If you were to build this today what would you change? I built me a 14900k with 128gb ddr 5 ecc 5400 mhz on a ASUS w680 ace impi but it’s leaving me lacking for the ability to add u.2 ssds. I’m finding u.2 ssds are about half the price of m.2, can endure x10 more, and have PLP. My other option would be to shell big on a bifurcating pcie card. Thoughts? Running about 20 mechanical hdds, a couple of 8tb Samsung qvos, and have a handful of u.2 drives id like to use and move away from sata and m.2 drives.  

2 hours ago, engin33rh3r3 said:

If you were to build this today what would you change? I built me a 14900k with 128gb ddr 5 ecc 5400 mhz on a ASUS w680 ace impi but it’s leaving me lacking for the ability to add u.2 ssds. I’m finding u.2 ssds are about half the price of m.2, can endure x10 more, and have PLP. My other option would be to shell big on a bifurcating pcie card. Thoughts? Running about 20 mechanical hdds, a couple of 8tb Samsung qvos, and have a handful of u.2 drives id like to use and move away from sata and m.2 drives.  

 

The only thing I would have changed is I'd have gone with 4x3.84TB or 4x7.68TB U.2/U.3 drives. And the reason being, I need more flash capacity. I have Mover set to move stuff quite quickly at the moment because well:

 

 

I'm like this constantly. There is terabytes more data I want held in the cache for performance reasons and I'm often at the limit. I think I was at 97% the other day. These 2TB SSD's are great but I need more capacity.

 

I still have a x16 slot unpopulated and two of these connectors on the board each of which can take 2 x U.2 or M.2 drives (with adapters). So I could potentially add 8 x U.2 drives total though I'm more likely to just buy 4. So that's probably what I'll change up next.

 

The 980 Pro's are fast. I've managed to hit 18GB/s with kernel updates and other changes so they're not at all slouches, love the performance and the temperature characteristics but the capacity, 2TB per drive just wasn't enough.

 

Endurance wise I'm not concerned as much, each drive is rated to 1.2PB so with 4 in RAID0 like I have now thats 4.8PB of write endurance. I think I've used 2% on each drive since I isntalled them just over a year ago. But capacity, I need more and I think U.2 enterprise drives are the way to go, the cost per TB is so much lower as you pointed out.

 

Apart from that I don't think I'd change anything to be honest. I love the CPU and Motherboard, the RAM quantity was perfect for what I do, I'm happy with the 2 x 2TB SN850's which I'm using for VM's and Dockers in RAID1.

 

Based on what I just said here I'd probably advise you to also go with U.2's if you need the capacity, consistent performance acros the whole drive or the high endurance.

Got another update to the build today. I added 4 x 7.68TB Kioxia CD6 PCIe 4.0 NVMe SSD's which I've put in a ZFS RAID1 so I have some redundancy. I'll be using these much in the same manner I had been using the 4 x 980 Pro 2TB SSD's, as cache for the array and a hot pool of fast storage for other things I'm doing.

 

I don't have a proper mount for these inside my chassis and I know they need fairly high airflow to stay cool so I got a 3D printed mount for them which lets air flow above and below each drive. I've positioned the stack of drives in front of one of my 120mm fans inside my chassis.

 

 

And here is the business end of the SSD's with their U.3 connector. Little tip about this, U.3 drives will fit and work in U.2 enclosures and with U.2 cables/adapters but not the other way around, U.2 won't work in a U.3 enclosure.

 

 

And finally one of the two costly cables I had to buy to use these drives with my motherboard. Below is a SlimSAS x8 to dual U.2/U.3 adapter. My motherboard features two SlimSAS x8 connectors so I can run all four SSD's at their full PCIe 4.0 x4 rated speed.

 

 

Performance is pretty good, I used FIO running natively on unRAID to test read and write performance. Here is a write result which I think is the more meaningful metric:

 

 

In read testing, I'm seeing 21.5GB/s though I'm not sure how accurate this is due to ZFS memory caching even with -direct flags the numbers don't correspond to what unRAID shows in the dashboard like the above write test does where FIO and unRAID are 1:1 in sync.

 

 

So that's the update for today. I don't have anything else planned with the server currently, but you never know what'll pop up. I may remove the 4 x 980 Pro and place that entire card (they're in a Hyper M.2 card from Asus) into my next desktop system.

