Graphics card-like bandwidth with graphics card-like compute performance.

A workstation with overclocked memory? :rolleyes:
This must be one of the most pointless products ever. Outside of curiosity in benchmarking, this product makes no sense.
A workstation needs reliability, while running a lot of memory at a very high overclock is going to cause not only stability issues, but lots of data corruption.

This is the sort of idea people would get from various YouTube tech channels, several of which annoyingly suggest people just overclock to get "free" performance. :facepalm:

By all means, overclock you hardware if you enjoy it. But don't overclock hardware intended for productive work.

efikkanA workstation with overclocked memory? :rolleyes:
This must be one of the most pointless products ever. Outside of curiosity in benchmarking, this product makes no sense.
A workstation needs reliability, while running a lot of memory at a very high overclock is going to cause not only stability issues, but lots of data corruption.

This is the sort of idea people would get from various YouTube tech channels, several of which annoyingly suggest people just overclock to get "free" performance. :facepalm:

By all means, overclock you hardware if you enjoy it. But don't overclock hardware intended for productive work.

- If ECC, it can correct its errors.
- If it works faster, it can finish work before it fails
- stability issues are tested by stress-testing first, to be sure it will work
- sometimes you just need a threadripper to be able to use 8 graphics cards on same motherboard and care only about performance of a simulation

tugrul_SIMD- If ECC, it can correct its errors.

While this says nothing about ECC, ECC wouldn't make a difference here.
DDR5 have built in ECC for the transfer itself, and ECC memory offers parity on the memory module side, but the problem I'm talking about isn't about the memory itself, as the DRAM itself probably can handle both the frequency, voltage and timings (assuming top quality DRAM).
The stability issue with OC memory is caused by the memory controller becoming unreliable over time, as it degrades, and the more load you put on it, the faster it degrades. ECC would do nothing to mitigate this issue.

tugrul_SIMD- If it works faster, it can finish work before it fails

Makes no sense.
So, constantly reboot the computer, and then re-run the tests to verify the results?
Data is worthless if it's not reliable. :P
And how much time would you waste on this? For professionals, troubleshooting stability issues is usually not worth the hassle. If someone is seriously running a lot of heavy simulations, they probably need a system that just works.

tugrul_SIMD- stability issues are tested by stress-testing first, to be sure it will work

So, stress test every time then?
The system can work fine today, but start to get memory errors a month from now. By the time you see a crash, there would be thousands of memory errors that didn't cause a crash, and some of them may have affected results or worse, corrupted a file.

Testing for faulty memory itself is fairly easy, testing for timing issues with a memory controller is a bit more difficult. A system can pass a round of Memtest and Prime95, and yet when running a timing sensitive workload get all kinds of random errors, e.g. code compilation, which isn't that stressful, yet can exhibit all kinds of errors. Remove the OC, and the problems goes away.

Hardware, incl. memory controllers are designed with safety margins, since they do degrade over time. When you buy e.g. a Threadripper, a workstation grade motherboard and ECC memory, they are usually certified for 24-7 load throughout the warranty period. But if you choose to OC instead, you use this safety margin to achieve the OC, but then it's only a matter of time before you get issues.

tugrul_SIMD- sometimes you just need a threadripper to be able to use 8 graphics cards on same motherboard and care only about performance of a simulation

You know, there are 8-channel memory on some motherboards, you seriously need more than that but don't care about reliability? ;)

efikkanA workstation with overclocked memory? :rolleyes:
This must be one of the most pointless products ever. Outside of curiosity in benchmarking, this product makes no sense.
A workstation needs reliability, while running a lot of memory at a very high overclock is going to cause not only stability issues, but lots of data corruption.

This is the sort of idea people would get from various YouTube tech channels, several of which annoyingly suggest people just overclock to get "free" performance. :facepalm:

By all means, overclock you hardware if you enjoy it. But don't overclock hardware intended for productive work.

You say that like this kit wasn't tested at all at the speeds that they're claiming it will run stable at. You don't need to run at JEDEC speeds to have a stable memory kit. If anything, the kit has been validated at these speeds but the CPU and your setup may not. Even still, if it has ECC, there should be a lot of headroom.

