AMD Details Bulldozer Processor Architecture

[bear jesus]

Honestly I'm fine with DDR2 and low timings. My board can run DDR2 @1333. But currently I run at 1067. But look at my timings.

I kind of agree, one of the things making it easy for me to wait to go with dd3 is the timings, when i move to ddr3 i want 8gb across 2 modules and the best that's easily available to me is 2000mhz at 9-10-9-27, I'm sure running it slower than 2ghz would possibly let me lower the timings but going from 5-5-5-15 ddr2 i would kind of want at least cas 6 or 7 with ddr3.

I hope in the coming months more memory will be released with lower timings.

 

[CDdude55]

I kind of agree, one of the things making it easy for me to wait to go with dd3 is the timings, when i move to ddr3 i want 8gb across 2 modules and the best that's easily available to me is 2000mhz at 9-10-9-27, I'm sure running it slower than 2ghz would possibly let me lower the timings but going from 5-5-5-15 ddr2 i would kind of want at least cas 6 or 7 with ddr3.

I hope in the coming months more memory will be released with lower timings.

Ya, as always the timings will get lower as time goes on. Currently im running 6GB DDR3 sticks at 1333 with timings of 7-7-7-20,which is a decent balance for me.

 

[Neo4]

Ya, as always the timings will get lower as time goes on. Currently im running 6GB DDR3 sticks at 1333 with timings of 7-7-7-20,which is a decent balance for me.

Won't get much better than what you have there unless you want to spend a ton of money with little performance improvement. My whole point all along is that RAM is RAM and you might as well get the most value because there's no point in getting high dollar stuff. Manufacturers could justify that cost if there was a real world improvement of at least 10% or more. I guess if you have deep pockets though...

 

[CDdude55]

Won't get much better than what you have there unless you want to spend a ton of money with little performance improvement. My whole point all along is that RAM is RAM and you might as well get the most value because there's no point in getting high dollar stuff. Manufacturers could justify that cost if there was a real world improvement of at least 10% or more. I guess if you have deep pockets though...

It really depends, each Memory standard is going to add more and more bandwidth, speed etc., even if there isn't a significant boost in performance, it's still going to be a standard on high end platforms, so you have pretty much have no choice in the uber high end market. I definitely agree though get what you need for a good price, as currently your aren't going to see a big difference. Even for me, i could of easily stayed with DDR2 800 memory, but if i was to upgrade, and i have the money, might as well throw in the towel with the latest standards.

 

[bear jesus]

Ya, as always the timings will get lower as time goes on. Currently im running 6GB DDR3 sticks at 1333 with timings of 7-7-7-20,which is a decent balance for me.

That reminds me, i'm sure i read somewhere that he latency with ddr3 is lower than with ddr2 so the timings are not really the same so ddr3 cas 7 is more like ddr2 cas 6 or 5 so that is a good balance of size, speed and timings, i only want a dual channel 8gb, cas 6, 2ghz ddr3 kit as i'm a whore

*edit*
With faster ram as far as i knew the only time you would see a big improvement is when something needs more bandwidth than is available, i only want so much bandwidth to go with 6 or 8 cores and multiple virtual machines to run 24/7 while carrying on all other normal usage.

 

[TheMailMan78]

I mean as far as I know the AM2, AM2+, AM3 don't even support tri-channel.

 

[bear jesus]

I mean as far as I know the AM2, AM2+, AM3 don't even support tri-channel.

They don't, and from what i have been reading desktop versions of bulldozer with be dual channel as well.

 

[InnocentCriminal]

I got all swallowed up when the details of the original Phenom started to circulate. The idea behind a completely native quad-core was and still is fantastic. I convinced myself that Phenoms were going to kick some serious ass, but unfortunately that didn't pan out. Learning from my experiences, I'm going to hold out until more information comes out on Bulldozer.

I'm with Wile E on this one, I want the best components for my rig(s) and if Bulldozer can compete with Intel on both performance and then it'll be my next purchase. I'm also hoping AMD have some fight in them and if the deal with Apple is true, then that'll be some much needed revenue that'll hopefully give them more resources to make all the right moves resulting in more competitive prices for the consumer... me!

