Share your AIDA 64 cache and memory benchmark here

[CraZyNoMaD]

 

[Zemach]

3600 Cl 16 OC 5000 Cl16

 

[NoJuan999]

New results with tighter timings for my G.SKILL TridentZ RGB (2x8GB) F4-3200C16D-16GTZR @ 3400 MHz:

 

[phanbuey]

3600 Cl 16 OC 5000 Cl16View attachment 121371

Is this real life?

holy crap. I actually had to open CPU-z on my rig to make sure i read that right.

 

[pippinfort]

Great thread. I've been doing a bit of testing on my X99 recently. DDR4-3200 with 5930K, 5960X and 6850K.

Ram is Corsair Vengeance® LPX 16GB (4 x 4GB) DDR4 DRAM 3200MHz C15

VENGEANCE® LPX 16GB (4 x 4GB) DDR4 DRAM 3200MHz C15 Memory Kit - Black

VENGEANCE LPX memory is designed for high-performance overclocking. The heatspreader is made of pure aluminum for faster heat dissipation, and the eight-layer PCB helps manage heat and provides superior overclocking headroom.

www.corsair.com

5930K x44 CPU x42 Cache

5960X x44 CPU x40 Cache

6850K x45 CPU x39 Cache

From Asus Sabertooth X79 with E5-1650v2 CPU x45

Ram is Corsair Dominator® Platinum — 16GB (4 x 4GB) DDR3 DRAM 2133MHz C9 Memory Kit

DOMINATOR® PLATINUM — 16GB (4 x 4GB) DDR3 DRAM 2133MHz C9 Memory Kit

DOMINATOR® PLATINUM high-performance DDR3 memory is designed for world-class system builds. It's built with hand-screened ICs, undergoes rigorous performance testing, and incorporates state-of-the-art cooling for reliable performance in demanding environments.

www.corsair.com

From Asus P9X79 with E5-2660v2 stock.

RAM is Hynix DDR3 PC3-14900 1866MHz 4x 4GB

 

[metalfiber]

G.SKILL Ripjaws V Series 32GB DDR4 3333 4 x 8GB

 

[KMBakPL]

Hi, it is my first post on this forum.

I have i7-8086K on stock clock.

I remember (or I think that I remember) that just after mount my new rig (September 2018) my CPU bench was bit over 8700K. Now bit below.


Cache performance are much lower then in results in this posts: https://www.techpowerup.com/forums/...ry-benchmark-here.186338/page-31#post-3938375

Also results in this review are higer then my CPU. I am a bit confused by these discrepancies. I used other version of AIDA... But...

What do You think? Any advices?

And one more observation.
Turning on virtualization (@bios) causes a small drop in AIDA results.

 

[mapnam charun]

 

[er557]

This is the current result, after latest windows updates, and all mitigations of new vulnerabilities enabled, however, @EarthDog maintains this is not per se eight channel ram, but two quad channel implementations(what's the difference?).

So, for comparison I provide the benchmark with only one cpu, yet still buffered and multi threaded, as it is supposed to sqew the results, and it seems it is the proper quad channel speed, about half

Also, at those ddr4 speeds, 2133~2400, it is very hard to reach above 60gb/s without it being more then four channels(i.e octal channel)
and of course the i7 5820k result @ddr4 2400 quad channel shows it very well.

So, despite the large latency, I think the dual processor system does work in at least virtual eight channels.

 

[Vlada011]

Your North Bridge Clock is very low. You run Benchmark probably on Balanced Power Mode???
OC Cache Frequency you will get higher speeds with 2400MHz. My memory results are very similar without OC Cache, 2666MHz.
With Cache to 4.0GHz go up to 61-62.000 all settings write-read-copy.

Change Bus Speed on 100.1MHz in BIOS, on that way you will get 3300MHz Default speed, NOT 3250MHz.
But guys with 4x 8GB kits 3200MHz and special if they use Broadwell-E or i7-5960X could reach and 70-75.000MB/s, on double sided B-Die 8GB per DIMM kits could be reach faster speeds. But they arrive later 2016 I think. In 2014 we didn't had so much choices.

