alternatives to mac pro

[h9826790]

i don't mind to debate it, i think everyone finds it interesting and is having fun to try and prove me wrong! the debate is complicated to follow as it gets longer so i am going to put underline and bold on things which i think are the most important points or can be confusing.



the performance of non-ecc memory in games is well understood because many benchmarks have been done on it. but i don't think you will find many gaming benchmarks of ecc memory, and that's what we are interested in. you can't assume that the trends in benchmarks of non-ecc ram will match that of ecc, they are two separate things.

its also important to know that the benchmarks you posted earlier of ecc ram were done on ddr3 modules, not ddr4. nobody is using ddr3 anymore, we really need tests showing ddr4. do we assume dual-channel ddr3 behaves the same way as 6-channel ddr4 on some test? we can't assume that, we would be guessing.

there are some benefits to faster ram speeds that while small aren't completely trivial. i could try to dig up that info if anyone wants to know about it. i think the benchmarks are all for non-ecc ram though.



performance of games is tricky. some games are more CPU than GPU dependant, and vice versa. some games will run better on ati vs nvidia hardware, and vice versa. there are so many factors, this is why you need to benchmark a number of games to eliminate those biases in the testing.



true, but in real world performance it means you do have a disk system, otherwise how can you fill the ram or cache with data? benchmarks are going to be influenced by the speed of those components. ssd vs platter disk, cas latency on the ram, did they have a raid 0 array? it's why i said the benchmark should really be done on your own system to make it most relevant. it may be impossible to know exactly how many fps you lost with all these variables in play that differ with each system.



real benchmarks of games are definitely needed, that's what everyone is questioning the most, how big is the fps hit? if these app benchmarks are demonstrating heavily optimized cache hits, it may not be the case for a game which needs to load tons of assets in and out of memory. anything that doesn't fit into the texture memory of the graphics card loads from system memory or disk for example.

which brings us to another point: if ecc is so great and doesn't affect the speed in any appreciable way, why isn't the GPU using it? hmm!

also, why would it be considered OK to render frames using non-ecc ram on my GPU, but you guys consider it dangerous to render frames with the non-ecc system-wide memory? explain to me how it's not the exact same thing.




the benchmarks ActionableMango showed come from puget systems here:
https://www.pugetsystems.com/labs/articles/ECC-and-REG-ECC-Memory-Performance-560/

on that page, you will also find:


so that's not my invention, its stated as a fact by at least one manufacturer, and it's referenced in many places on the net. puget systems didn't see a 2% penalty in that specific battery of tests. but it's possible they did the tests wrong because they have suspicious results with registered ecc being faster than non-ecc. and registered ecc has an extra step involved that standard ecc does not have. furthermore, this test was of ddr3 not ddr4 ram, it's not even the right type of memory module. tests from anandtech and toms hardware would be more reliable if any exist.

a 10 fps hit in games was my own ballpark estimate of what you may see with ecc ram. this is what you guys seem to disagree with the most.

ActionableMango combined two separate data points and came up with a formula to link them together:
2% of 500fps = 10fps

his suggestion is that a 10fps hit means we are getting 500fps. but that formula is not correct at all. you cannot just take some numbers and create a linear equation out of it and expect it to make sense.

You make more assumption than we do.

1) RAM speed is just RAM speed, you assume that 2% come from non ECC cannot be compared to from clock speed. Where is the reference? It's just your assumption.

2) You assume the effect on DDR3 and DDR4 is different. Again, RAM speed is just RAM speed, 2% is just 2%, where is your reference that 2% on DDR4 is different than DDR3? It's just your assumption.

3) you even assume the memory manufacture did the wrong test!!! But your initial agreement was because the memory manufacture claim that there are 2% difference!

4) You assume 2% penalty = 10FPS and with a very non-linear scale. Where is the reference? It's again just your own assumption.

2% is just 2%. If there are 2% performance penalty on a 100FPS situation, then the penalty is 2FPS. That's just some simple maths.

And why gaming card's don't use ECC? Because that increase the manufacturing cost. It's so simple.

And why we don't use ECC VRAM for rendering. In fact, quite a few of us prefer to do that, just limited by the software / driver support etc. Some members here get the workstation card simply because want to use ECC VRAM.

Also registered memory has nothing to do with ECC. And registered memory do perform better when multi DIMM installed on a single channel. However, most non ECC RAM are unregistered, and registered RAM often found in ECC RAM, that makes people confused. It's totally possible that registered ECC RAM run faster than non ECC unregistered RAM. You can goggle and study what is registered memory, then you should know the reason.

