8 PACK MEMORY RANGE GROWING: SAY HELLO TO 8 PACK RIPPED EDITION & 32GB KITS!!!

Skynet5 · 24 May 2020 at 23:57

opethdisciple said:
I'm uptodate on all drivers.

The things I haven't applied from the calculator are the VDDG and VDDP voltages. I've left them at stock.

I've been using the PC over the last 10 days and gaming on it so it's stable.

---

I just tried applying the 'rec' voltages for vddg and vddp and it made no difference what's so ever to my membench score.

To disprove the theory around core count, I could run 16 threads to match yours and see it my number goes up.

Will test tomorrow

opethdisciple · 25 May 2020 at 00:10

lnoton said:
To disprove the theory around core count, I could run 16 threads to match yours and see it my number goes up.

Will test tomorrow

You could do that, but also you could just see what % of the 240% total task each thread is doing.

My prediction is 10%.

Skynet5 · 25 May 2020 at 00:21

opethdisciple said:
I'm uptodate on all drivers.

The things I haven't applied from the calculator are the VDDG and VDDP voltages. I've left them at stock.

I've been using the PC over the last 10 days and gaming on it so it's stable.

---

I just tried applying the 'rec' voltages for vddg and vddp and it made no difference what's so ever to my membench score.

The thing is the test takes 240% divides it by the number of CPU threads and then runs the test. So it makes sense that a CPU with more threads finishes the test faster.

When I run the test each thread does 15% of the task. 15 * 16 = 240

I reckon your 3900x each thread is only doing 10% of the task meaning the job finishes 33% faster.
10 * 24 = 240

I'll check tomorrow, but I'm sure each thread goes to 15% for me.

Ive never really thought about it. Other than what is the point of a memtest bench mark that can't be compared with out cpu consideration.

opethdisciple · 25 May 2020 at 00:23

lnoton said:
I'll check tomorrow, but I'm sure each thread goes to 15% for me.

Ive never really thought about it. Other than what is the point of a memtest bench mark that can't be compared with out cpu consideration.

OK, cool. Check tomorrow. If you think it goes to 15% too that means the test automatically scales the task scope to 15% * threads = taskscope which would make sense.

Shac · 25 May 2020 at 01:45

@opethdisciple
The membench pic you posted earlier it appears you have not imported your XMP from Taiphoon so it is using a generic set of data. (as long as that is your current settings ofc)
You really should import the XMP to at least ensure you have the correct figures generated.

It may not be the same make but we both have good "B" die so I would have thought you should have similar figures generated to mine.

Anyway tras + Trp = tRC so using that data yours should be 45 not 48.
My tRRDS is 4 x 4 =tFAW of 16.
Lower tFAW is good.

Lots of other differences between my data and yours as well.

Your lower scores if stable is nothing to do with the voltages its your timings.

Membench task scope shows 15% for 3900X before you start.
When running it shows 360% which is 15% x 24 so for the extra cores they increase the workload to keep it a fair test.
So unfortunately you need to reckon again

Skynet5 · 25 May 2020 at 11:15

opethdisciple said:
OK, cool. Check tomorrow. If you think it goes to 15% too that means the test automatically scales the task scope to 15% * threads = taskscope which would make sense.

As I thought, and as @Shac mentioned. The test is designed to be comparable across machines with difference core counts.

opethdisciple · 25 May 2020 at 11:57

Shac said:
@opethdisciple
The membench pic you posted earlier it appears you have not imported your XMP from Taiphoon so it is using a generic set of data. (as long as that is your current settings ofc)
You really should import the XMP to at least ensure you have the correct figures generated.

It may not be the same make but we both have good "B" die so I would have thought you should have similar figures generated to mine.

Anyway tras + Trp = tRC so using that data yours should be 45 not 48.
My tRRDS is 4 x 4 =tFAW of 16.
Lower tFAW is good.

Lots of other differences between my data and yours as well.

Your lower scores if stable is nothing to do with the voltages its your timings.

Membench task scope shows 15% for 3900X before you start.
When running it shows 360% which is 15% x 24 so for the extra cores they increase the workload to keep it a fair test.
So unfortunately you need to reckon again

I most definitely imported my timings from taiphoon burner.

My process was:

1. Taiphoon burner report, then show in nano seconds.
2. Export as complete html and import in to the dram calc.
3. Set all my settings: x570, bdie, rank etc.... the only one I dont set is the dram PCB revision. Mine is A0 I think.
When you import the timings it defaults to manual so I left it as such. And when I tried to change it all the timings disappeared.
4. Press fast timings button and then put primary and secondary timings in to bios and set voltage.
5. Finally membench and get a crap score.

What am I missing?

Also my ram is not the 3600CL14 8 Pack ram! Mine is the 3200MHz 8 Pack stuff I have overclocked to similar timings.