How you finding the temps on the a4000? I just got myself one for €400. Had to put the stock cooler back onto the card (it had an aftermarket 4070 cooler). I'm idling at 36. It gets up to 60 real quick (and holds there, 5min test) when transcoding and I havent had an extended transcode yet to see where the load temps stop. Or put a heavier load on it. I know it runs hot and that 60 is absolutely fine, i just want to have something to compare it to. 

1 minute ago, tazire said:

How you finding the temps on the a4000? I just got myself one for €400. Had to put the stock cooler back onto the card (it had an aftermarket 4070 cooler). I'm idling at 36. It gets up to 60 real quick (and holds there, 5min test) when transcoding and I havent had an extended transcode yet to see where the load temps stop. Or put a heavier load on it. I know it runs hot and that 60 is absolutely fine, i just want to have something to compare it to. 

I've found it idles around 45c in my server and a 41% fan duty cycle. When transcoding it gets up to 55-60c. When I'm doing any inference on it, if it's something that takes a few minutes I've seen it hit 90c and begin to throttle due to temperatures.

 

So far I really like the card, it's quite fast especially for LLM's. I've not had it crash or stop working due to temperatures at all it seems to throttle itself quite well when it needs to. For short loads say under a minute sometimes the power limit kicks in before the temperature limit.

That's good to know. I'm more checking for my own peace of mind that I remounted the stock cooler correctly. I don't do anything with it to put it under a serious load just yet.

SO........2.5 yrs later, has this system done everything you've ever wanted it to do?  I ask because I just built my 1st Unraid box this week, from parts I had lying around.  It works great.  I've built hundreds of PCs & servers in my day.

Also, how do you like the performance on the Kioxia's in ZFS RAID ?

I imagine the whole box is highly performant, right?

 

On 12/6/2024 at 6:52 PM, CHoffman said:

SO........2.5 yrs later, has this system done everything you've ever wanted it to do?  I ask because I just built my 1st Unraid box this week, from parts I had lying around.  It works great.  I've built hundreds of PCs & servers in my day.

Also, how do you like the performance on the Kioxia's in ZFS RAID ?

I imagine the whole box is highly performant, right?

 

It has been a great system for sure. I'm very happy with it and with some of the additions I made like the RTX A4000 and the 7.68TB SSD's it has been able to fulfill more roles that I have for the server. Most of the CPU usage on it is actually from my own software that I write so there has been some situations where more cores would have been beneficial to me but that can be remedied, I'll probably pick up an EPYC 7773X (64 Core with 3D V-Cache) when they're cheaper and swap that in as another upgrade.

 

The SSD's perform well. I'm hitting up against the FUSE overhead on unRAID quite often so using these SSD's while bypassing FUSE to get maximum performance is basically required or I'm only getting 1/10th what they can do. Similarly I've found ZFS to be quite hard on the processor so I think that would also benefit from the 7773X if and when I do that upgrade.

 

Looking back at the system in hindsight I think the only things I would have changed is I wouldn't have purchased those M.2 PCIe cards from Amazon the basic looking ones. They struggled with higher powered M.2's. I would have just bought 2 x Hyper M.2 from Asus right from the start. And also the power supply I bought while quite efficient (80+ platinum rated) isn't as efficient as some other units that cost a bit more. I would have definitely gone for a titanium power supply from the get-go to get that extra few percent of efficiency.

 

Also just to touch on stability, the system has been rock solid. I encountered a few bugs with Docker that broke unRAID a few times, but those are not hardware issues they're just software bugs which I was able to work around. The motherboard from Supermicro has been very dependable and everything has "just worked" including those U.3 SSD's which make use of the SlimSAS x8 connectors on the board. Whenever you use cables for high speed niche connectivity like that it's a question mark about whether it really works properly and I'm happy to say it does, I've not had any issues with it or ZFS since adding those drives.

 

Looking towards the future apart from the potential 7773X upgrade I'm looking forward to future linux kernels which apparently will include directed I/O for FUSE shares so that reads and writes can head directly to the disks which should significantly reduce overhead, that'll likely be a huge efficiency win for my usage.

Have a new update today a prelude to another update coming in a couple weeks time. Today I've installed 1TB of DDR4 3200MHz CL22 LRDIMM ECC memory (Model: M386AAG40AM3-CWE). That's 8 x 128GB modules from Samsung.

 

The upgrade went without a hitch even though I had read online about the H12SSL series of boards having some trouble posting when LRDIMM's are installed I didn't have any such problems, the system posted after a minute and all was good. This may be due to me using EPYC Milan and not EPYC Rome combined with much newer modules, I think first-generation DDR4 LRDIMM's that were around 2400MHz are the problematic ones when it comes to EPYC.