 

[LAN_deRf_HA]

Won't get much better than what you have there unless you want to spend a ton of money with little performance improvement. My whole point all along is that RAM is RAM and you might as well get the most value because there's no point in getting high dollar stuff. Manufacturers could justify that cost if there was a real world improvement of at least 10% or more. I guess if you have deep pockets though...

That's why I think the unlocked uncore is the best feature of 1366. Let's you make cheap memory that only does tight timings at low speed perform like it's running at a much higher mhz.

 

[mastrdrver]

You all realize that DDR3 has lower latencies then DDR2 right?

 

[bear jesus]

You all realize that DDR3 has lower latencies then DDR2 right?

The latency with ddr3 is lower than with ddr2 so the timings are not really the same so ddr3 cas 7 is more like ddr2 cas 6 or 5

but i have no idea how much lower i was just guessing.

 

[CDdude55]

You all realize that DDR3 has lower latencies then DDR2 right?

When did that happen? lol

Here's an DDR2 OCZ kit of memory with cas 5 timings: http://www.amazon.com/dp/B0017SA5ZY/?tag=tec06d-20

And here's the DDR3 version of that same kit with higher speed but still higher timings (cas 7): http://www.amazon.com/dp/B0013HC36S/?tag=tec06d-20

Still fairly high from what im looking at, which DDR3 kits are lower?

 

[bear jesus]

When did that happen? lol

Here's an DDR2 OCZ kit of memory with cas 5 timings: http://www.amazon.com/dp/B0017SA5ZY/?tag=tec06d-20

And here's the DDR3 version of that same kit with highwer speed but still higher timings (cas 7): http://www.amazon.com/dp/B0013HC36S/?tag=tec06d-20

He means as in cas5 on ddr3 has lower latency than ddr2 cas5 as apparently the clock cycles are shorter. but i have not really read into it.

 

[mastrdrver]

When did that happen? lol

Here's an DDR2 OCZ kit of memory with cas 5 timings: http://www.amazon.com/dp/B0017SA5ZY/?tag=tec06d-20

And here's the DDR3 version of that same kit with higher speed but still higher timings (cas 7): http://www.amazon.com/dp/B0013HC36S/?tag=tec06d-20

Still fairly high from what im looking at, which DDR3 kits are lower?

The timings given by any dimm is only relative to its speed (in nanoseconds).

Timings =! latency

DDR3 1600 CAS 9 has a lower latency then DDR3 1333 CAS 9 because the speed (in nanoseconds) of DDR3 1600 is 1.25 and DDR3 1333 is 1.50 hence you get 10.5ns for 1600 and 13.5ns for 1333.

DDR2 666 CAS 4 has a latency of 12ns because the speed of the dimm is twice that of DDR3 1600 at 3ns.

Secrets of PC Memory

Read that and understand this:

DDR refresh clock (tCLK):
DDR 200 is 10.0ns
DDR 266 is 7.52ns
DDR 333 is 6.02ns
DDR 400 is 5.00ns

DDR2 refresh clock (tCLK):
DDR2 400 is 5.00ns
DDR2 533 is 3.76ns
DDR2 667 is 3.00ns
DDR2 800 is 2.50ns
DDR2 1066 is 1.876ns

DDR3 refresh clock (tCLK):
DDR3 1066 is 1.876ns
DDR3 1333 is 1.50ns
DDR3 1600 is 1.25ns
DDR3 1866 is 1.07ns
DDR3 2000 is 1.00ns

To find the latency of any dimm you need to take its tCLK and multiply it by its listed timings to find the latency of that timing.

 

[CDdude55]

Very good info guys, thanks.

 

[TheMailMan78]

The timings given by any dimm is only relative to its speed (in nanoseconds).

Timings =! latency

DDR3 1600 CAS 9 has a lower latency then DDR3 1333 CAS 9 because the speed (in nanoseconds) of DDR3 1600 is 1.25 and DDR3 1333 is 1.50 hence you get 10.5ns for 1600 and 13.5ns for 1333.

