• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

which speed is better? 2000/1600/1333? check out the results!

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
Interesting to see speed doesn't really matter (or favors lower latency) on the i-series either. Thanks for posting ... :thup:

That is surprising to me I do not think it would be the case on X58 as QPI is tied directly to mem speed (as is CPUNB speed or uncore)

From what I read the other day on p55 though many boards say QPI on p55 they do not actually have one because QPI was developed for interCPU communication and 1166 will never support that.
 
Last edited:
That is surprising to me I do not think it would be the case on X58 as QPI is tied directly to mem speed (as is CPUNB speed or uncore)

From what I read the other day on p55 though many boards say QPI on p55 they do not actually have one because QPI was developed for interCPU communication and 1166 will never support that.

right LGA1156 doesn't have or use QPI buss, it uses DMI buss. its a bit slower and maybe a bit narrow then QPI.
http://www.intel.com/Assets/Image/diagram/p55_diagram.gif
i really dont know if it is but most testing put the X58 ahead of P55. with base clock at the same speed with ram at the same speed as well.

none of this is really surprising to me, its all stuff we have had access to for ages now.
 
The intel side of DDR3 was covered on the front page a few months ago... boiled down to P55 needs higher speed.

I do not know much about P55 the platform never held uch interest for me since for hte longest time it was just as pricey as X58.

edit: Yes but side by side that with CPUz cpu and mem tabs open and you will see much higher uncore and QPI speeds.

I take too long to type on this craptop, i need to start quoting...
 
You mean the RAM test results or the Intel stuff ...?

the intel memory test results but even more to the point of speed vs timings. that has been the same thing we have dealt with since got SDR PC100 cas2 or PC133 cas 3...
It seems on X58 his claim has some weight to it

if you want a kicker watch your latency drop when you removed stick of ram. the LGA1366 platform takes a slight latency hit (ie increases) when you use the third channel.
 
if you want a kicker watch your latency drop when you removed stick of ram. the LGA1366 platform takes a slight latency hit (ie increases) when you use the third channel.

You on the benching team? :clap:


EDIT: ARGH!!!!

Yes dual is better for benching 2D for certain tests in more than one way ;)
 
interesting, so dual channel would be better for 2d benches?

that would explain the lower latency on my P55 vs my X58
well programs that need fast access to ram but given how fast things are now. DDR3-200 with cas 8 or cas 9 in dual imo would be best even for gaming. still the big kicker would be to have a ssd to improve latency from HD to ram for lvl's ect to load faster.

you could say using the extra channel is a good thing for the extra bandwidth. though its a weird thing to try to explain since i havent been around much. i would rather have a lower latency access to ram then higher bandwidth. since the cpu will only need so much, what would be the point of throwing extra at it. like driving a super car on a american highway, yea it can do 200+ but the speed limit is 65.

i guess after that point, you are getting to diminishing returns. with regards to what we are talking about. not only that but using the extra channel does increase cpu temp as well, not by much but 2-3c. ocing the cpu should be better with using dual channel on LGA1136.

You on the benching team? :clap:


EDIT: ARGH!!!!

Yes dual is better for benching 2D for certain tests in more than one way ;)
well i am on it but i have done any thing for them lately. not only that but i knew about this before i joined up on the bench team. it goes back to the dual vs tri reviews going on when LGA1336 hit the streets. they were showing = BW with tri DDR3-1333 cas7 to dual DDR3-1600 cas8. the other thing was they also showed the tri setup to have a higher latency then the dual on LGA1336.
 
1156 does have QPI, it's what links the on-package NB to the CPU core itself.
DMI links it to the PCH on the mobo.

That makes QPI very, very important for memory performance, like FSB was back in 775 days.
 
1156 does have QPI, it's what links the on-package NB to the CPU core itself.
DMI links it to the PCH on the mobo.

That makes QPI very, very important for memory performance, like FSB was back in 775 days.
So the QPI (at least for memory access) fills the same roll as cpuNB speed on an AMD ...?
 
1156 does have QPI, it's what links the on-package NB to the CPU core itself.
DMI links it to the PCH on the mobo.

