MP 7,1 4 VS 6 channels RAM in Real-world Lightroom TEST! It is 39.2% FASTER?! [UPDATE RAMDISK TEST]

[zhpenn]

You are correct in that it is impossible to configure a four and six channel configuration with the same memory capacity. The best you could do is attempt to configure the system as close in capacity as possible.

Since the theory is the speed gain could be the result of memory caching the configuration would be such as to minimize it if possible. If I recall correctly the reference configuration was 128GB (4 x 32GB = 128GB). Thus for a six channel configuration you would configure 96GB (6 x 16GB = 96GB). If the six channel configuration was still 39% faster that strongly suggest caching isn't the reason for the gain.

Of course we could always go the other way and configure a four channel configuration with more than 192GB. The closest we could configure is 256GB (4 x 64GB = 256GB), If caching were the reason I would expect to see little, if any, increase in speed with the additional memory.

None of these are ideal but I would be willing to concede six channels is 39% faster if a 96GB six channel configuration equals the speed of a six channel 192GB configuration.

Yes, but you need to buy new DIMMS for those test, I'm not a professional system tester. I just see the differences and sharing here, the closet way I can do is to limited the avaliable free RAM space in a software way, and it is clear enough in the activity monitor how many ram is used and free space and no swap memory, and LR has a lot more than enough RAM Space it need to play with, in all those configurations, so and I dont see any "unfair" that I use 92gb of 6 channels vs a 192gb of 6 channels have almost the same results. maybe still some other factors that need to be consider, but this I think is a close enough way to testing without buying new DIMMs.

also there is also another bandwidth test here, can see how big different can it be

2019 Mac Pro: Memory Bandwidth

macperformanceguide.com

 

[defjam]

Yes, but you need to buy new DIMMS for those test, I'm not a professional system tester. I just see the differences and sharing here, the closet way I can do is to limited the avaliable free RAM space in a software way, and it is clear enough in the activity monitor how many ram is used and free space and no swap memory, and LR has a lot more than enough RAM Space it need to play with in all configuration, so and I dont see any "unfair" that I use 92gb of 6 channels vs a 192gb of 6 channels have almost the same results.

also there is also another bandwidth test here, can see how big different can it be

2019 Mac Pro: Memory Bandwidth

macperformanceguide.com

As such you shouldn't be making blanket statements that six channel memory provides a roughly 39% performance increase over a four channel memory configuration.

 

[zhpenn]

For me, the best way to test memory bandwidth comparing number of memory channels being employed is as follows...

Using 32 GB R-DIMMs (as that is what I have)

1 Memory Chanel used

1. Place 32GB R-DIMMs in slot 5 & 8
2. Run STREAM to obtain best memory bandwidth

2 Memory Chanels used

1. Place 32GB R-DIMMs in slot 3, 5, 8 & 10
2. Run STREAM to obtain best memory bandwidth

3 Memory Chanels used

1. Place 32GB R-DIMMs in slot 1, 3, 5, 8, 10 & 12
2. Run STREAM to obtain best memory bandwidth

4 Memory Chanels used

1. Place 32GB R-DIMMs in slot 1, 3, 5, 6, 7, 8, 10 & 12
2. Run STREAM to obtain best memory bandwidth

5 Memory Chanels used

1. Place 32GB R-DIMMs in slot 1, 3, 4, 5, 6, 7, 8, 9, 10 & 12
2. Run STREAM to obtain best memory bandwidth

6 Memory Chanels used

1. Place 32GB R-DIMMs in slot 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 & 12
2. Run STREAM to obtain best memory bandwidth

im thinking, if I move Only one DIMM to the others slot,
was Place 32GB R-DIMMs in slot 1, 3, 5, 8, 10 & 12
Now Place 32GB R-DIMMs in slot 2, 3, 5, 8, 10 & 12,
Can it be turn on and use all the DIMMs but less bandwidth?

 

[AidenShaw]

Yes, but you need to buy new DIMMS for those test, I'm not a professional system tester.

Actually, you started out doing the right thing - testing an application that seems to be important to you.

Benchmarks like GeekBench and STREAM are written to bypass the caches and drill down to the raw streaming memory bandwidth.

