchasing manager' often has their hand tied. In a mid size organization (say 15,000 employees) then the IT department don't need to bid out every job; they need simple management and simple support. Mixing as many different vendors as your system would provide in that mix would create a political and implementation nightmare, as each vendor passes the buck, cites violation of contract, or just plain isn't able to deliver their promises.

If you jump vendor everytime you find a promise of green pastures and golden fields, you end up on the turf over the septic tank. And it stinks.

BTT - AMD's quad core solution will be superior in the multi-socket arena, and in the single socket market there are so few applications that could use the power that only highly interested individuals will care if one has greater 'straight line' speed than the other.

I believe that AMD's quad core will exceed the performance of Kentsfield, and only the top end penryn models will be faster - but they will be overpriced, undersupplied monsters.

[moshpit]

But you highlighted my point all together and didn't even realize it! If FB-DIMMS are the only way to buy a current Intel server, yet you can go buy equal or BETTER performance out of a Sun Opteron server which DOESN'T have FB DIMMs and has FULL compatability with that Intel server for any task you need of it, then why should I care about the ones who mistakenly trusted thier OEMs but should have known better? That was my whole point. I can't feel sorry for any IT manager who makes that mistake. It's lack of research pure and simple when the info is out there to find about Opteron, Xeon, and their memory requirments.

[caveman-jim]

Agreed - NOW. But what about those who couldn't buy an Opteron server that doesn't have FULL compatibility, or performance?

ndant tCase in point.

I bought 6 PE2950's, 2 CX3-20's for a virtualization project. Time frame was last July. Only vendor I can use is Dell, because of state bid laws, company policy, and time to procure funding. Already I'm in a box. But then I note that only Intel has server boxes ready to go with VMware approval, critical to my application vendors also giving continued support. Another mark against AMD.
Every indepeest I saw showed major improvements going with Intel for the purposes I was going to use them for.

Fast forward to today - would I buy Intel again? Yes; for my application, dual woodcrests are still higher performing than the opteron equivalents. Now, I have the choice; then I didn't. Will I still pick Intel when the K8L is available? I don't think so, but without hard data is impossible to make a firm decision.

However, for this project, drop in quad cores would be a benefit - but the memory now makes a significant impact in that equation.

[moshpit]

Wait a tic. I have to get my brain around a couple of your points so bear with me. VMware approval for virtualization? I know that VMware is supported in hardware by Xeon, but as far as I know, VMware works fine on AMD chips as far as I knew. I used to use VMware to run a virtual DC and a couple virtual clients to do a bunch of my NT and 2000 labs on an Athlon MP no less, let alone how well Opteron would handle it, and this was LONG before hardware virtualization.

The other part I had trouble understanding was the part about vendor specific requirements. It seems like state law would do the OPPOSITE of forcing you to go with a single vendor, I'm not understanding this point. And as far as the company policy part goes, any policy forcing you to ONLY buy from a specific vendor is short sighted and while not YOUR mistake, is still a HUGE mistake on the part of the board that approved that policy and shows their lack of understanding how technology advances. And that single vendor loyalty can put a company out of business if that vendor ever screws you over, which happens enevitably with ALL vendors in one way or another eventually.

I know how rigid company policy can be tho, and just how self-defeating it often seems. Doesn't change my distaste for it and actually is a perfect example for why I got out of the corporate side of things and started working for small shops, just couldn't stand corporate stupidity anymore, lol!

[Raged_Fanatic]

Originally Posted by caveman-jim

its not that old, it only hit the native quad core discussion in the AMD Processor forum this afternoon, maybe and hour ago.

I'm disappointed also, as previous sources I'd seen had stated penryn was native dual core... but that means AMD is going to be first to market with a native quad which should boost their server sales nicely, especially combined with (a) HTT 3.0 and (b) Intel's bandwidth problems with scaling multi socket quad cores (c) cooler cheaper DDR2 registered ram vs. hot slow expensive FB-DIMMs.

Who cares? It's basically going to be based upon the same performance and architecture as their current stuff so it's going to suck even compared to Intel's current Core 2 Quad.

Why anyone would get AMD now is beyond me...

[moshpit]

Originally Posted by Raged_Fanatic

Who cares? It's basically going to be based upon the same performance and architecture as their current stuff so it's going to suck even compared to Intel's current Core 2 Quad.

Why anyone would get AMD now is beyond me...

Wow, that was a bit far to the left. I've been steadily defending Intel (for the most part) in this thread, but I gotta say that your statement is WAY over-generalized and almost nonsensical. K8 as it is, doesn't "suck" compared to Core 2 Quad in ANY way. A bit lower performance for a bit higher wattage, true, but the term "suck" would lead you to believe we're compairing Xeons to Celerons which would be completely inaccurate of a comparison level.

[cjl64]

Originally Posted by moshpit

Wow, that was a bit far to the left. I've been steadily defending Intel (for the most part) in this thread, but I gotta say that your statement is WAY over-generalized and almost nonsensical. K8 as it is, doesn't "suck" compared to Core 2 Quad in ANY way. A bit lower performance for a bit higher wattage, true, but the term "suck" would lead you to believe we're compairing Xeons to Celerons which would be completely inaccurate of a comparison level.

hence why he is one of the very few on my ignore list.... i suggest you do that same as he is an ass-hat.