DDR2 666 CAS 4 has a latency of 12ns because the speed of the dimm is twice that of DDR3 1600 at 3ns.

Secrets of PC Memory

Read that and understand this:

DDR refresh clock (tCLK):
DDR 200 is 10.0ns
DDR 266 is 7.52ns
DDR 333 is 6.02ns
DDR 400 is 5.00ns

DDR2 refresh clock (tCLK):
DDR2 400 is 5.00ns
DDR2 533 is 3.76ns
DDR2 667 is 3.00ns
DDR2 800 is 2.50ns
DDR2 1066 is 1.876ns

DDR3 refresh clock (tCLK):
DDR3 1066 is 1.876ns
DDR3 1333 is 1.50ns
DDR3 1600 is 1.25ns
DDR3 1866 is 1.07ns
DDR3 2000 is 1.00ns

To find the latency of any dimm you need to take its tCLK and multiply it by its listed timings to find the latency of that timing.

Going by this and the fact an AMD platform is only duel channel going from DDR2 to DDR3 have very little benefit. Am I correct?

 

[Wile E]

AMD gets a small boost from DDR3.

DDR3 is worth it to me just because it is getting cheap, and it runs a lot cooler.

 

[Deleted member 67555]

I guess if you consider 7-12% overall small..

Sure some programs wont even have improvement at all.

But tbh an fair it's better to increase your storage transfer rates before your memory.

 

[Wile E]

I guess if you consider 7-12% overall small..

Sure some programs wont even have improvement at all.

But tbh an fair it's better to increase your storage transfer rates before your memory.

7-12% in synthetics maybe. Not nearly that much in real world apps.

 

[mastrdrver]

One of the reasons (there are others too) for going to DDR3 (if I understand what I've read correctly) is that 4GB DDR2 dimms is the limit because of the way it retrieves data. With DDR it was 2GB.

Again I'm not 100% on that part but it is from what I've come to understand from what I've read. I'm not 100% on why the way the data is retrieved has anything to do with the size limit of the dimm.

If any one wants to make their mind bleed from trying to understand memory read: Everything you always wanted to know about sdram memory but were afraid to ask on Anandtech. Also their article ASUS ROG Rampage Formula: Why we were wrong about the Intel X48 may need reading first to be able to follow along in the other.

I got to about page 4 before I said f- this and then got saved on page 5 with the Youtube video. I kind of skimmed from there because it was stretching my mind mentally trying to follow along.

 

[lane]

I kind of agree, one of the things making it easy for me to wait to go with dd3 is the timings, when i move to ddr3 i want 8gb across 2 modules and the best that's easily available to me is 2000mhz at 9-10-9-27, I'm sure running it slower than 2ghz would possibly let me lower the timings but going from 5-5-5-15 ddr2 i would kind of want at least cas 6 or 7 with ddr3.

I hope in the coming months more memory will be released with lower timings.

I have lower timing access (memory access in ns) ( Aida64 ) with 1600mhz modules @ 9-9-9-24 i had with DDR2 ultra low latencies ... (Micron D9 )

now i use 1600mhz 6-8-6-24 modules DDR3 triple channel, and they keep Cas6 untill 1800mhz... and Cas7-8 @ 2000mhz ...

You can 't compare latencies using CAS between different type of memory, DDR1-2-or 3 .... it's a false idea.

for give you an idea: Sandrasisoftware give me a 29go/s bandiwth with my DDR3@1600mhz ( CPU@200x20).... how much you have with DDR2 ? 15go/s? ( this have nothing to do with latencies, but well it's just for information )

 

[mastrdrver]

Even worse is that the wiki articles on ddr/ddr2/ddr3 confuses memory latency with timings making the claim that higher timings means higher latencies. :shadedshu

 

[Wile E]

Even worse is that the wiki articles on ddr/ddr2/ddr3 confuses memory latency with timings making the claim that higher timings means higher latencies. :shadedshu

Well, if all else is equal, it does.

 

[mastrdrver]

Problem is they rarely are.