That makes QPI very, very important for memory performance, like FSB was back in 775 days.

well there is no NB on the cpu, its the IMC the only thing there. unless we are talking the dual cores with on package video. the PCH still has all the functions of the NB just minus the memory controller. according to intel it doesnt have QPI, not even to link the IMC to the CPU, it is still the DMI buss. i could link all the LGA1156 cpus but here is the i5 750.
http://ark.intel.com/Product.aspx?id=42915
DMI 2.5 GT/s
vs
LGA1336 i7 920
http://ark.intel.com/Product.aspx?id=37147
Intel® QPI Speed 4.8 GT/s
# of QPI Links 1


So the QPI (at least for memory access) fills the same roll as cpuNB speed on an AMD ...?
QPI or DMI = HTT, not sure if they changed the names around again for HTT.
 
well programs that need fast access to ram but given how fast things are now. DDR3-200 with cas 8 or cas 9 in dual imo would be best even for gaming. still the big kicker would be to have a ssd to improve latency from HD to ram for lvl's ect to load faster.

you could say using the extra channel is a good thing for the extra bandwidth. though its a weird thing to try to explain since i havent been around much. i would rather have a lower latency access to ram then higher bandwidth. since the cpu will only need so much, what would be the point of throwing extra at it. like driving a super car on a american highway, yea it can do 200+ but the speed limit is 65.

i guess after that point, you are getting to diminishing returns. with regards to what we are talking about. not only that but using the extra channel does increase cpu temp as well, not by much but 2-3c. ocing the cpu should be better with using dual channel on LGA1136.

Thanks, that makes sense to me, guess I'll be tweaking and testing this weekend thanks to this thread :D
 
Thing is, the DMI is a 4x PCIe 2.0 link with intel bonus features.
The QPI ratio set in bios directly effects benchmark speed, meaning it is used for something.

The IMC on package also has the iGPU in it, it's a G45 on crack essentially, with the PCIe ripped out and stuck into the cpu core. It's my understanding that the IMC talks to the CPU via QPI. Seems like it'd have to, as the 2.5GT/s DMI can do isn't even close to what people are posting memory bandwidth wise.

Uncore = cpuNB, and bclk = htt as far as intel to amd labels go.
 
Thing is, the DMI is a 4x PCIe 2.0 link with intel bonus features.
The QPI ratio set in bios directly effects benchmark speed, meaning it is used for something.
yea i read that in wiki but what can i say its the wiki. not really that solid for info since anyone with a login can change things. not even back in the day did they mention the QPI/DMI were PCIE buss lanes being used. though this link here
http://www.villagegeek.com/Archives/WhatsNew/i7_and_i5_architecture.html
this guy really did his homework...

The IMC on package also has the iGPU in it, it's a G45 on crack essentially, with the PCIe ripped out and stuck into the cpu core. It's my understanding that the IMC talks to the CPU via QPI. Seems like it'd have to, as the 2.5GT/s DMI can do isn't even close to what people are posting memory bandwidth wise.

then i guess we have a problem then? if we look at IMC cpus vs on-package. everything in the past i read/saw showed the IMC being connected to the CPU's L3 cache. need to go look... I did just find this from back from 2008.
http://arstechnica.com/hardware/news/2008/04/what-you-need-to-know-about-nehalem.ars
as i thought from the start a cpu with a IMC as in not on package tied to a video processor. the L3 connects the IMC to the rest of the cpu, not via a DMI or QPI buss. though we are back to another area, if a on-package memory controller is not connected via the L3 ie the i3/i5 dual cores. then they must be connected another way, but how, via DMI or a wider/faster DMI link just for the ram. going on that link though the LGA1156 i5 750 or IMC cpus, the IMC would still be or need to be connected to the L3. it is all right there on one package, doesn't make sense to take a detour and use a DMI buss.

right now this is getting to the point my head hurts. spending 9hrs a day in the sun/heat in TX, makes for a interesting night when your required to think.

*edit*
i said my head hurt, i think that ares link messed me up bad... need to come back to this after some sleep... :x
 
Last edited:
I was reading the intel papaers on QPI earlier and although 1156 has QPI Link arch in it... it never said exactly what it is used for. I would assume connecting the different cores and IO buses (PCIe and DMI) (Actual it is registered as PCI BUS 0 device 16 :S)

Considering it does everything and talks to the NB on the X58 platform I can only assume that it is similar on p55... except there is no "NB" on p55 since the pcie and southbridge IO is integrated to the CPU....

Sounds exactly like HT on amd.
 
Yeah the location of the L3 on P55 surprised me greatly, i had always assumed it was in the IMC, but then there it is shown in the CPU. Maybe the QPI links L3 to IMC?

All i really know for a fact is that faster QPI* = faster benchmarks. Beyond that it's guesswork based on articles out in the wild.


* or whatever the QPI setting on gigabyte p55 mobos actually does.
 
Back