Sign in with
Sign up | Sign in
Your question

XEON 5160 vs X6800 Comparison

Last response: in CPUs
Share
September 17, 2006 11:41:26 AM

GamePC pits the X6800 up against the XEON 5160 in an interesting comparison of server based processor against desktop based processor. The XEON is at a decided disadvantage due to the necessary use of FB-DIMM's and their lack of high-speed options.

It will be interesting to see what happens once Intel goes back to DDR2, as they have said they are going to focus on that, rather than FB-DIMM.

http://www.gamepc.com/labs/view_content.asp?id=5160vs68...
September 17, 2006 12:02:50 PM

Really? Bah, I wish they wouldnt, FB-DIMMs rock.

There is no *reason* for them to be slow other than the fact that atm they only make server class high capacity, ecc/registered DIMMs, and its the ultimate in upgradeability with an FB-DIMM mobo supporting DDR2, DDR3, DDR4, etc etc with no changes.

Fewer mobo traces also mean Quad or Octo Channel is perfectly doable, and more DIMM slots is also fine.

FB-DIMMs do add slightly to latency but the should increase bandwidth substantially if implimented correctly
September 17, 2006 4:18:01 PM

Quote:
Really? Bah, I wish they wouldnt, FB-DIMMs rock.

There is no *reason* for them to be slow other than the fact that atm they only make server class high capacity, ecc/registered DIMMs, and its the ultimate in upgradeability with an FB-DIMM mobo supporting DDR2, DDR3, DDR4, etc etc with no changes.

Fewer mobo traces also mean Quad or Octo Channel is perfectly doable, and more DIMM slots is also fine.

FB-DIMMs do add slightly to latency but the should increase bandwidth substantially if implimented correctly
Actually, this is just rumor so far, as the INQ is the source of the story. I read it on techreport, who linked to the INQ. Acccording to the story, they will cut down FB-DIMM use in 2008, but we shall see.

http://techreport.com/onearticle.x/10727

As far as FB-DIMM increasing BW, look at the BW numbers in SANDRA, only about half the bandwidth of the Conroe. :?
Related resources
September 17, 2006 5:24:08 PM

The demise of FB-DIMMs may well be exagerrated. For instance, the "microbuffer" DIMM technology that The Inquirer saids is it's replacement doesn't exist. Industry experts have never heard of microbuffers before:

http://www.realworldtech.com/forums/index.cfm?action=de...

I have to admit though that the full benefits of FB-DIMMs is really only felt in 4P+ setups which is why a 2P pull back is possible.

Still I think FB-DIMMs are still in their infancy and will continue to improve. For instance, it's unclear whether the low bandwidth that FB-DIMMs displays is fundamental to the technology or whether it's just because this is Intel's first quad channel FB-DIMM controller which isn't optimized yet.

Concerns on power are also going to be alleviated by 2007. The Original AMB's consumed 6W and now they currently consume 5W. Intel is hoping to get it down to 4W which will help. The bigger benefit is the transition to DDR3 which will lower power consumption even further. I'm sure going to DDR3 800 with 5-5-5 or possibly 4-4-4 timings along with a 2nd generation memory controller will help solve some of the bandwidth and latency problems.
September 17, 2006 5:50:56 PM

Quote:
The demise of FB-DIMMs may well be exagerrated. For instance, the "microbuffer" DIMM technology that The Inquirer saids is it's replacement doesn't exist. Industry experts have never heard of microbuffers before:

http://www.realworldtech.com/forums/index.cfm?action=de...

I have to admit though that the full benefits of FB-DIMMs is really only felt in 4P+ setups which is why a 2P pull back is possible.

Still I think FB-DIMMs are still in their infancy and will continue to improve. For instance, it's unclear whether the low bandwidth that FB-DIMMs displays is fundamental to the technology or whether it's just because this is Intel's first quad channel FB-DIMM controller which isn't optimized yet.

Concerns on power are also going to be alleviated by 2007. The Original AMB's consumed 6W and now they currently consume 5W. Intel is hoping to get it down to 4W which will help. The bigger benefit is the transition to DDR3 which will lower power consumption even further. I'm sure going to DDR3 800 with 5-5-5 or possibly 4-4-4 timings along with a 2nd generation memory controller will help solve some of the bandwidth and latency problems.
I entirely expect that. Actually, i hated to even mention it, considering the source....when i saw the write-up at the techreport, i was interested, then when i followed the link, i thought...."i shoulda known". It's too bad that you can't get the server boards with overclocking abilities, as it might be fun to see how they would fare with some overclocking, and also how/if the FB-DIMM's would handle it. :) 
September 17, 2006 6:21:56 PM

Quote:
As far as FB-DIMM increasing BW, look at the BW numbers in SANDRA, only about half the bandwidth of the Conroe. :?


Yes I saw that, but that shouldnt be the case.

The theory of FB-DIMMs is that you have a minature memory controller on the DIMM, controlling the chips onboard.

This means that the main 'memory controller' only sees the smaller 'subcontrollers' on each DIMM, and that the 'subcontrollers' can control any memory type, from DDR2 to whatever new RAM type comes out in the future.

It also has a vastly reduced trace count, (although it still uses a DDR2 240pin connector, most of this pins are not connected) and therefore is much easier to route with multiple channels, and the onboard controllers cut down dramaticly on signal reflection etc, enabling more DIMMs per channel to run at a faster clock speed.

On paper therefore, it should be faster, and this is one of the main benefits of FB-DIMMs quoted by the designers. The onboard controllers do add slightly to latency.

Therefore, I'd be tempted to think that the results there were more down to an imperfect implimentation in the Intel chipset, than a factor of FB-DIMM usage. Certainly, if the FB-DIMM contains DDR2-667 modules there is no reason for it to be any slower than a non-buffered DDR2-667 module.

As for power usage, we are talking less than 6w here, so I really dont give a damn, as that much is probably added by extra onboard SATA controllers and USB joysticks left plugged in when unused.
!