 

I've included some photos of the memory and thought I'd start with a humorous one where unRAID is displaying that I have 1024 TiB of memory installed. No doubt a niche visual bug.

 

 

The eight modules:

 

 

The information from a single module:

 

 

And I thought showing how they look in the Supermicro IPMI may also be interesting:

 

 

Unlike the previous 32GB RDIMM's I was using the IPMI isn't able to read the speed of these 128GB modules, but I did check them from within Linux and they are all running at 3200 MHz. Also the IPMI is able to read the temperature sensors from all the modules and boy do they get hot. In-fact I had to install an extra fan in my system just in front of the modules to keep them from overheating.

 

So that's the update for today, I have another one in a few weeks when my new CPU arrives which is an EPYC 7773X 64-core Milan-X processor with 3D-VCache. I decided to pretty much max out this build and get many more years out of it instead of switching to a newer socket, the main reason for that is the lack of PCIe slots on the newer systems combined with the performance of the Milan-X3D chips actually being around 6% higher than the same core-count Genoa chips that lack X3D so it was a good upgrade for my use case considering I didn't need to change motherboards, cooler or lose 2 PCIe slots.

So with this memory upgrade from my last post I have actually had a lot of problems with the temperatures. With my default cooling setup they immediately (within 5 minutes of booting) reached 98-100c when the system was idle and I had to shut the system down and figure out a solution. By contrast the previous memory only hit 62c or so, way below the 85c warning and 90c thermal shutdown alerts in the Supermicro IPMI.

 

So at first I tried of course increasing fan speeds, that didn't do anything. I tried inserting an extra 120mm fan pointed just slightly below the CPU heatsink so it hit the memory and also left the lid of the chassis slightly open to let more air in. This didn't resolve the overheating issue and the memory still hit 82-83c under load and was still climbing in temperature before I shut the system down to try something else.

 

Here is a photo showing the setup (which didn't work) with the 120mm fan added in front of the CPU heatsink assembly, pardon the dust.

 

 

As you can see I had a stack of U.3 drives positioned in-front of the CPU heatsink in the top right corner and I thought that was perhaps part of the issue as they do get quite warm, consuming up-to 19 Watts each and I have 4 of them for 76 Watts total. However, the CPU was still very cool only 35c idle and 45c under a high load with the fan RPM's at 900.

 

So I purchased a PCIe to 4 x U.2/U.3 card and moved those U.3 drives out of there and in doing so removed that rats nest of cables. This surprisingly did not change the temperatures of the RAM or the CPU at all. I tried positioning two 92mm fans connected together side-by-side aimed directly at the RAM below the CPU heatsinks intake fan, this did have an effect but the memory at idle was still 72c and would likely go over 80c in the summer when the room is hotter or under a high load.

 

Back to the drawing board. I thought about the airflow situation and what exactly is happening, well that big CPU heatsink is sucking in all the air before it has a chance to reach the RAM modules underneath. Ordinarily not a problem with low-to-medium sized UDIMM & RDIMM's as they consume so little energy but these 128GB LRDIMM's are a different beast entirely, I would estimate they are consuming somewhere in the range of 120 Watts under load across the full 8 sticks vs about 40 Watts for the 32GB RDIMM's I had previously.

 

So the solution to the problem, firstly I moved around a few of my PCIe cards, moved the U.3 drives to a PCIe card as mentioned previously and then I propped up two 120mm Noctua fans off-set from each other, one in front and one behind the current CPU heatsink and this has actually reduced the RAM temperatures to between 58c-62c idle and 64-66c under a stressed load. Very far away from the 85c warning / 90c shutoff temperatures I was seeing before.

 

Here is how that looks in the system:

 

 

I was afraid that removing the fans from the heatsink would harm the CPU temperatures due to the loss of static pressure but to my surprise the temperature has remained about the same 40c or so.

 

So some lessons here, firstly I knew before I got this RAM that these are intended for very high airflow server chassis. But since I work in the same room as this server I needed it to be quiet, in-fact the build with these 7 fans seen in the photo (3 black Noctuas in the middle, 5 brown ones in the main chamber) have had the needed result, every component is now running cool and the build is still whisper quiet. It's a little janky.. but it works and honestly if it didn't the RAM would have to go.