These are the timings it gives me after I import using taiphoon burner.

Skynet5 · 25 May 2020 at 12:23

opethdisciple said:
I most definitely imported my timings from taiphoon burner.

My process was:

1. Taiphoon burner report, then show in nano seconds.
2. Export as complete html and import in to the dram calc.
3. Set all my settings: x570, bdie, rank etc.... the only one I dont set is the dram PCB revision. Mine is A0 I think.
When you import the timings it defaults to manual so I left it as such. And when I tried to change it all the timings disappeared.
4. Press fast timings button and then put primary and secondary timings in to bios and set voltage.
5. Finally membench and get a crap score.

What am I missing?

Also my ram is not the 3600CL14 8 Pack ram! Mine is the 3200MHz 8 Pack stuff I have overclocked to similar timings.

I think this is just evidence that timings alone do not define a score. perhaps being a 3200 base kit means that although you can up the spped and low timings, to the new 8 pack kit, doesnt mean you'll get the same score. Other factors are at play, including the quality of the silicon across cpu and ram chips.

we need @Space Monkey to report in with the new kit and his easy and default scores.

Space Monkey · 25 May 2020 at 12:42

lnoton said:
I think this is just evidence that timings alone do not define a score. perhaps being a 3200 base kit means that although you can up the spped and low timings, to the new 8 pack kit, doesnt mean you'll get the same score. Other factors are at play, including the quality of the silicon across cpu and ram chips.

we need @Space Monkey to report in with the new kit and his easy and default scores.

This is stock XMP with 4 sticks (32GB) at 3600C14.

opethdisciple · 25 May 2020 at 12:52

So it seems my rubbish score of 245 is OK. Nothing to be concerned about it seems.

Space Monkey · 25 May 2020 at 12:52

opethdisciple said:
So it seems my rubbish score of 245 is OK. Nothing to be concerned about it seems.

Yeah, I've noticed your sub timings are tighter than my XMP, so seems about right?

opethdisciple · 25 May 2020 at 12:58

Space Monkey said:
Yeah, I've noticed your sub timings are tighter than my XMP, so seems about right?

I hope so. I exported the timings from the dram calc and was concerned by my seemly crap membench score. But comparing to your stock settings it seems OK.

Inline.

Space Monkey · 25 May 2020 at 13:01

opethdisciple said:
I hope so. I exported the timings from the dram calc and was concerned by my seemly crap membench score. But comparing to your stock settings it seems OK.

Inline.

The best time quoted of 206s will have been set by someone like 8Pack as 3600C14 is no slouch, so to get such a disparity between the two timings is not from normal users getting such low scores!

Space Monkey · 25 May 2020 at 13:08

Membench on Easy mode;

opethdisciple · 25 May 2020 at 13:22

Thanks @Space Monkey

opethdisciple · 25 May 2020 at 14:51

So it seems there is another way to use the calculator. Rather than importing timings from taiphoon burner which will revert the PCB revision parameter to manual, if you just select the PCB parameter from the drop down list instead (and don't import anything) and choose the rest of the settings it will spit out a different set of timings. If you try it your see what I mean. If you import your timings and then change the PCB revision it will reset everything to 0.

Or maybe I have the order of this wrong.

Maybe you have to set all the settings in the calculator including the PCB parameter and THEN import.

The release notes for 1.7.1 says this:

DRAM PCB Revision - select PCB RAM for more accurate calculation of timings. In most cases it is recommended to use "A0" for better compatibility.

But as I said, import your timings and it goes to manual. Change the PCB revision and it changes to 0.

Skynet5 · 25 May 2020 at 15:19

Space Monkey said:
Membench on Easy mode;

I think tfaw and trfc tweaking will massively improve your scores.

Trrds seems high. Aim for 4 and trrds*4 for tfaw.

Then get trfc to about 300-350.

Youll knock 20seconds off...

opethdisciple · 25 May 2020 at 15:46

@lnoton I think what affecting the membench score is not the latency of the ram but the bandwidth. You have an FCLK of 1866MHz if I am correct. This could be why you get a much better score than me.

Skynet5 · 25 May 2020 at 15:56

opethdisciple said:
@lnoton I think what affecting the membench score is not the latency of the ram but the bandwidth. You have an FCLK of 1866MHz if I am correct. This could be why you get a much better score than me.

Aye. This is true. I can actually run my FCLK at 1900 full stable it appears. At least running known stress test programs for 4+ hours works. But I keep it 1 click down just for the comfort factor.

Must say, if it wasn't for the 3900XT rumour I felt like I bought into this platform at the right time for me and my wallet.

doopydug · 25 May 2020 at 16:01

I am thinking of using this memory with the new X570 MSI Tomahawk.

Question is, will I be better with the 8 PACK 3600 RAM or the 4000 RAM with slightly slower timings?