 

[applesaucePro]

1. i don't understand your question.

2. they are 2 different parts, cannot compare apples to oranges.

3. puget systems did that test, they are a system builder, not a memory manufacturer. you are confusing 2 different companies. crucial.com is the memory manufacturer.
https://www.pugetsystems.com
http://www.crucial.com/

4. yes i did assume this, nobody has found any ecc ddr4 gaming benchmarks to prove it wrong yet.

And why gaming card's don't use ECC?

because it's slow, expensive, and unnecessary!

Some members here get the workstation card simply because want to use ECC VRAM.

workstation cards did not always have ecc, and many still do not have it. take a look here:
https://www.anandtech.com/show/11102/nvidia-announces-quadro-gp100

it says... GPU: GP100, P6000, M6000, K6000 => ECC: Full, Partial, Partial, Full

so my questions:
1. wow, full ecc is not needed on the mighty P6000, top of the line quadro!!? why not?
2. why is the P6000 $5000 and you don't get full ecc ram? you said cost is the only reason not to add it.
3. is it too dangerous now to use the quadro to render with, because it has no full ecc support?
4. what makes quadro safe to render on, and not a geforce card, and not system-memory? all use non-ecc ram...

registered memory has nothing to do with ECC.

registered ecc ram is a type of ecc ram. registered ecc should be slower than unbuffered non-ecc ram.

the test from puget systems did not show this, so it is suspicious. even if true, that test was in regards to ddr3, and did not even have any gaming benchmarks, let alone any fps numbers.

 

[h9826790]

1. i don't understand your question.
2. they are 2 different parts, cannot compare apples to oranges.
3. puget systems isn't a memory manufacturer, its a system builder.
4. yes i did assume this, nobody has found any ecc ddr4 gaming benchmarks to prove it wrong yet.


because it's slow, expensive, and unnecessary!


workstation cards did not always have ecc, and many still do not have it. take a look here:
https://www.anandtech.com/show/11102/nvidia-announces-quadro-gp100

it says... GPU: GP100, P6000, M6000, K6000 => ECC: Full, Partial, Partial, Full

so my questions:
1. wow, full ecc is not needed on the mighty P6000, top of the line quadro!!? why not?
2. why is the P6000 $5000 and you don't get full ecc ram? you said cost is the only reason not to add it.
3. is it too dangerous now to use the quadro to render with, because it has no full ecc support?
4. what makes quadro safe to render on, and not a geforce card, and not system-memory? all use non-ecc ram...



registered ecc ram is a type of ecc ram. registered ecc should be slower than unbuffered non-ecc ram.

the test from puget systems did not show this, so it is suspicious. even if true, that test was in regards to ddr3, and did not even have any gaming benchmarks, let alone any fps numbers.

Both ECC and non ECC RAM can be registered. Please do some study. Registers memory has nothing to do with ECC, and definitely not a subset of ECC RAM.

UDIMM is faster than RDIMM if there is only one DIMM installed on a memory channel, but RDIMM will be faster if there is more than one DIMM installed on a single memory channel. Again, please do your own study. Then you will know the “why”.

Not all workstation card has ECC VRAM, and I never say all of them have. Different card fit different usage, that’s some simple logic. Not one size fit all.

I agree that ECC VRAM is unnecessary for gaming, and no point to use that on gamin card becasue of the unnecessary extra cost. However if there no evidence to show they are not slower, that also means no evidence to prove that they are slower. Then why you insist it’s slower in real world with absolutely zero prove?

So far, all evidence pointed to there is no noticeable performance penalty by using ECC RAM. And you still unable to provide a single solid reproducible situation that ECC RAM can reduce gaming frame rate by 10FPS.

I am more than happy to discuss / debate with you, and allow both of us to know the truth. However, can we do it properly with prove / evidence / solid info? I tried both ECC and non ECC on my own Mac, and there is absolutely no difference in any real world usage. And I do quite a bit of gaming on my cMP. I can tell you that the FPS difference is literally zero. This is not theory, but real world experience by using my own Mac. Not any rumour / theory on the internet.

 

[throAU]

I'm super happy with my new Ryzen 7 2700X desktop with dual RX Vega 64s that i built for less than an iMac Pro.

 

[h9826790]

I'm super happy with my new Ryzen 7 2700X desktop with dual RX Vega 64s that i built for less than an iMac Pro.

