.

[tidy]

.

 

[Concerned Citizen]

dood
Them ain't no killer RAMS.
There will be no more,you're 2-3 years late.
You ain't got no Sansungs that run @ 10-12-12-37 2800 daily.
Chances are the chip would.Your RAMs will not.
Oh DDr4
Hmm
TridentZ cas 16-17-17-something.
That's what holds the world record, too.
I have a kit.I bought a "Higher binned" kit.The 3600 kit is better.
$124 it's a 3600 kit.Trident Z 3600 CAS 16.
Latency on ddr3 is better.
cache is better on newer stuff

 

[tidy]

.

 

[Concerned Citizen]

well,32-3600 is easily doable.yannow?
wait,are you on ddr3 or 4?
for ddr3 you need the Samsungs that are CAS10 @ 2600 Mhz
10-12-12 28 2600
They sold about 3 years ago so ebay is the place.
They come in 4gb sticks.
I wonder what the chances are I could return 3 :?
What's your cache clock?

 

[Hail The Brain Slug]

I'm not interested in world records or overclocking RAM. 2400MHz is Intel's maximum officially supported spec for this chip (as 1600MHz was for the 4790K). Presumably their 76.8 GB/s figure is based on 2400MHz RAM. I'm wondering why mine is so far off. I doubt tighter timings would make up the 30GB/s difference.

What are you using to measure your memory bandwidth? If you have AIDA64, can you run the cache and memory benchmark and post the result for us to look at?

 

[tidy]

.

 

[Hail The Brain Slug]

I can give you a shot of the trial version ($40 is a bit steep for a piece of software I'll probably use twice a year):

Since you have a trial version, if you go in the left navigation menu and expand 'Benchmarks' you can manually select each of the memory benchmarks and run them individually for results. Go through all four and post them up here.

 

[mnewsham]

With DDR4 3200MHz here is what I get

Read
Quad DDR4-3200 16-18-18-38 CR2 54467 MB/s

Write
Quad DDR4-3200 16-18-18-38 CR2 47993 MB/s

Latency
Quad DDR4-3200 16-18-18-38 CR2 60.1 ns

 

[dave_the_nerd]

Overclocking doesn't really matter, since that changes the maximum throughput as well. The real question is "Why does quad-channel DDR4 seem to only benchmark at about 60% of its theoretical max in synthetic benchmarks?"

After perusing a few benchmarks/reviews, it looks like that's about right for quad-channel DDR4. In the ballpark anyways.

http://www.guru3d.com/articles-pages/core-i7-5960x-5930k-and-5820k-processor-review,13.html

But yes, the theoretical max throughput is that 76.8GB/sec number. I just have no idea what additional overhead on the bus is preventing people from hitting it. But OP can at least relax, since he's not nutso and there's nothing wrong with his system in particular.

My guess is that the higher latency/timings of DDR4 are slowing you down. If you could get some 10-10-12 DDR4 running at 2400MHz, it's probably be a different story.

You could also try a member benchmark like the one build into memtest. Less stuff going on w/ the system should mean a benchmark score closer to the theoretical max.

 

[Hail The Brain Slug]

I was going to wait until I had a friend's AIDA64 run on his 6800K as he has almost the same MB and identical RAM and settings to my 5960X to illustrate a point. I will update once I get it.

It would seem a lot of factors can play a role in observed memory bandwidth. Since I got my 5960X setup in November 2015, I've seen wild variances in my memory bandwidth up and down after BIOS and CPU microcode updates. With similar settings I've had my bandwidth range from 64 GB/s to the current 85 GB/s I have now.

Interestingly enough, Haswell-E got a mysterious bandwidth bump after Intel released a microcode update and ASUS (and hopefully other manufactureres) pushed it out on a bios update. See here.

If you have a Haswell-E, you might see about making sure you update to the latest bios for your board to make sure you get the microcode update, given your manufacturer pushed it out.