 

If I were to do another build in the future I would probably just go for the largest RDIMM's perhaps 8 x 64GB for 512GB total to not have such a complicated cooling setup necessary for LRDIMM power consumption. But the RAM is really good, the performance (latency and throughput) is identical to the RDIMM's I had before and apart from the tinkering I had to do with the cooling, works great.

 

Perhaps someone else will stumble across this post and find it useful, my clear warning here is, you need high airflow or a large case where you can pepper multiple fans to cool these high capacity LRDIMM's and perhaps even if I used a less-wide CPU heatsink that allowed for a RAM-specific cooling apparatus that would be a viable solution too.

6 hours ago, Pri said:

So with this memory upgrade from my last post I have actually had a lot of problems with the temperatures. With my default cooling setup they immediately (within 5 minutes of booting) reached 98-100c when the system was idle and I had to shut the system down and figure out a solution. By contrast the previous memory only hit 62c or so, way below the 85c warning and 90c thermal shutdown alerts in the Supermicro IPMI.

 

So at first I tried of course increasing fan speeds, that didn't do anything. I tried inserting an extra 120mm fan pointed just slightly below the CPU heatsink so it hit the memory and also left the lid of the chassis slightly open to let more air in. This didn't resolve the overheating issue and the memory still hit 82-83c under load and was still climbing in temperature before I shut the system down to try something else.

 

Here is a photo showing the setup (which didn't work) with the 120mm fan added in front of the CPU heatsink assembly, pardon the dust.

 

 

As you can see I had a stack of U.3 drives positioned in-front of the CPU heatsink in the top right corner and I thought that was perhaps part of the issue as they do get quite warm, consuming up-to 19 Watts each and I have 4 of them for 76 Watts total. However, the CPU was still very cool only 35c idle and 45c under a high load with the fan RPM's at 900.

 

So I purchased a PCIe to 4 x U.2/U.3 card and moved those U.3 drives out of there and in doing so removed that rats nest of cables. This surprisingly did not change the temperatures of the RAM or the CPU at all. I tried positioning two 92mm fans connected together side-by-side aimed directly at the RAM below the CPU heatsinks intake fan, this did have an effect but the memory at idle was still 72c and would likely go over 80c in the summer when the room is hotter or under a high load.

 

Back to the drawing board. I thought about the airflow situation and what exactly is happening, well that big CPU heatsink is sucking in all the air before it has a chance to reach the RAM modules underneath. Ordinarily not a problem with low-to-medium sized UDIMM & RDIMM's as they consume so little energy but these 128GB LRDIMM's are a different beast entirely, I would estimate they are consuming somewhere in the range of 120 Watts under load across the full 8 sticks vs about 40 Watts for the 32GB RDIMM's I had previously.

 

So the solution to the problem, firstly I moved around a few of my PCIe cards, moved the U.3 drives to a PCIe card as mentioned previously and then I propped up two 120mm Noctua fans off-set from each other, one in front and one behind the current CPU heatsink and this has actually reduced the RAM temperatures to between 58c-62c idle and 64-66c under a stressed load. Very far away from the 85c warning / 90c shutoff temperatures I was seeing before.

 

Here is how that looks in the system:

 

 

I was afraid that removing the fans from the heatsink would harm the CPU temperatures due to the loss of static pressure but to my surprise the temperature has remained about the same 40c or so.

 

So some lessons here, firstly I knew before I got this RAM that these are intended for very high airflow server chassis. But since I work in the same room as this server I needed it to be quiet, in-fact the build with these 7 fans seen in the photo (3 black Noctuas in the middle, 5 brown ones in the main chamber) have had the needed result, every component is now running cool and the build is still whisper quiet. It's a little janky.. but it works and honestly if it didn't the RAM would have to go.

 

If I were to do another build in the future I would probably just go for the largest RDIMM's perhaps 8 x 64GB for 512GB total to not have such a complicated cooling setup necessary for LRDIMM power consumption. But the RAM is really good, the performance (latency and throughput) is identical to the RDIMM's I had before and apart from the tinkering I had to do with the cooling, works great.

 

Perhaps someone else will stumble across this post and find it useful, my clear warning here is, you need high airflow or a large case where you can pepper multiple fans to cool these high capacity LRDIMM's and perhaps even if I used a less-wide CPU heatsink that allowed for a RAM-specific cooling apparatus that would be a viable solution too.

Is there any ram heatsinks that fit this ram? Also do you think you may have to reconsider the CPU upgrade in future because of this? Glad you were able to find a solution. 