Can that run MacOS?

 

[pl1984]

Not the thread where users buy or modify or use "supported" cards. That would obviously be user fault. I meant the thread with the 2013 Mac Pros running on faulty D300 and D500s, causing crashes and freezes. The 2013 Mac Pros that Apple has acknowledged that has problems.
https://forums.macrumors.com/thread...u-driver-issues.1860297/page-63#post-26186654

In a tech forum, there are many complaints, but people complaining about non-ECC memory problems are quite rare.

Thank you for the clarification.

 

[applesaucePro]

Please do some study. Registers memory has nothing to do with ECC, and definitely not a subset of ECC RAM.

registered ecc ram does exist. we know that there is such a thing. it has a buffer. i was merely telling you, i expect any type of ecc to be slower than non-ecc. we don't need to argue about semantics of the word subset that's not productive.

Not all workstation card has ECC VRAM, and I never say all of them have. Different card fit different usage, that’s some simple logic. Not one size fit all.

the P6000 is the top quadro card, nothing is better. if the top card doesn't need full ecc support, nobody needs it. just like nobody needs ecc in their system anywhere. it's a "want". you want it. it makes you feel safer, even though it only protects from single bit errors and nothing else. its an expensive waste of money for most people, but they want it anyway, because they don't know any better. that's one reason why i started this thread to begin with, to help people learn if they need it or not.

I agree that ECC VRAM is unnecessary for gaming, and no point to use that on gamin card becasue of the unnecessary extra cost.

at least we can both agree that ecc is not needed for playing games!

And you still unable to provide a single solid reproducible situation that ECC RAM can reduce gaming frame rate by 10FPS.

it seems you are mainly disagreeing now about my theoretical 10fps drop caused by using ecc. well, i did admit already that there are many variables which affect fps, and it will vary by computer and game. just because you do not see such a drop doesn't mean it cannot happen.

lots of tests would be needed. most games are capped at 60FPS, and top end cards could produce 200fps. so how can you notice if 10fps was dropped off the top of 200 when its already capped at 60??? you can't! and can you always tell by eyeball 50 from 60fps... no. this is why real benchmarks are needed.

things you noticed when playing a game on your computer does not count as a benchmark.

 

[pl1984]

the performance of non-ecc memory in games is well understood because many benchmarks have been done on it. but i don't think you will find many gaming benchmarks of ecc memory, and that's what we are interested in. you can't assume that the trends in benchmarks of non-ecc ram will match that of ecc, they are two separate things.

I think the point was to show memory speed doesn't have much of an affect on FPS. Think of ECC RAM being the DDR4-2133 speed memory (the slowest in the benchmark) and non-ECC RAM being the DDR4-3200 speed memory (the fastest in the benchmark). The 2133 memory is 33% slower than the 3200 memory. Yet the FPS for the two is essentially the same.

In essence you're saying slower memory (i.e. ECC) results in lower FPS over faster memory (i.e. non-ECC). While his example doesn't directly compare ECC versus non-ECC it does show that slower memory does not automatically translate into slower game performance.

It's not perfect but it's the best we've got unless someone wants to perform some benchmarks.

its also important to know that the benchmarks you posted earlier of ecc ram were done on ddr3 modules, not ddr4. nobody is using ddr3 anymore, we really need tests showing ddr4. do we assume dual-channel ddr3 behaves the same way as 6-channel ddr4 on some test? we can't assume that, we would be guessing.

The graph in post 23 shows the memory was DDR4

which brings us to another point: if ecc is so great and doesn't affect the speed in any appreciable way, why isn't the GPU using it? hmm!

Some do. Several workstations grade graphics cards do.

also, why would it be considered OK to render frames using non-ecc ram on my GPU, but you guys consider it dangerous to render frames with the non-ecc system-wide memory? explain to me how it's not the exact same thing.

I doubt anyone considers it dangerous. Just not preferable.

​

the benchmarks ActionableMango showed come from puget systems here:
https://www.pugetsystems.com/labs/articles/ECC-and-REG-ECC-Memory-Performance-560/

on that page, you will also find:

so that's not my invention, its stated as a fact by at least one manufacturer, and it's referenced in many places on the net. puget systems didn't see a 2% penalty in that specific battery of tests. but it's possible they did the tests wrong because they have suspicious results with registered ecc being faster than non-ecc. and registered ecc has an extra step involved that standard ecc does not have. furthermore, this test was of ddr3 not ddr4 ram, it's not even the right type of memory module. tests from anandtech and toms hardware would be more reliable if any exist.

