• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Epyc has problems when you max out PCIe lanes

andy_mk3 said:
System is this one: https://www.gigabyte.com/Rack-Server/R272-Z32-rev-100#ov

Using CNV3024 NVMe risers.
Click to expand...

Thats the server the l1 tech guy loves. Not sure id deploy gigabyte into enterprise but that does look a nice chassis. My Rome servers are literally pcie gen 3 (hp dl380 g10) naples chassis with an updated bios. Dual socket servers but only one populated.

I get their workloads but i just cant fathom how a single 24 drive array would be the way forward, does editing 8k red footage really require that kind of I/O?
 
Vince said:
Thats interesting, I wonder if the same number of drives in a dual socket would alleviate the issue, im really going to have to watch the video arent I? Commenting up here when i havent even looked. Noob.

Do we know what controller they were using?
Click to expand...

I haven’t watched the video myself yet.
 
Vince said:
Thats the server the l1 tech guy loves. Not sure id deploy gigabyte into enterprise but that does look a nice chassis. My Rome servers are literally pcie gen 3 (hp dl380 g10) naples chassis with an updated bios. Dual socket servers but only one populated.

I get their workloads but i just cant fathom how a single 24 drive array would be the way forward, does editing 8k red footage really require that kind of I/O?
Click to expand...

I guess it does when you have 5 editors plus others that may need to access it. I believe they are all editing off a single PC now too.
 
jigger said:
I haven’t watched the video myself yet.
Click to expand...

I tend to avoid LTT, not really sure why but something about it annoys me. I don't think it's particularly Linus but it just grinds me slightly. Mind you I stopped watching that south African channel after one day he was talking about tech companies financials and he skipped over some massive tech companies like Fortinet with a comment like "dunno what that is", err dude, how can I now take a tech review company seriously that has 0 exposure to Fortinet? I believe they have been in the to 50 ish fastest growing tech companies for the last 5 years.

andy_mk3 said:
I guess it does when you have 5 editors plus others that may need to access it. I believe they are all editing off a single PC now too.
Click to expand...

I guess so, im not really in a position to argue either way tbh. I just find it hard to believe that you need that sort of bandwidth but then I don't know and it would be silly of me to make assumptions. I wonder then how "big" studios deal with this. FC Connected workstations at a guess.
 
Vince said:
I guess so, im not really in a position to argue either way tbh. I just find it hard to believe that you need that sort of bandwidth but then I don't know and it would be silly of me to make assumptions. I wonder then how "big" studios deal with this. FC Connected workstations at a guess.
Click to expand...

I guess the big studios have much bigger deployments with arguably better setups. LTT are a tech channel so I guess some of their interests are cutting edge tech which is interesting to viewers. I quite enjoy watching LTT, he's a knowledgeable guy, I can see why some would find him annoying though.
 
Vince said:
I tend to avoid LTT, not really sure why but something about it annoys me. I don't think it's particularly Linus but it just grinds me slightly. Mind you I stopped watching that south African channel after one day he was talking about tech companies financials and he skipped over some massive tech companies like Fortinet with a comment like "dunno what that is", err dude, how can I now take a tech review company seriously that has 0 exposure to Fortinet? I believe they have been in the to 50 ish fastest growing tech companies for the last 5 years.



I guess so, im not really in a position to argue either way tbh. I just find it hard to believe that you need that sort of bandwidth but then I don't know and it would be silly of me to make assumptions. I wonder then how "big" studios deal with this. FC Connected workstations at a guess.
Click to expand...

Yeah I don’t watch much of this type of YouTube personality myself.

This is a strange use case and configuration, but the cost of EPYC opens up the enterprise market somewhat.
 
Someone was flogging a ton of those drives on Ebay, some data centre was upgrading their storage and the Intel drives being replaced were still under good warranty, its where Linus got them, so i had a look thinking i would get some good Intel mass storage drives cheap, £250 a pop at the time and it turns out that is quite cheap for what they are (4TB Intel P4500) but they are not the cheap i was looking for :(
 
Quartz said:
Linus Tech Tips has an interesting video about the problems they had when they maxxed out the PCIe lanes on their Epyc server with umpteen NVME drives

TLDR There are major performance issues; it's all too fast and the bandwidth is overloaded. I'm wondering if Intel solutions have the same problems?
Click to expand...

Why you try to spin it as "EPYC problem"?

a) If you see the problem was the RAM, having to hold 3 times the data it copies while writing the data on the drives. Due to the ZFS redudancy feature. That alone slows down the whole process and especially the CPU and RAM.

b) Interupt base communication, was the initial issue, because of the sheer amount of drives. Again that was resolved by using polling, however that needs higher CPU power to deal with 24 way raid


Quartz said:
I wonder if reducing the number of cores would have worked.
Click to expand...

And from the above post clearly you didn't understood the whole video and the only thing that you got out of it was "AMD Epyc has problems when you max out PCIe lane".

Is actually sad seeing such clickbait threads in this forum like it is WCCFTECH even to this day where the forum standards are pretty low of what was to be up to 7 years ago.
 
Panos said:
Why you try to spin it as "EPYC problem"?
Click to expand...

Because that's what it appears to be.

Panos said:
And from the above post clearly you didn't understood the whole video
Click to expand...

Are you sure? One of the highlighted problems was interrupts and processes having been switched away and the solution was to switch interrupts off and move to polling. Managing umpteen processes via polling is difficult and, as LTT found, CPU-intensive. So by reducing the number of cores you obviate that. Each process / core has more to do, of course, but it's just shifting stuff between the NIC, RAM, and the SSDs. You could do it as three processes: one moving data between the NIC and RAM, one process moving data between RAM and SSD, and one master process. That's a lot simpler with much less overhead.
 
Quartz said:
Because that's what it appears to be.



Are you sure? One of the highlighted problems was interrupts and processes having been switched away and the solution was to switch interrupts off and move to polling. Managing umpteen processes via polling is difficult and, as LTT found, CPU-intensive. So by reducing the number of cores you obviate that. Each process / core has more to do, of course, but it's just shifting stuff between the NIC, RAM, and the SSDs. You could do it as three processes: one moving data between the NIC and RAM, one process moving data between RAM and SSD, and one master process. That's a lot simpler with much less overhead.
Click to expand...

You missed the point that they are using ZFS with raid redudancy. It must hold 3 copies of the data in the memory while it writes them.
 
LTT explained their workload in a previous video a while back. They basically wanted nvme drives on the working server because the high iops allowed video editors high speed scrubbing and better timeline performance as they skipped through video source files.

If you think that LTT are trying to push enterprise server advice then I think your viewed it wrong. I've always considered them to be a bit of let's see what happens and have a laugh.
 
Donnie Fisher said:
LTT explained their workload in a previous video a while back. They basically wanted nvme drives on the working server because the high iops allowed video editors high speed scrubbing and better timeline performance as they skipped through video source files.

If you think that LTT are trying to push enterprise server advice then I think your viewed it wrong. I've always considered them to be a bit of let's see what happens and have a laugh.
Click to expand...

This for sure. They like doing stuff a slightly different way that is both awesome and functional with the latest tech. It's a great showcase.
 
You must log in or register to reply here.
Back
Top Bottom