7 hours ago, Pri said:

I would estimate they are consuming somewhere in the range of 120 Watts under load across the full 8 sticks vs about 40 Watts for the 32GB RDIMM's I had previously.

Reasonable, never think those high capacity modules will draw so much power. ( I use eight 32G module, some build also Gooxi case )

Edited February 23, 20251 yr by Vr2Io

11 hours ago, Pri said:

Perhaps someone else will stumble across this post and find it useful,

Thanks, I had no idea that LRDIMMs used more power, my largest sticks are 32GB RDIMMs, will keep this in mind for upgrades, like my cool temps without any special cooling, just a little higher for the ones that are above the CPU, which is expected:

 

 

 

 

5 hours ago, tazire said:

Is there any ram heatsinks that fit this ram? Also do you think you may have to reconsider the CPU upgrade in future because of this? Glad you were able to find a solution. 

I would say heatsinks would help for sure, especially just to add some thermal mass that the incidental airflow can hit. Honestly it surprises me these kinds of high-powered modules don't come with heatsinks like FB-DIMM's used to back in the day which also ran very hot.

 

I've actually put the 7773X upgrade on pause as the seller I was buying from tried a bait and switch for an engineering sample version after I'd already paid for a brand new retail 7773X. Then a second seller informed me their 7773X's are locked to Dell motherboards so sourcing the chip has become complicated. But I don't think there will be any issues cooling wise, the 7443p I have now is a 200 Watt chip and it remains around 35-45c (idle-load). This heatsink from Arctic is quite good.

 

4 hours ago, Vr2Io said:

Reasonable, never think those high capacity modules will draw so much power. ( I use eight 32G module, some build also Gooxi case )

Mhm it is surprising.

 

1 hour ago, JorgeB said:

Thanks, I had no idea that LRDIMMs used more power, my largest sticks are 32GB RDIMMs, will keep this in mind for upgrades, like my cool temps without any special cooling, just a little higher for the ones that are above the CPU, which is expected:

Those are some great temperatures and indeed I like them to stay as cool as possible for longevity.

Today I upgraded the processor from the 7443p 24-core to the 7773X 64-core with 3D V-Cache. I've been surprised by the performance, the cores actually do want to hit 3.5GHz with regularity (with turbo-boost enabled of course).

 

 

I performed only a couple benchmarks from within Windows due to the extremely slow loading time of the benchmarks from a USB 2.0 thumb drive that I was running Windows from. Neither of these benchmarks will take advantage of the 3D V-Cache and so this is more just a raw core compute test than anything.

 

 

 

Obligatory Windows Task Manager shot for the hilarity that is displaying 128 threads.

 

As this was only Windows 10 Home it did detect the 1TB of RAM but only lets you use 128GB of it:

 

unRAID's main dashboard does actually struggle a little bit to keep the information refreshing smoothly with this many threads showing. Feels like it's missing some frames when it renders the animations on the bar charts.

 

 

But everything is working well, very happy with this chip already as you can see it is singing with my software and there's still headroom left left for more activities. So what is the next upgrade for the server? - Today I don't have anything planned but I think in the far off future (few years time) I'll likely buy a 100Gb NIC to finish it off, but only if I end up getting faster internet at home that would warrant it, currently the multiple 10Gb connections the server has is perfectly fine for my needs.

Edited March 11, 20251 yr by Pri

Nice build. Just built a machine last week that maybe you'd appreciate. Originally-slated as a NAS server / ProxMox VE, but decided against it, figuring it was too overkill for such a use-case. AND, it seems that virtualizing UnRAID is also a very bad idea. Just coming into the HomeLab space and getting my feet wet and figuring things out. The machine also had 19 physical drives connected to it the other day... no external HBA card—all connected via motherboard connectors (ASRock TRX50 WS). The DIMMs are (4) 256GB ECC at 4800 MT/s. 

Took this screenshot last week... CB nT score blew my mind.

Edited March 11, 20251 yr by dondi

That's an excellent score. I was thinking about going Threadripper myself in the future due to the clock speed advantage they have. Hopefully AMD continues to release Threadripper Pro

Little update on this CPU upgrade. So 4 days after I installed the 7773X the system began to randomly shutdown (complete power off) then start again on its own. The logs in unRAID revealed absolutely nothing. Never any errors before or afterwards, just completely random restarts.

 

The only common denominator between the restarts is that the system only restarted when it was idle. Even being completely stressed for 40 minutes it wouldn't restart but when left idle for a while there was a high probability it would shut off. In-fact I was able to induce this shutdown several times by switching from a high-load to low-load scenario on the CPU.