You can't discount the results of these benchmarks merely because you disagree with them. If you've got some benchmarks which disagree with these benchmarks we'd be happy to look them over. Do you?

ActionableMango combined two separate data points and came up with a formula to link them together:
2% of 500fps = 10fps

his suggestion is that a 10fps hit means we are getting 500fps. but that formula is not correct at all. you cannot just take some numbers and create a linear equation out of it and expect it to make sense.

Why not?

 

[h9826790]

registered ecc ram does exist. we know that there is such a thing. it has a buffer. i was merely telling you, i expect any type of ecc to be slower than non-ecc. we don't need to argue about semantics of the word subset that's not productive.


the P6000 is the top quadro card, nothing is better. if the top card doesn't need full ecc support, nobody needs it. just like nobody needs ecc in their system anywhere. it's a "want". you want it. it makes you feel safer, even though it only protects from single bit errors and nothing else. its an expensive waste of money for most people, but they want it anyway, because they don't know any better. that's one reason why i started this thread to begin with, to help people learn if they need it or not.


at least we can both agree that ecc is not needed for playing games!


it seems you are mainly disagreeing now about my theoretical 10fps drop caused by using ecc. well, i did admit already that there are many variables which affect fps, and it will vary by computer and game. just because you do not see such a drop doesn't mean it cannot happen.

lots of tests would be needed. most games are capped at 60FPS, and top end cards could produce 200fps. so how can you notice if 10fps was dropped off the top of 200 when its already capped at 60??? you can't! and can you always tell by eyeball 50 from 60fps... no. this is why real benchmarks are needed.

You still don't do your study!!!!!

Registered NON ECC RAM exist. Registered or not has absolutely nothing to do with ECC. If you don't do your study properly, then this discussion is non productive. And Registered ECC RAM can be faster than Unbuffered non ECC RAM WHEN THERE IS MORE THAN ONE DIMM INSTALLED ON A SINGLE MEMORY CHANNEL. Can you read that? Can you study that? Please don't be lazy but just argue. DO YOUR STUDY!!!

No, I am not just focus on that 10FPS, but there isn't any performance penalty on my own tests.

I am not talking about 200FPS gaming but mainly around 60FPS. I always tune up the graphic settings to just above 60FPS. So far, I can't find one game has 1FPS difference because of (non) ECC RAM. That's with dozens of games. All of them are either CPU limited or GPU limited, but not ECC RAM limited.

And again, you still can't provide a single case that any game has noticeable performance penalty by using ECC RAM. I am not talking about any extreme difference, but just measurable difference that outside normal error margin.

 

[pl1984]

registered ecc ram is a type of ecc ram. registered ecc should be slower than unbuffered non-ecc ram.

This statement has two variables:

1. ECC versus non-ECC
2. Registered versus non-registered

The two or not dependent on one another. ECC is intended to detect and correct errors, registered memory is intended to increase memory capacity over non-registered memory. Both ECC and registered memory due introduce latency over the non-ECC / registered variants.

The question is: Does the latency appreciably affect performance? IME neither does unless the application under test is memory limited (such as benchmarks). I have yet to see a gaming benchmark which shows an appreciable decrease in performance. We're happy to review any you may have.
[doublepost=1530017326][/doublepost]

the P6000 is the top quadro card, nothing is better. if the top card doesn't need full ecc support, nobody needs it. just like nobody needs ecc in their system anywhere. it's a "want". you want it. it makes you feel safer, even though it only protects from single bit errors and nothing else. its an expensive waste of money for most people, but they want it anyway, because they don't know any better. that's one reason why i started this thread to begin with, to help people learn if they need it or not.

