Woodcrest previewed

[Donnie27]

wasn't samsung part of the group that pricefixed rambus? I'm pretty sure it was them too.

It's not that. I still think there's something really wrong with those benchmarks. You take a CPU that has access to more memory and bandwith then its single CPU brother and you get a smaller performance jump. There's something wrong.

NO, Samsung was RAMBUS' only real partner due to close Intel ties. The problems was called Micron and Infineon the Big 5 then Later Big 7, they cheated and got caught, it just took 5.5 Years. The funny thing in all of this mess was Infineon and Micron Hate made Intel vs. AMD or nVidia vs ATI look small time.

Let's see, there's Fully Buffered DIMMS, Smart Cache, Smart access to Fully Buffered DIMMS while every ounce of cache is smartly used, ummmm? IMHO, I would be asking why Woodcrest didn't do even better with 21GB of Bandwidth. THat's skipps advanced arbitration, motherboard buffers that are processors in their own right, serialized DMI and links to FBDIMMS and etc...

I have NO doubt AMD will pick up on some of this tech to make its IMC even stronger for Servers and Desktops. Just as I know Intel will use and Alpha-like IMC just like AMD. While all the Intel vs AMD wars rage, Alpha was the real hero that should have drawn Fanb0y$.

 

[Donnie27]

I don't like the implication that AMD doesn't contribute to the Industry and just takes the cheap rout to copy Intel's tech.

Who introduced AMD64 and who coppied the instruction set?
Who set up the open HyperTransport standard that is used by dozens of companies world wide?

Just like Intel and any other smart company, AMD uses what it needs if it is available and invents it if it isn't.

Of course you wouldn't like that said about AMD, you like AMD. Fans always brag without understanding the implication. AMD was smart not to go with DDR2 was bragged about countless times. Where DDR2 be without Intel. Intel pushed DDR1 ahead to great links, with P4 there was not need for PC-3200 single channel and where would real Dual Channel be?

Intel had it own version of 64bit and MS killed IA32-64, "We'll support only version of 64bit and we choose AMD's version" was that Intel's fault? No I linked to Hypertransport guys before, unlike PCI-Sig, you have to pay to join, woopty doooo! Who else is using Hypertransport? In the words of Janet Jackson, "What have you done for me lately?"

AMD mooces more than they contribute. Like nVidia, they want their IP paid for but want others' IP for free, sorry, that's not right. See XBox beggings?

 

[NulloModo]

Of course you wouldn't like that said about AMD, you like AMD. Fans always brag without understanding the implication. AMD was smart not to go with DDR2 was bragged about countless times. Where DDR2 be without Intel. Intel pushed DDR1 ahead to great links, with P4 there was not need for PC-3200 single channel and where would real Dual Channel be?

Intel had it own version of 64bit and MS killed IA32-64, "We'll support only version of 64bit and we choose AMD's version" was that Intel's fault? No I linked to Hypertransport guys before, unlike PCI-Sig, you have to pay to join, woopty doooo! Who else is using Hypertransport? In the words of Janet Jackson, "What have you done for me lately?"

AMD mooces more than they contribute. Like nVidia, they want their IP paid for but want others' IP for free, sorry, that's not right. See XBox beggings?

Well, Intel killed off IA32 before it was born (at least I thought... from what I remember way back when wasn't IA32 supposed to replace x86 and IA64 supposed to take over all the server duties?) which may or may not have been a bad call... Personally I would have liked to see the Itanium (and whatever the consumer variant was to be called) chips have a real go and some serious R&D for the mass market invested into them, but I guess the early failures gave Intel cold feet. MS supported, and still supports as far as I know, IA64 in server class versions of windows.

Or, from a quick google, I could be really wrong... I thought IAxx was synomous with Itanium architecture, which I didn't think was exactly the same as x86, maybe I just had it wrong back then.

Either way, IA64 chips were out a long time before A64s.

 

[thecoldanddarkone]

NO, Samsung was RAMBUS' only real partner due to close Intel ties. The problems was called Micron and Infineon the Big 5 then Later Big 7, they cheated and got caught, it just took 5.5 Years. The funny thing in all of this mess was Infineon and Micron Hate made Intel vs. AMD or nVidia vs ATI look small time.

Let's see, there's Fully Buffered DIMMS, Smart Cache, Smart access to Fully Buffered DIMMS while every ounce of cache is smartly used, ummmm? IMHO, I would be asking why Woodcrest didn't do even better with 21GB of Bandwidth. THat's skipps advanced arbitration, motherboard buffers that are processors in their own right, serialized DMI and links to FBDIMMS and etc...


I 21 gb of bandwidth actually thats incorrect. Both 2cpu.com and techreport had ddr2 fb dimms at rated 533. 533 is pc 4200 ram, or 4.2 gb of bandwidth. ddr2 667 is 5.4 and ddr2 is 6.4 of bandwidth. 4.2 x 4 = 16.8 (roughly 80% percent of max bandwidth) I quote, "Our system's FB-DIMMs packed 1GB of DDR2 533 ECC memory with 4-4-4 timings" techreport page 3. 2cpu page page 2 says pc 4200 ram as well. So unless pc 4200 fb ddr2 fb dimms are rated at a different bandwidth I am not going insane.

I am not saying how much of a difference it i will make because I don't own one, however I just wanted to point this out to you.

btw techreport didn't botch the review it just happens to be the testbed which happens to be different than some other testbeds. This is why (although I believe they are good reviewers) I believe that it is important to get information from more than one source to see as many different applications as you can for whats being reviewed.

 

[savantu]

I was thinking the same thing. You have a 2.66 Ghz Conroe destroying a FX60 OCed to 2.8 Ghz in some benchmarks, while in this case it's two 3.0 Ghz Woodcrests, which are essentially the same chip with a different memory interface, beating two 2.6 Ghz Opterons by less of a margin, even though an Opteron 2xx is just an A64 with a different name and dual Proc. capability.

Woodcrest scales less well than the Opteron from 1S to 2S.

Having 2 FSBs means you need very complex logic inside the chipset to arbitrate and keep coherency , that adds latency and limits scaling.

One thing I expect Woody to do better is really intesive I/O.I mean several GBs...

 

[Donnie27]

I 21 gb of bandwidth actually thats incorrect. Both 2cpu.com and techreport had ddr2 fb dimms at rated 533. 533 is pc 4200 ram, or 4.2 gb of bandwidth. ddr2 667 is 5.4 and ddr2 is 6.4 of bandwidth. 4.2 x 4 = 16.8 (roughly 80% percent of max bandwidth) I quote, "Our system's FB-DIMMs packed 1GB of DDR2 533 ECC memory with 4-4-4 timings" techreport page 3. 2cpu page page 2 says pc 4200 ram as well. So unless pc 4200 fb ddr2 fb dimms are rated at a different bandwidth I am not going insane.

I am not saying how much of a difference it i will make because I don't own one, however I just wanted to point this out to you.

btw techreport didn't botch the review it just happens to be the testbed which happens to be different than some other testbeds. This is why (although I believe they are good reviewers) I believe that it is important to get information from more than one source to see as many different applications as you can for whats being reviewed.

I don't own one either and have to read up on the Chip Set in detail but from what I've seen so far is that yes, it's DDR2 533 and I was thinking similar after I wrote it, but thank you.