 

So I tried the following:

1. Disabling C-States

2. Updating BIOS to latest (I was 1 version behind)

3. Reducing chip TDP from 280 Watts to 225 Watts

4. Reducing RAM speed from 3200MHz to 2933MHz

5. Re-seating the CPU in the socket

6. Disabling unRAID power states (trying to eliminate low-energy idle states to rule things out).

7. Removing the USB cord I had my unRAID drive loading from so it was plugged into the mobo directly.

 

None of this worked. The maximum amount of uptime I was able to achieve apart from the initial first 4 days was about 2 days. Most of the time it shutdown within 2 to 6 hours. Sometimes I even had it shut down immediately upon reaching the unRAID GUI.

 

Final thing I've tried, I put the old 7443p CPU back in the system. It has been stable ever since, that's almost three days uptime and prior to this the day before I changed CPU's it crashed three times in one day with the 7773X.

 

So this all leads me to believe the CPU is faulty, doesn't appear to like going idle. Possibly it degraded quickly after that first initial four days of uptime, it is a brand new retail chip and these things do happen especially with X3D chips that are lets say a bit more fragile. With regards to the IPMI on the motherboard, it only ever gave me a single log entry when these shutdowns occurred and it said the CPU was "Out of Spec Definition" which again points to the CPU, I would receive this error every single time the system started back up after shutting off and no other errors were ever reported.

 

Right now I'm just going to wait until I have I think 5 or 6 days uptime on the 7443p to fully rule out anything else and then I'll RMA the 7773X and try again with a replacement. It's a real shame but I've only lost some time to this problem and luckily I have the old chip to use in the meantime. I really love the performance the 7773X has delivered for my workloads and it actually runs about the same temperature as the 7443p even while having almost 50% higher TDP requirements.

Edited March 22, 20251 yr by Pri

Thought I'd update on the CPU situation. So the 7443p I was originally using in this build was stable for 11 days after I removed the 7773X which kept crashing. That was definitely confirmation to me that the problem was the 7773X.

 

After that 11 days of uptime my replacement 7773X arrived and I put that in one week ago. Since then the system has been completely stable and I'm able to enjoy the new processor to its full potential. So the upgrade was a success but that first chip definitely suffered from infant mortality.

 

For those curious I bought the chip brand new from tugm on ebay, he's quite a well known seller in the Homelab community and with good reason. He was happy to replace the chip for me and even pre-paid for the return shipping with FedEX express which only took 3 days all the way back to Hong Kong. Really good seller and I just wanted to give him a shout out for the excellent service I received

13 minutes ago, Pri said:

Thought I'd update on the CPU situation. So the 7443p I was originally using in this build was stable for 11 days after I removed the 7773X which kept crashing. That was definitely confirmation to me that the problem was the 7773X.

 

After that 11 days of uptime my replacement 7773X arrived and I put that in one week ago. Since then the system has been completely stable and I'm able to enjoy the new processor to its full potential. So the upgrade was a success but that first chip definitely suffered from infant mortality.

 

For those curious I bought the chip brand new from tugm on ebay, he's quite a well known seller in the Homelab community and with good reason. He was happy to replace the chip for me and even pre-paid for the return shipping with FedEX express which only took 3 days all the way back to Hong Kong. Really good seller and I just wanted to give him a shout out for the excellent service I received

Have to agree with you on that seller. That's who I used for all my gear and it's never missed a beat. Glad everything has worked out. 

Small update to the server. So I've had the RAM in there a while and I did solve the overheating issue a long time ago by cutting some plastic packaging with scissors to make an air baffle. I wasn't really happy with that solution long term so I have designed and printed a proper baffle at home.

The left baffle is a bit thinner than the right side so that it fits with a PCIe card in the top slot. But of course this may differ based on your motherboard.

Link to the model for printing: https://makerworld.com/en/models/1712778-arctic-freezer-4u-m-ram-air-guide-baffle

And I've included some images of the baffle installed too. Temperature wise I now see both sides of memory at 61-64c where as before the right side would regularly hit 71c with the janky cut-up plastic baffle I made from some packaging material. Quite an improvement and a lot more sturdy and dependable.

Included in the model is a mount for a 120mm fan so I could in the future mount a fan back on the heatsink in the middle, I may do that and use two rows of 40mm fans on either side for the RAM baffles but right now I'm happy with this as-is, CPU temps around 60c under load aswell.