Perhaps I'm missing something? The specs for the P6000 (and P5000) list ECC memory:

GPU Memory 24 GB GDDR5X ECC​

http://www4.pny.com/MarketingPromotions/ResellerProgram/email/2016-09-30-P6000-P5000-CAA.html

Frame Buffer Memory 24 GB GDDR5X
ECC Memory Yes

https://hothardware.com/reviews/nvidia-quadro-p6000-and-p5000-workstation-gpu-reviews?page=2

 

[applesaucePro]

Registered NON ECC RAM exist.

no it doesn't, try and buy some, nobody sells it! i don't care if it exists on paper, for all intent and purposes it doesn't exist. let me know if you find some on newegg.com the biggest parts supplier in the world, i don't think that even they have any so you are being ridiculous!

and i already told you i know the difference of registered and non registered, you were not specific about what you were talking about when you tell me just to go research it. why do you not provide me any benchmarks or resources if you wanted me to see something...

you still can't provide a single case that any game has noticeable performance penalty by using ECC RAM. I am not talking about any extreme difference, but just measurable difference

there was a 1% difference on that uniengine benchmark with ecc ddr3, but i suspect even if it was 2% you would say its not significant or it's in the error margin. if you just want to lie to yourself that ecc is as fast as non-ecc, i will not stop you. checking errors takes time vs not checking errors, where no extra steps are needed. lets lie and say checking errors takes no time at all... that is one hell of an optimized algorithm!
[doublepost=1530017814][/doublepost]

Perhaps I'm missing something:

GPU Memory 24 GB GDDR5X ECC
http://www4.pny.com/MarketingPromotions/ResellerProgram/email/2016-09-30-P6000-P5000-CAA.html

it has only partial ecc support, we were discussing that here:
https://forums.macrumors.com/threads/alternatives-to-mac-pro.2124642/page-2#post-26188809

 

[pl1984]

no it doesn't, try and buy some, nobody sells it! i don't care if it exists on paper, for all intent and purposes it doesn't exist. let me know if you find some on newegg.com the biggest parts supplier in the world, i don't think that even they have any so you are being ridiculous!

Registered memory is intended to permit higher memory capacities over non-registered memory. The primary use for high memory capacities is found in workstations and servers where, ECC memory, has value. It makes sense ECC is found in registered memory modules. However the two are not dependent on one another.

it has only partial ecc support, we were discussing that here:
https://forums.macrumors.com/threads/alternatives-to-mac-pro.2124642/page-2#post-26188809

What does partial ECC support mean? Everything I've read indicates the 24GB of RAM is ECC protected.

 

[applesaucePro]

What does partial ECC support mean? Everything I've read indicates the 24GB of RAM is ECC protected.

if you spot me $5000 i will get one and let you know all about it.

 

[pl1984]

if you spot me $5000 i will get one and let you know all about it.

Uh, I'm not that interested

 

[h9826790]

no it doesn't, try and buy some, nobody sells it! i don't care if it exists on paper, for all intent and purposes it doesn't exist. let me know if you find some on newegg.com the biggest parts supplier in the world, i don't think that even they have any so you are being ridiculous!

and i already told you i know the difference of registered and non registered, you were not specific about what you were talking about when you tell me just to go research it. why do you not provide me any benchmarks or resources if you wanted me to see something...


there was a 1% difference on that uniengine benchmark with ecc ddr3, but i suspect even if it was 2% you would say its not significant or it's in the error margin. if you just want to lie to yourself that ecc is as fast as non-ecc, i will not stop you. checking errors takes time vs not checking errors, where no extra steps are needed. lets lie and say checking errors takes no time at all... that is one hell of an optimized algorithm!
[doublepost=1530017814][/doublepost]

it has only partial ecc support, we were discussing that here:
https://forums.macrumors.com/threads/alternatives-to-mac-pro.2124642/page-2#post-26188809

You are so care that 2% on paper, and now your don't care registered non ECC RAM's existence (on paper). Is that ridiculous?

You can't find one because it's rare, but it does exist. Also, no matter you can buy it or not, the point is, registered has ABSOLUTELY NOTHING TO DO WITH ECC. This is the point. You keep shifting the focus. When I tell you the fact, you said this only exist on paper.

We show you the benchmarks evidence, then you said that 2% exist on paper but we can't observe it at all.

What you want to say???

On paper? or in real world.

If on paper, registered has nothing to do with ECC. and ECC DRIMM can be faster than non ECC UDIMM if condition fit.

If in real world, benchmarks shows that 2% doesn't exist.

 

[BlueTide]

One needs to understand their workload and test suitable configurations to determine the best solution that fits their needs. This is a lot simpler with Apple as they offer a limited number of options. Unfortunately some of those options are slowly fading away as the technology has aged and Apple hasn't updated them.

It also means that you don't necessarily get what would benefit you the most. Same with this ECC discussion - nothing wrong with it, but I'd rather have an option not to have it if it would cost (a lot) more or make the system otherwise cost more. For the same price, sure. But in visual media, I ain't gonna need that stuff. Same with the pro-GPU nonsense. Outside macOS, the bigger reason for those is 10bit colors which is just a money grab driver issue.