Any normal person looking at this kit is likely going to stress test it before throwing a multi-hour workload at it, but let's be real. A single bit flipping will render any kit useless for any use, not just productivity workloads. I suspect the bigger question mark would be the IMC on whatever CPU you're using. Either way, systems can be stress tested prior to such use. However if the kit has been validated at these speed and has ECC, then I'd feel fairly confident using such a kit.

By your logic, the DDR3-2133Mhz kit in my SBe machine that I've had for 12 years should never have been used for professional workloads, yet I did and it worked out just fine. I think you're just being hyperbolic to be completely honest. I get the point, but it's a mole hill, not a mountain.

AquinusYou say that like this kit wasn't tested at all at the speeds that they're claiming it will run stable at. You don't need to run at JEDEC speeds to have a stable memory kit. If anything, the kit has been validated at these speeds but the CPU and your setup may not. Even still, if it has ECC, there should be a lot of headroom.

Any normal person looking at this kit is likely going to stress test it before throwing a multi-hour workload at it, but let's be real. A single bit flipping will render any kit useless for any use, not just productivity workloads. I suspect the bigger question mark would be the IMC on whatever CPU you're using. Either way, systems can be stress tested prior to such use. However if the kit has been validated at these speed and has ECC, then I'd feel fairly confident using such a kit.

By your logic, the DDR3-2133Mhz kit in my SBe machine that I've had for 12 years should never have been used for professional workloads, yet I did and it worked out just fine. I think you're just being hyperbolic to be completely honest. I get the point, but it's a mole hill, not a mountain.

Yup, JEDEC is overly cautious, slow to update and leaves too much on the table.

efikkanWhile this says nothing about ECC, ECC wouldn't make a difference here.
DDR5 have built in ECC for the transfer itself, and ECC memory offers parity on the memory module side, but the problem I'm talking about isn't about the memory itself, as the DRAM itself probably can handle both the frequency, voltage and timings (assuming top quality DRAM).
The stability issue with OC memory is caused by the memory controller becoming unreliable over time, as it degrades, and the more load you put on it, the faster it degrades. ECC would do nothing to mitigate this issue.


Makes no sense.
So, constantly reboot the computer, and then re-run the tests to verify the results?
Data is worthless if it's not reliable. :p
And how much time would you waste on this? For professionals, troubleshooting stability issues is usually not worth the hassle. If someone is seriously running a lot of heavy simulations, they probably need a system that just works.


So, stress test every time then?
The system can work fine today, but start to get memory errors a month from now. By the time you see a crash, there would be thousands of memory errors that didn't cause a crash, and some of them may have affected results or worse, corrupted a file.

Testing for faulty memory itself is fairly easy, testing for timing issues with a memory controller is a bit more difficult. A system can pass a round of Memtest and Prime95, and yet when running a timing sensitive workload get all kinds of random errors, e.g. code compilation, which isn't that stressful, yet can exhibit all kinds of errors. Remove the OC, and the problems goes away.

Hardware, incl. memory controllers are designed with safety margins, since they do degrade over time. When you buy e.g. a Threadripper, a workstation grade motherboard and ECC memory, they are usually certified for 24-7 load throughout the warranty period. But if you choose to OC instead, you use this safety margin to achieve the OC, but then it's only a matter of time before you get issues.


You know, there are 8-channel memory on some motherboards, you seriously need more than that but don't care about reliability? ;)

IMC degradation ? Really ? at some stage yeah sure I believe but there is definitely an option to safely oc a workstation and run it for years stable, I'm sitting here on ASRock X79 Extreme11 with 4930K and G.Skill 2400 9-11-11-31 @ 2450 9-11-11-28-160 T1 quad channel over a decade now OCed stable 4750@1.5v cpu 1.3VTT 1.3 VCSSA bclk strap 125 some will say volts are too high but what I had learned over the years on this platform as long as temps are ok and runs stable it won't really degrade for another decade or even two. Asrock X79 Extreme11 is more of a workstation motherboard imo but that doesn't mean you can't OC safely and stable with 9 SSD 3 HDD and almost 40TB of data. I couldn't afford any data corruption, workstations definitely can be safely OCed and be stable.

Workstation CPUs are highest binned. They can be OCed easily.