Best quote:

While the typical latencies for a JEDEC DDR2 device were 5-5-5-15, the standard latencies for the JEDEC DDR3 devices are 7-7-7-20 for DDR3-1066 and 7-7-7-24 for DDR3-1333.

The DDR3 article mixes latency with dimm clock cycles (tCLKmin) needed to complete an action (CAS, RAS, etc). This would only be true if tCLKmin was equal to 1ns, but that only happens at 2000mhz for DDR3. Just 2 paragraphs later they get it right:

As with earlier memory generations, faster DDR3 memory became available after the release of the initial versions. DDR3-2000 memory with 9-9-9-28 latency (9 ns) was available in time to coincide with the Intel Core i7 release.[7] CAS latency of 9 at 1000 MHz (DDR3-2000) is 9 ns, while CAS latency of 7 at 667 MHz (DDR3-1333) is 10.5 ns.

The first part should say "typical timings" instead of "typical latencies" because it makes it sound like 5-5-5-15 is the latency but its not. Its just the number of cycles need to complete that action at the given speed (i.e 800mhz for ddr2). The latency changes once the speed goes to 1066mhz for DDR2 and it would be a mistake to say 5-5-5-15 equals latencies.

Example:

DDR2 Timings of 5-5-5-15 at
-------------------------------------------------
1066mhz is about 9.5 - 9.5 - 9.5 - 28 (in nanoseconds)
800mhz is about 12.5 - 12.5 - 12.5 - 37.5 (in nanoseconds)

DDR3 Timings of 7-7-7-20 at
-----------------------------------------------------
1066mhz is about 9.5 - 9.5 - 9.5 - 28 (in nanoseconds)
1333mhz is about 10.5 - 10.5 - 10.5 - 30 (in nanoseconds)

This also shows that not only is 1066 DDR3 marginally faster (then 1333 DDR3) given those timings but, unless you have some bandwidth sensitive real world application, there is no real gain by going to 2000mhz DDR3 CAS 9 dimms when the latencies between the two are almost the same. The difference only shows up if the program moves a lot of data in and out of memory otherwise its just wasted (as in most cases). This is also why lower the refresh (tREF) makes the system "feel" fast and more responsive because the system is waiting less and less for the data to be refreshed and minimal changes show up quicker.

 

[Wile E]

Problem is they rarely are.

Best quote:


The DDR3 article mixes latency with dimm clock cycles (tCLKmin) needed to complete an action (CAS, RAS, etc). This would only be true if tCLKmin was equal to 1ns, but that only happens at 2000mhz for DDR3. Just 2 paragraphs later they get it right:



The first part should say "typical timings" instead of "typical latencies" because it makes it sound like 5-5-5-15 is the latency but its not. Its just the number of cycles need to complete that action at the given speed (i.e 800mhz for ddr2). The latency changes once the speed goes to 1066mhz for DDR2 and it would be a mistake to say 5-5-5-15 equals latencies.

Example:

DDR2 Timings of 5-5-5-15 at
-------------------------------------------------
1066mhz is about 9.5 - 9.5 - 9.5 - 28 (in nanoseconds)
800mhz is about 12.5 - 12.5 - 12.5 - 37.5 (in nanoseconds)

DDR3 Timings of 7-7-7-20 at
-----------------------------------------------------
1066mhz is about 9.5 - 9.5 - 9.5 - 28 (in nanoseconds)
1333mhz is about 10.5 - 10.5 - 10.5 - 30 (in nanoseconds)

This also shows that not only is 1066 DDR3 marginally faster (then 1333 DDR3) given those timings but, unless you have some bandwidth sensitive real world application, there is no real gain by going to 2000mhz DDR3 CAS 9 dimms when the latencies between the two are almost the same. The difference only shows up if the program moves a lot of data in and out of memory otherwise its just wasted (as in most cases). This is also why lower the refresh (tREF) makes the system "feel" fast and more responsive because the system is waiting less and less for the data to be refreshed and minimal changes show up quicker.

I understand that. Was essentially just pointing out that your statement was a little vague.