Applications, however, utilize cache - therefore your attempts to measure LightRoom are good. Unfortunately, you had an experiment in two variables - number of channels and total amount of RAM.

 

[curmudgeonette]

im thinking, if I move Only one DIMM to the others slot,
was Place 32GB R-DIMMs in slot 1, 3, 5, 8, 10 & 12
Now Place 32GB R-DIMMs in slot 2, 3, 5, 8, 10 & 12,
Can it be turn on and use all the DIMMs but less bandwidth?

These should be the same bandwidth, except...

...the second configuration might not work, or might cause the memory to be down-clocked: Presumably, Apple wired the DIMM sockets by routing from the Xeon to slot 2, and then on to slot 1. Putting a DIMM in slot 2 but not slot 1 will mean the wiring between these slots forms an unterminated transmission line. As such, a signal will bounce off the end of the wire and the (inverted) reflection will come back to slot 2. This may prevent proper functioning of the RAM, or might require a slower clock to give time for the reflection to be damped.

 

[zhpenn]

These should be the same bandwidth, except...

...the second configuration might not work, or might cause the memory to be down-clocked: Presumably, Apple wired the DIMM sockets by routing from the Xeon to slot 2, and then on to slot 1. Putting a DIMM in slot 2 but not slot 1 will mean the wiring between these slots forms an unterminated transmission line. As such, a signal will bounce off the end of the wire and the (inverted) reflection will come back to slot 2. This may prevent proper functioning of the RAM, or might require a slower clock to give time for the reflection to be damped.

I tried Now Place 32GB R-DIMMs in slot 2, 3, 5, 8, 10 & 12,
the amber light is on, can not turn on the mac

 

[Snow Tiger]

I tried Now Place 32GB R-DIMMs in slot 2, 3, 5, 8, 10 & 12,
the amber light is on, can not turn on the mac

you are using the wrong half of channel F . Use the primary ( number 1 ) half of each channel with six matching modules .

 

[zhpenn]

you are using the wrong half of channel F . Use the primary ( number 1 ) half of each channel with six matching modules .

View attachment 892148

Yes, because I want to try using less bandwidth but same amount of RAM to do the test again,if I put all channels on 2 will it be turn on?

 

[zhpenn]

Just Saw this thread there, there a memory bandwidth documents shows different DIMMs setup differences

Best RAM configuration for the 2019 Mac Pro

Yes Definitely... 4xDimms is actually only 66% Performance ( bit more in reality ) you have 6 channels of memory access and you are Currently using 4. it's like 95% performance with 6 sticks and 100% with all 12 ( so long as they are all the same ) From page 18...

forums.macrumors.com

From page 18

https://lenovopress.com/lp1089.pdf

(actually that whole section explained it all... balanced and near balanced etc)

6channels 95% performance
4channels 66% performance ( bit more in reality )
That is 43.9% increase, which is similar to the test results

 

[defjam]

I tried Now Place 32GB R-DIMMs in slot 2, 3, 5, 8, 10 & 12,
the amber light is on, can not turn on the mac

My recommendation would be to put everything back to the way it was and enjoy your new system. Finding the answer appears to be taking a lot of your time.

 

[Snow Tiger]

My recommendation would be to put everything back to the way it was and enjoy your new system. Finding the answer appears to be taking a lot of your time.

He's playing with his new toy .

I remember when I first got my Apple II back in the early 1980s . I bought a directory of all the 8 bit programs that were available for it . If memory serves , there were 20,000 apps and utilities available then all listed in a book the size of a good ol' fashioned phonebook . That was quite a selling point for Apple back in the day . I read that darn book cover to cover . What can I say ? I'm a geek ? .

 

[defjam]

He's playing with his new toy .

If he wants to do it far be it from me to tell him not to. However I didn't want him to feel obligated on my behalf.

I remember when I first got my Apple II back in the early 1980s . I bought a directory of all the 8 bit programs that were available for it . If memory serves , there were 20,000 apps and utilities available then all listed in a book the size of a good ol' fashioned phonebook . That was quite a selling point for Apple back in the day . I read that darn book cover to cover . What can I say ? I'm a geek ? .

I hear ya. My current projects (3 SEs and 3 SE/30s):

 

[Snow Tiger]

If he wants to do it far be it from me to tell him not to. However I didn't want him to feel obligated on my behalf.