 

[er557]

I run ultimate performance power scheme(windows enterprise), and the north bridge clock is maxxed out by efi patch driver, pegged @3000 mhz,

edit: Oh i see you are reffering to the i7 system, I dont have it anymore, it is an old screenshot, I was talking about the xeon.

 

[EarthDog]

however, @EarthDog maintains this is not per se eight channel ram, but two quad channel implementations(what's the difference?).

This isnt the place for this discussion but since you brought it up here...I'll repeat what I found. Maybe someone else can clarify.

All I know is what I quoted from Intel about the cpus (quad channel) AMD the motherboard (quad channel).

Product Specifications

quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

Max # of Memory Channels 4

ASRock Rack > EP2C612 WS

www.asrockrack.com

1. Supports Quad Channel DDR4 2400/2133/1866/1600 RDIMM and LRDIMM, 8 x DIMM slots

EDIT: As I said in the other thread, AIDA64 memory tests scale with cores/threads. Drop in a higher core count CPU (as opposed to the 5820K) and bandwidth improves.

 

[er557]

Indeed, but intel spec refers to ONE processor, how many channels are TWO? they do work together via QPI links to enable both cpus to access all ram and share it via NUMA, each of the procs providing quad channel, working TOGETHER.
As to the motherboard spec, I will check this directly with ASROCK, as the benchmarks show the actual situation.

 

[EarthDog]

Two processors are dual quad channel AFAIK.

I quoted the ASRock motherboard page above. One would think that if it supported actual "octo" channel memory, it would say octo channel on the page, no?

What does CPUz say the memory is running as?

I could be wrong...just waiting to see something more compelling than a single piece of software showing an aggregate of the system.


EDIT: Again, PM me when you have something as this thread is for results, not discussion of quad/octo/NUMA.

 

[heky]

5960X @ 4.5Ghz, Corsair Vengeance 32Gb 3200 c16 2666 c13

Does someone have a clue how to raise the memory write performance on Haswell-E?

 

[er557]

can you overclock the ram a bit? also your performance is fine already, for what purpose do you want it raised?

 

[heky]

can you overclock the ram a bit? also your performance is fine already, for what purpose do you want it raised?

Thats the thing, the ram is actually 3200mhz c16, but is not stable with the xmp enabled (Its dual-rank Samsung E-die). Its rock stable @ 2666mhz c13. There is no particular purpose as to why i want it raised, just one of those "why not" moments.

 

[EarthDog]

You need to add some voltage for the memory controller I would imagine. 3200 C16 32GB is possible on that platform.

But that is the way to do it... running it at speed.

 

[heky]

You need to add some voltage for the memory controller I would imagine. 3200 C16 32GB is possible on that platform.

But that is the way to do it... running it at speed.

I know it should be possible, but the thing is that my motherboard (AsRock Taichi) doesnt show the actual SA voltage, just an offset value. And since i dont know what i am offsetting to, i am affraid to give it more than +0.100v, since i really dont want to fry the 5960x`s memory controller. I read that happened quite a lot on the x99 platform.

The other thing is the XMP profile actually sets the SA voltage to +0.350 or so i think, but the ram is not stable with the profile enabled.

 

[EarthDog]

Try VccIO voltage as well...

But I would think with 0.350+ offset that should be plenty... maybe it just isn't in the cards. Did you try the same timings (as XMP) but at 3K?

 

[heky]

Yeah i tried 2800 and 3000, but no go for both speeds at XMP timings, which is really weird, since 2666@13-15-15-30-1t should be harder on the memory than say 2800@16-18-18-36-2t. Maybe i should try playing with the CPU strap?

 

[1986nath]

 

[Zemach]

DDR4 5162 cl 17 16 16 28

 

[KMBakPL]

Are L1, L2 and L3 caches speed related with RAM speed? I see CPU on stock clock and caches so much faster then my CPU... ?

 

[heky]

Are L1, L2 and L3 caches speed related with RAM speed? I see CPU on stock clock and caches so much faster then my CPU... ?

They are related to uncore (cache) speeds...so in Zemach`s case that is running at 4860mhz.