 

[pl1984]

It also means that you don't necessarily get what would benefit you the most. Same with this ECC discussion - nothing wrong with it, but I'd rather have an option not to have it if it would cost (a lot) more or make the system otherwise cost more. For the same price, sure. But in visual media, I ain't gonna need that stuff. Same with the pro-GPU nonsense. Outside macOS, the bigger reason for those is 10bit colors which is just a money grab driver issue.

I agree. The best way to get what benefits you the most is to consider the PC space. IMO, a lot of the holdouts prioritize macOS first and then everything else.

 

[ISanych]

the poor film student who needs a ton of ram can buy 2 or maybe even 3 times as much non-ecc ram for the same price. right? is he better off getting more student loans to buy ram, or is he better off buying non-ecc?

Actually if you need a ton of ram, ecc ram is way to go. For last home server I've got 256GB of DDR3 ECC (cheaper than adding 32Gb to iMac pro) - and non ECC is not an option if you need TBs of RAM

 

[applesaucePro]

It also means that you don't necessarily get what would benefit you the most. Same with this ECC discussion - nothing wrong with it, but I'd rather have an option not to have it if it would cost (a lot) more or make the system otherwise cost more. For the same price, sure. But in visual media, I ain't gonna need that stuff. Same with the pro-GPU nonsense. Outside macOS, the bigger reason for those is 10bit colors which is just a money grab driver issue.

i agree with everything you said here 100% a designer has different needs than a engineer, and even among design there are differing needs such as with 10bit. i am seeing so much opposition to the idea that we should take a more active role in choosing system specs, rather than have it hand-fed to us with limited configuration options that don't make sense or are just expensive extras we don't need.

Actually if you need a ton of ram, ecc ram is way to go. For last home server I've got 256GB of DDR3 ECC (cheaper than adding 32Gb to iMac pro) - and non ECC is not an option if you need TBs of RAM

ecc is best for a server, definitely!

You are so care that 2% on paper, and now your don't care registered non ECC RAM's existence (on paper). Is that ridiculous?

hahaha! did you find proof that registered non-ecc ram exists yet? show it to us. i guess the worlds biggest part supplier newegg didn't have one. did you try looking on ebay or digging in the ground for fossils of one that ever existed? why am i the only one that needs to prove things around here?

registered has ABSOLUTELY NOTHING TO DO WITH ECC. This is the point. You keep shifting the focus.

YOU keep shifting the focus, we were talking about ecc, and now you start talking about registered! and you admit it has nothing to do with ecc.

that 2% exist on paper but we can't observe it at all

you told me 2% doesn't matter and is not within what you consider an error margin, how can you observe something you aren't willing to observe?

 

[pl1984]

i agree with everything you said here 100% a designer has different needs than a engineer, and even among design there are differing needs such as with 10bit. i am seeing so much opposition to the idea that we should take a more active role in choosing system specs, rather than have it hand-fed to us with limited configuration options that don't make sense or are just expensive extras we don't need.

I find people in this forum are advocating for this ability.

 

[applesaucePro]

i didn't have a chance to address post#33, but here's the post i'm referencing.
https://forums.macrumors.com/threads/alternatives-to-mac-pro.2124642/page-2#post-26189095

1. horses and mules are related, but you can't time a horse race and then expect a mule to run as fast when you have never observed it run. you have to observe the actual thing and scientifically measure the performance, that is why we don't assume non-ecc benchmarks apply to ecc devices, we can't extrapolate data from tangentially related things.

2. the benchmark in post #15 is for ecc ddr3, and the one in post #23 is for non-ecc ddr4, don't confuse these.

3. most gpu cards don't have ecc, and many workstation cards don't fully support it either even when it says it does on the box.

4. there was originally a lot of concern about data loss, so i was pointing out that most gpu cards don't have ecc, you all are running fine without it and the world hasn't ended

5. the people who did this benchmark even mentioned themselves that the results were suspicious and re-ran the tests. they should have had another lab verify the findings. but since the tests are for ddr3 it doesn't apply to ddr4 at all. so i'm not discounting benchmark results out of hand here.

6. you guys took 2 data points i mentioned and made it into a linear equation. computer performance is seldom linear, you can see this in things like amdahls law. if we could make such elementary assumptions about performance why are benchmark tests even done? we can't extrapolate a graph from nothing, you would need other data points to show a trend!