I hear ya. My current projects (3 SEs and 3 SE/30s):

View attachment 892192

be careful with those CRTs - but , you probably already know .

Here's a cool pile of junk . All my G4 projects I abandoned when the Mac Pros first got released . Buried in there somewhere is also a working Quadra 700 , a working Macintosh IIx , a MDD G4 with a dual Sonnet CPU upgrade , a 1.8 GHz PMG4 Digital Audio with a genuine GeForce 3 Mac Edition and goodness knows what else .

I hardly touched this stuff since 2007 - 08 . Every now and then I'll fire one up and - yup - it still works . I cut my teeth learning how to remanufacturer Macs with these units - a skill that became useful later on . I completely tore them down , cleaned them , upgraded the cooling systems , upgraded the components , added new drives ( some of the G4s have SSDs ) and finally an OS and a few games . Lots of fun .

 

[AidenShaw]

That is 43.9% increase, which is similar to the test results

A couple of possibilities:

• By sheer coincidence, the mismatched memory sizes on your test helped by an amount similar to the raw (uncached) bandwidth differences.
• Your LightRoom test happened to be one of the rare programs that with your script was bandwidth constrained, and didn't utilize the cache.

One can't make any conclusions based on a single program - unless that program and the script that you ran does indeed reflect your critical workflow.

Even GeekBench can be useful if you ignore the two main numbers, and look at the component tests. If most of the component tests show a small difference, and a couple show a big difference - that should be a clue that the results are highly application dependent.

 

[zhpenn]

There is a similar test, which has a similar conclusion

Mac Pro review - A photographer and Videographers' perspective

I am going to update my post but here is a snippet with RAM change to a Single capacity modules as opposed to mixed. As confirmed in other threads, there is a massive boost.

forums.macrumors.com

 

[AidenShaw]

There is a similar test, which has a similar conclusion

With the same problem of having two variables in the experiment....
Sorry, bad arithmetic.

 

[darthaddie]

With the same problem of having two variables in the experiment....

May I ask what are the 2 variables. except the different RAM setup? Its identical in my opinion. Both have 12 Core/96GB/Vega II and exactly the same drive setup etc.
[automerge]1581564797[/automerge]

There is a similar test, which has a similar conclusion

Mac Pro review - A photographer and Videographers' perspective

I am going to update my post but here is a snippet with RAM change to a Single capacity modules as opposed to mixed. As confirmed in other threads, there is a massive boost.

forums.macrumors.com

View attachment 893722

Thanks for linking this.
[automerge]1581564941[/automerge]

May I ask what are the 2 variables. except the different RAM setup? Its identical in my opinion. Both have 12 Core/96GB/Vega II and exactly drive etc.
[automerge]1581564797[/automerge]


Thanks for linking this.

And just to make sure it isn't just Lightroom, Geekbench also sees a performance boost.

Plus Borderlands 3 on Bootcamp saw a 12-15% FPS boost. No changes or updates done on the software side. I just changed the RAM config.

 

[defjam]

With the same problem of having two variables in the experiment....

It's not ideal but I think this test is closer to it than the previous test. It eliminates a key issue I had with the previous test (disparate memory sizes). Outside some laboratory testing I think this is the best we're likely to get and I'm willing to acknowledge the move from four to six channel appears to have significant gains in this specific real world use case. I guess my next question is why is there such a variance between the two? Especially with a 12-core processor.

 

[AidenShaw]

May I ask what are the 2 variables. except the different RAM setup? Its identical in my opinion. Both have 12 Core/96GB/Vega II and exactly the same drive setup etc.

Sorry, I didn't do the math right.

 

[worldburger]

If he wants to do it far be it from me to tell him not to. However I didn't want him to feel obligated on my behalf.


I hear ya. My current projects (3 SEs and 3 SE/30s):

View attachment 892192

Have you heard of (or played with) retr0bright gel? Phenomenal little trick for the Bromine in those cases ?

 

[defjam]

Have you heard of (or played with) retr0bright gel? Phenomenal little trick for the Bromine in those cases ?

I haven't heard of retr0bright gel. I use the traditional hair care product methods which have typically worked well. I'll have to investigate it.