Regarding Broadwell-E, it seems a lot of people have noticed poor observed memory bandwidth. My buddy's 6800K system with identical memory and memory settings to mine scores in the 58-65GB/s range in AIDA64 compared to, well, what you see in my other thread. It doesn't appear to be hurting the performance of BW-E at all, however through BIOS and microcode updates, we might see the performance go up over time.

 

[mnewsham]

If you have a Haswell-E, you might see about making sure you update to the latest bios for your board to make sure you get the microcode update, given your manufacturer pushed it out.

Which seems fairly hit or miss, and if they do it's not like they include that information in the release notes, almost all of my BIOS updates are for "Improved memory compatibility" and adding broadwell-E/EP support going back a few months.

 

[Hail The Brain Slug]

Which seems fairly hit or miss, and if they do it's not like they include that information in the release notes, almost all of my BIOS updates are for "Improved memory compatibility" and adding broadwell-E/EP support going back a few months.

You can check your microcode revision in HWINFO64. The latest HW-E code is 38 as pictured here:

Obviously it would be nice if they listed the microcode update in the notes, but ASUS didn't list it either despite rolling it out on all their x99 motherboards. Your only choice is to flash and wish. If you do go for it and get the microcode update, please let us know how your bandwidth changes.

The ASUS update was around 7/25. I see there is an MSI bios update for your board from around the same time. Give it a whirl!

 

[tidy]

.

 

[Hail The Brain Slug]

Here are the other AIDA64 memory results:

Read: 48814 MB/s
Write: 60788 MB/s
Copy: 57399 MB/s
Latency: 71.9 ns

If this comes down to waiting for BIOS updates, then I don't think much of my chances. It seems like my board is really unpopular, and it's BIOS gets less attention from ASUS as a result. There have been barely any revisions as far as I can tell, although the BIOS download section is absent from the board's web page (which is telling in and of itself).

Those look right about in line with what I've been seeing others reporting around the web for BW-E.

As for bios updates, there is one newer that was released on 8/29 for your board here. If you're up for the update, you could give it a shot. There's really no telling what will change with each bios update, ASUS seems to do a lot of work on them.

In general, low bandwidth scaling from dual to quad channel has been a complaint ever since there were quad channel consumer level parts. a ~50% increase over dual channel bandwidth is what the historic return seems to have been, and HW-E wasn't an exception until the microcode update.

On a side note, I would not be surprised if the bios and other updates for your board are very slow. It's intended as a workstation board, so stability and time-tested updates are very critical. The other boards are the testing ground for the latest updates, and lots of people have issues if they stay on the leading edge of them.

 

[tidy]

.

 

[Hail The Brain Slug]

How did you find that BIOS?

If you click the "GO" button on any of the CPUs listed on the compatible list, it takes you to a link to download the latest bios.

Edit: To be clear, I don't think updating your bios today is going to give you better memory bandwidth. I think your bandwidth may change over time with successive bios updates and maybe Intel will even update the microcode like they did with HW-E to improve the performance of the memory controller. The haswell update came out almost two years after the CPUs did.

Unfortunately for all the increases in memory performance I've gotten, the change in overall performance is negligible. You can find articles showing little to no performance difference between dual and quad channel DDR3/DDR4. The bandwidth being a little low really shouldn't affect your performance.

For funsies, when I am finished with work today I'm going to clock my memory down to the same configuration as yours. I wouldn't be surprised if mine didn't really perform all that much better.

 

[mnewsham]

You can check your microcode revision in HWINFO64. The latest HW-E code is 38 as pictured here:

Obviously it would be nice if they listed the microcode update in the notes, but ASUS didn't list it either despite rolling it out on all their x99 motherboards. Your only choice is to flash and wish. If you do go for it and get the microcode update, please let us know how your bandwidth changes.

The ASUS update was around 7/25. I see there is an MSI bios update for your board from around the same time. Give it a whirl!

Well I updated my BIOS, I was on the one that came out just before the latest one. The latest BIOS does include the microcode update, but because I currently only have 3x8GB instead of 4x8GB because one of the sticks is being RMA'd, my results are hard to compare to the previous.