[moshpit]

I try not to put people on ignore if I can avoid it in ANY way, I still even have good old Heymanyo off of ignore and he can REALLY be a buthead. But even buttheads make good points upon occassion, would hate to miss that rare word of wisdom buried in the rest of the slew of rudeness just because he irritates me the rest of the time...

[cjl64]

Originally Posted by moshpit

I try not to put people on ignore if I can avoid it in ANY way, I still even have good old Heymanyo off of ignore and he can REALLY be a buthead. But even buttheads make good points upon occassion, would hate to miss that rare word of wisdom buried in the rest of the slew of rudeness just because he irritates me the rest of the time...

I totally agree with you moshpit as the reason why it takes that special person to be on mine.... and raged is that special someone... i just dont see him ever eeking out any wisdom... he is always way out in left field and shits on threads with nothing but BS to say. He can say what ever he wants about me. I am a fair and kind person that will listen to almost anyone... hell even MazeWing is not on my ignore list. Trust me, that wisdom is never coming from raged_fanatic.

[caveman-jim]

Originally Posted by moshpit

Wait a tic. I have to get my brain around a couple of your points so bear with me. VMware approval for virtualization? I know that VMware is supported in hardware by Xeon, but as far as I know, VMware works fine on AMD chips as far as I knew. I used to use VMware to run a virtual DC and a couple virtual clients to do a bunch of my NT and 2000 labs on an Athlon MP no less, let alone how well Opteron would handle it, and this was LONG before hardware virtualization.

The other part I had trouble understanding was the part about vendor specific requirements. It seems like state law would do the OPPOSITE of forcing you to go with a single vendor, I'm not understanding this point. And as far as the company policy part goes, any policy forcing you to ONLY buy from a specific vendor is short sighted and while not YOUR mistake, is still a HUGE mistake on the part of the board that approved that policy and shows their lack of understanding how technology advances. And that single vendor loyalty can put a company out of business if that vendor ever screws you over, which happens enevitably with ALL vendors in one way or another eventually.

I know how rigid company policy can be tho, and just how self-defeating it often seems. Doesn't change my distaste for it and actually is a perfect example for why I got out of the corporate side of things and started working for small shops, just couldn't stand corporate stupidity anymore, lol!

Hardware virtualization support only appeared with Rev E AMD chips, and the Socket F opterons. Basically if you don't have an AM2/Socket F AMD, you don't have hardware VT. Makes a small difference - reduces a 10% overhead to 2%. This is significant when you scale up.

The policy for having preferred vendors is to stop each department from having to spend months getting 3 competitive bids to fulfil state law. For the most part it works, but sometimes it ties your hands and gives you less options.

I don't make the world, I just live in it

[caveman-jim]

Originally Posted by Raged_Fanatic

Who cares? It's basically going to be based upon the same performance and architecture as their current stuff so it's going to suck even compared to Intel's current Core 2 Quad.

Why anyone would get AMD now is beyond me...

2 Fallacies here -
1 AMD's performance sucks compares to Intel. In some cases AMD still outperforms Intel. For most of the gamers on here, that never happens. But for some of us in the enterprise sphere, there are other considerations.

2 Based upon the same performance and architecture. Core 2 Duo is just a die shrink and respin of a Pentium 3 Mobile, right?

[moshpit]

Originally Posted by caveman-jim

2 Based upon the same performance and architecture. Core 2 Duo is just a die shrink and respin of a Pentium 3 Mobile, right?

I know you said that part semi-sarcasticly, but I have to say something on that one. As I said earlier, I LOVE my C2D, but cannot help but think of it as a direct descendant of Pentium 3 that snagged the few good things about Netburst. That's a good thing tho, I respected Pentium 3 far more then I respect Pentium 4 and Netburst. So in my mind, C2D IS a respin of Pentium 3, but with dual cores, beefed up execution units, and a crazy fast fsb (compared to P3's). It helps me forget Intel made a Pentium 4...

[caveman-jim]

Originally Posted by moshpit

I know you said that part semi-sarcasticly, but I have to say something on that one. As I said earlier, I LOVE my C2D, but cannot help but think of it as a direct descendant of Pentium 3 that snagged the few good things about Netburst. That's a good thing tho, I respected Pentium 3 far more then I respect Pentium 4 and Netburst. So in my mind, C2D IS a respin of Pentium 3, but with dual cores, beefed up execution units, and a crazy fast fsb (compared to P3's). It helps me forget Intel made a Pentium 4...

whatever gets you through the night, man

[moshpit]

Originally Posted by caveman-jim

whatever gets you through the night, man

Meh, it works for me

[{0}Salamander{0}]

Any idea when Penryn quad core is due out?

[Grov]

Im to lazy to read, will 45nm Penryn Quad, work in my mobo?

[cjl64]

man... this thread is old... OT.. but some of the stuff i posted in here does not even sound like me... I dont know... sorry just had kinda a wtf moment.

[gramboh]

Originally Posted by Grov

Im to lazy to read, will 45nm Penryn Quad, work in my mobo?

I haven't been able to find if Penryn chips will drop into P965 boards or not. Some people say they will with a new BIOS, others say the voltage regulator can't handle it.

It isn't even confirmed for P35, but apparently X38 will handle them.

[caveman-jim]

Early demo's of Penryn ES had them running in modded Bad Axe 2 mobo's; the modifications were to the VRM's. It's possible that a newer mobo design than the BA2 would have that capability already and only require a BIOS flash to work.