 

[pl1984]

i didn't have a chance to address post#33, but here's the post i'm referencing.
https://forums.macrumors.com/threads/alternatives-to-mac-pro.2124642/page-2#post-26189095

1. horses and mules are related, but you can't time a horse race and then expect a mule to run as fast when you have never observed it run. you have to observe the actual thing and scientifically measure the performance, that is why we don't assume non-ecc benchmarks apply to ecc devices, we can't extrapolate data from tangentially related things.

It's called an analogy. They're not perfect but this one illustrates memory does not appear to be a limiting factor. Until you can demonstrate memory is a factor the analogy is reasonable.

2. the benchmark in post #15 is for ecc ddr3, and the one in post #23 is for non-ecc ddr4, don't confuse these.

I wasn't. I clearly said:

The graph in post 23 shows the memory was DDR4​

Which was a reference to what I had written earlier in my response.

3. most gpu cards don't have ecc, and many workstation cards don't fully support it either even when it says it does on the box.

I don't recall anyone saying most did so I'm not sure what point it is you're trying to counter with this statement.

4. there was originally a lot of concern about data loss, so i was pointing out that most gpu cards don't have ecc, you all are running fine without it and the world hasn't ended

Unless one is using a GPU for compute purposes the lack of ECC on a graphics adapter is unlikely to result in any problems. At worst the displayed image would have an imperfection which, in all likelihood, would not be noticeable. The same cannot be said of the information stored in memory.

5. the people who did this benchmark even mentioned themselves that the results were suspicious and re-ran the tests. they should have had another lab verify the findings. but since the tests are for ddr3 it doesn't apply to ddr4 at all. so i'm not discounting benchmark results out of hand here.

Yes, you are. You have provided little more than your opinion which appears to be contradicted by available data and others experiences. We're happy to review any data and / or benchmarks which support your position and we've been waiting on it.

6. you guys took 2 data points i mentioned and made it into a linear equation. computer performance is seldom linear, you can see this in things like amdahls law. if we could make such elementary assumptions about performance why are benchmark tests even done? we can't extrapolate a graph from nothing, you would need other data points to show a trend!

Perhaps I missed something but wasn't it you who is saying it's a 2% decrease? If so let me ask you this: A 2% decrease from what? What is the reference point for which the 2% is being applied?

 

[ActionableMango]

why am i the only one that needs to prove things around here?

First of all, when someone make an assertion like "ecc is going to cost 10FPS at least", it is on the person making the assertion to prove it.

Secondly, I did not set out to prove you wrong. That's why I felt bad earlier that it seemed to be turning into an argument. It really isn't, not for me. It is an interesting topic, personally, since I have myself switched from Mac Pro with ECC to PC without ECC.

I set out to look for the answer, whatever it was. I could not find a "perfect" benchmark, which I would really have liked to find for my own satisfaction. I did find some approximations that are good enough for me. No, they do not absolutely disprove your assertion, but they do cast a large amount of doubt on it.

I'll point out that you are holding my references to a far higher standard than you are holding yourself. That 2% Crucial statement, which has been taken down, was written in 2001 or even earlier, which would have been DDR1 era or older.

So if your point is that multiple real-world bench marks for DDR4 and for DDR3-ECC are too different to be relevant, I'd counter that they are more relevant than a deleted 17-year old statement about DDR1 or older RAM. Also you took that 2% and did I don't even know what, in order to come up with the ballpark 10FPS figure.

If you don't want to believe my references, that's totally fine; I am personally satisfied with the answers I found.

However, I don't think you should attack my references with such high standards until you can find something better, because IMHO what I found is significantly more up to date and relevant than what you've provided.

If you do find something better, I'm genuinely interested in the results, whatever they are. My goal is only to know what the answer is. Maybe it is a >10FPS drop. But I don't think so.

 

[h9826790]

YOU keep shifting the focus, we were talking about ecc, and now you start talking about registered! and you admit it has nothing to do with ecc.

Because you said "it's possible they did the tests wrong because they have suspicious results with registered ecc being faster than non-ecc" in post #25...

You have no idea that you are the person to start with this topic??? I told you twice the fact registered ECC RAM can run faster than non registered non ECC RAM when more than one DIMM installed in a single memory channel. You just ignore it. When I re-emphasis this fact, you suddenly say I am the person who shift the focus???