I got a memory read speed of 51620MB/s So while that is slower than my previous score, it is triple channel vs quad channel, so if I had all 4 sticks i'd probably have a nice performance boost.


What I wasn't expecting from the BIOS update was windows decided my computer was no longer activated and I had to call M$ in order to get it re-activated. Took about 10-15 minutes on the phone with Paul in India.

 

[Hail The Brain Slug]

Well I updated my BIOS, I was on the one that came out just before the latest one. The latest BIOS does include the microcode update, but because I currently only have 3x8GB instead of 4x8GB because one of the sticks is being RMA'd, my results are hard to compare to the previous.

I got a memory read speed of 51620MB/s So while that is slower than my previous score, it is triple channel vs quad channel, so if I had all 4 sticks i'd probably have a nice performance boost.


What I wasn't expecting from the BIOS update was windows decided my computer was no longer activated and I had to call M$ in order to get it re-activated. Took about 10-15 minutes on the phone with Paul in India.

That's a nice result. I would expect it to be around 65GB/s+ with all four channels going.

 

[Hail The Brain Slug]

I as closely emulated your settings as I could. Same cache speed, core turbo profile, and memory settings. As you can see, my performance tanked.

I looked around and it would seem some of my performance drop was due to lowering my cache speed. I notice AIDA is reporting yours as 2800MHz, is that the default cache max multiplier? I can't find any documentation on what the cache speeds are supposed to be for those processors.

If you could increase your cache speed, I get the idea your memory bandwidth might get a little boost. If 28x is the maximum stock setting, that would count as overclocking.

 

[Hail The Brain Slug]

I've got a HW-E to BW-E comparison now.

I changed my core and cache configuration to match his setup to eliminate cache speed as a variable for the memory bandwidth. Both setups have identical memory and memory configuration. Almost identical MB (x99-A vs x99-A II).

Both are on the latest bios available.

HW-E

BW-E

 

[tidy]

.

 

[Hail The Brain Slug]

I updated my BIOS. The microcode revision didn't change. My AIDA64 memory bench results improved marginally:

Read: 50604 MB/s
Write: 61010 MB/s
Copy: 57679 MB/s
Latency: 68.5 ns

Incidentally, the new BIOS didn't fix my turbo issues either (turbo is stuck at 3.8GHz and never moves, as per the other thread). My hopes had risen briefly when I saw Turbo mentioned in the release notes.

Trust my luck to finally get around to buying HEDT parts, purely for the memory bandwidth, and just happening to get the generation with retrograde memory performance.

That's actually a pretty nice bump from what appears to be a minor bios revision. A few GB/s plus shaving off a new ns of latency sounds great to me.

The platform is very new, so I'm sure it will just take time. They iterate every bios update and if Intel can, I'm sure they will update the microcode to make improvements. Like I said, the HW-E update was nearly 2 years after launch.

After the update, I wouldn't bat an eye at your benchmarks given you aren't overclocking.

Edit: Consider this. From what I have seen, a 6700K with dual channel DDR4-2400 only scores around 33-35GB/s on the bandwidth benchmarks. You're getting ~60-70% increased bandwidth over that. Your read is a little low but it looks like it's a characteristic of the BW-E memory controller in general as all the benchmarks I've seen are low for read.

 

[tidy]

.

 

[ashetos]

Don't know how the AIDA benchmark really works but in my experience you need a multi-threaded benchmark to really get the most out of your memory's bandwidth. Writing a multi-threaded C program could possibly give better results.

 

[Hail The Brain Slug]

Don't know how the AIDA benchmark really works but in my experience you need a multi-threaded benchmark to really get the most out of your memory's bandwidth. Writing a multi-threaded C program could possibly give better results.

It's much better than the stream benchmark in Geekbench. My 82 GB/s copy from AIDA turns into 27.9 GB/s in Geekbench Multithreaded.

Watching CPU usage, it definitely uses several threads. CPU load ranges from 40-100% during testing.