From Wiki:

Although most registered memory modules also feature error-correcting code memory (ECC), it is also possible for registered memory modules to not be error-correcting or vice versa. Unregistered ECC memory is supported and used in workstation or entry-level server motherboards that do not support very large amounts of memory.[1]

AND

Normally, there is a performance penalty for using registered memory. Each read or write is buffered for one cycle between the memory bus and the DRAM, so the registered RAM can be thought of as running one clock cycle behind the equivalent unregistered DRAM. With SDRAM, this only applies to the first cycle of a burst.

However, this performance penalty is not universal. There are many other factors involved in memory access speed. For example, the Intel Westmere 5600 series of processors access memory using interleaving, wherein memory access is distributed across three channels. If two memory DIMMs are used per channel, this "results in a reduction of maximum memory bandwidth for 2DPC (DIMMs per channel) configurations with UDIMM by some 5% in comparison to RDIMM".[attribution needed][2](p. 14). This occurs because "when you go to two DIMMs per memory channel, due to the high electrical loading on the address and control lines, the memory controller uses a '2T' or '2N' timing for UDIMMs. Consequently, every command that normally takes a single clock cycle is stretched to two clock cycles to allow for settling time. Therefore, for two or more DIMMs per channel, RDIMMs will have lower latency and better bandwidth than UDIMMs."[3]

I told you why registered ECC can be faster than non ECC (this is exactly because non ECC RAM used for testing is non registered). And there is no point to doubt that result. And your refuse to accept it, because you refuse to listen, refuse to study. And just insist you were right without any further solid info or prove.

First of all, when someone make an assertion like "ecc is going to cost 10FPS at least", it is on the person making the assertion to prove it.

Secondly, I did not set out to prove you wrong. That's why I felt bad earlier that it seemed to be turning into an argument. It really isn't, not for me. It is an interesting topic, personally, since I have myself switched from Mac Pro with ECC to PC without ECC.

I set out to look for the answer, whatever it was. I could not find a "perfect" benchmark, which I would really have liked to find for my own satisfaction. I did find some approximations that are good enough for me. No, they do not absolutely disprove your assertion, but they do cast a large amount of doubt on it.

I'll point out that you are holding my references to a far higher standard than you are holding yourself. That 2% Crucial statement, which has been taken down, was written in 2001 or even earlier, which would have been DDR1 era or older.

So if your point is that multiple real-world bench marks for DDR4 and for DDR3-ECC are too different to be relevant, I'd counter that they are more relevant than a deleted 17-year old statement about DDR1 or older RAM. Also you took that 2% and did I don't even what, in order to come up with the ballpark 10FPS figure.

If you don't want to believe my references, that's totally fine; I am personally satisfied with the answers I found.

However, I don't think you should attack my references with such high standards until you can find something better, because IMHO what I found is significantly more up to date and relevant than what you've provided.

If you do find something better, I'm genuinely interested in the results, whatever they are. My goal is only to know what the answer is. Maybe it is a >10FPS drop. But I don't think so.

I totally agree if that almost 50% faster memory (due to clock speed) can't provide any benefit on gaming, then a merely 2% faster (on paper) non ECC RAM is impossible to show any effect on gaming. Especially applesaucePro keep emphasising that he only interested in DDR4 and gaming.

 

[applesaucePro]

It's called an analogy. They're not perfect but this one illustrates memory does not appear to be a limiting factor.

sorry, i'm not sure what specific benchmark you were making an analogy of, there are a few now being thrown around.

i do accept that memory speed is not a significant factor of performance. i do accept that error checking memory is not a significant factor either. i have been saying this all along. but the whole debate is about that small difference.

the difference is considered insignificant by most people, that's why there's so few benchmarks on it. but it doesn't mean this difference does not exist. the 2% figure came from manufacturers themselves, so it holds weight.

That 2% Crucial statement, which has been taken down, was written in 2001 or even earlier, which would have been DDR1 era or older.

haha i was waiting for someone to say that! you get the gold star. exactly right. an old statement of performance may not be true any longer. and the whole knowlegebase is gone, nobody else mentioned it yet, i don't think they even bothered to look. maybe that knowlegebase just got moved though.

what i find funny, hilarious actually, is that everyone here thinks 2% overall system degradation is insignificant. but a 10fps drop in a game, well that's just downright intolerable! proves my point that people are very upset about losing frames and will want to avoid ecc if it means even one fps is lost.

i'm a bit busy now but i will come back to discuss the other points in a bit.