All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
More SuperMicro Problems
Just received a batch of SuperMicro boards to deploy in Hong Kong. This time, a bunch of X11SSL-F boards.
Sadly, yet another couple of faulty boards were included in the batch. We order new gear every couple months or so, and much of it will include some supermicro equipment. At least every other shipment will contain at least 1 faulty piece, and this works out to be around 10% of all new SM equipment we receive having some sort of fault.
Luckily our latest order was from our EU vendor, rather than the HK vendor, and they are making the whole RMA process much easier for us, and apparently trying to get us a new one shipped out ASAP, so we need not wait for months while the faulty one gets repaired. This is much better than what I can say of any HK based vendors who not only do not keep stock, and so end up having long delivery times, but also have extremely lengthy RMA handling times which can (and have) literally last several months! The warranty, when going through your vendor, is worthless in such circumstances, so at least our EU vendor is really making things much smoother now.
This seems to be more an issue with NEW SM gear. We also buy a good deal of second hand gear, and I can honestly not remember any of the second hand stuff we've bought to have any problems, and again, I do not recall any of them ever failing while in production. I suspect SM are somewhat lacking in the QC department. Having spoken to our vendors about this, it would seem they have the same assessment...
Rant over.
Comments
Out of curiosity, when you receive your servers, what sticker do they have on them, assembled in the U.S.A. Or assembled in the Netherlands ? Or something else ?
Perhaps there’s an issue specific to one plant...
I just got a delivery of 144 nodes (12x 12 node system) and had 0 DoAs, we sometimes have some, but I’d say well below 5%, more like 1 or 2% - could be the small probe thing down to bad luck though...
We also got some new machines this month, so far never had issues with new SM stuff.
Are you getting them assembled and tested from your vendor or parts for local assembly?
This particular shipment is just for motherboards, not complete servers.
The boards are made in China I believe. And our MicroClouds are assembled in USA I believe (they were shipped FROM USA).
I've ordered new SM gear from 3 separate official vendors. 2 in HK and 1 in EU, and I've had faulty gear from all of them. Around 90% of what we receive is fine.
Aye that would also be a good question.
Worth getting them assembled and tested - if you negotiate right. We pay around 50 USD for the pleasure per system of 12 nodes, and naturally this includes testing, so it could potentially heavily influence the stats on our end. You don’t need to buy any drives for assembly at SM plant with MicroCloud - 1x CPU + 1x memory stick is the minimum config.
After all if you buy a bare one you’re not a consumer but an integrator so you absorbe some risks that otherwise an integrator would.
No, we are testing and assembling in house. And before someone suggests that it may be US mishandling the equipment, that would suggest we would see the same kind of problems across the board for all brands and second hand boards, and we are not. We We have handled quite a bit of second hand hardware, and as I said, I cannot remember the last time we had a DOA or faulty board shipped in. So it's not us being careless with NEW SM boards, and being super careful with everything else.
We get free assembly and testing from our vendor in Bulgaria and most likely this is why we've never had an issue so far.
Yeah, then it’s a different thing.
When you buy a new one from Dell you don’t buy a barebone, you buy assembled server. If you don’t buy an assembled unit from SM, but a bare one then we’re comparing apples with oranges.
Testing requires memory and CPU hence why other than some self test functionality in their boards I wouldn’t anticipate any more complete testing being done on the barebone / parts before shipping.
Negotiate having them assembled and tested either by this EU vendor or by SM themselves.
The problem is that getting complete systems adds massive cost.
For 1, official SM chassis are crazy expensive. At least compared to what we normally use. We also have a huge supply of unused chassis, so we WANT to use what we already have instead of buying more.
Secondly, the shipping cost is expensive enough as it is. A board + RAM + CPU + heatsink in original boxes weighs about 1KG. A fully built out system would be at least 10KG. And SM chassis seem to be made of lead cuz' they weigh a damn ton, and probably pushes the min weight upto 20KG. It's too much.
If we could get them assembled for $100 per box, and not worry about faults, because they server builder takes care of that, then I'm happy to pay it. But the reality is that it costs much more than that, and the local HK vendors are god awful. For example, if you want to make sure that what you get works, and have guaranteed same day replacement for the duration of your warranty, the price is almost twice as much! If you don't pay this premium, then you need to wait months for the RMA to go through. So it's cheaper to just buy 10 - 20% more than you expect you will need, rather than pay for their premium replacement service. Not to mention, they won't support your server unless you buy ALL components from them. That includes RAM, CPUs, HDDs/SSDs AND chassis. So the warranty is truly worthless in HK.
Push the vendor to bring the price down, or calculate the risk of DOA in as an integrator. I don’t think that adding a CPU and a single stick of memory pushes the shipping cost much if at all, also, they can be aggressive on component costs too . Not too many more options to be honest. Any barebone you buy will have similar issues - again probably not close to 10% but a visible single digit number is to be expected.
We already order the CPUs and RAM from the same vendor. Actually its sometimes even more expensive than buying from HK, but it's just easier to deal with them because they will handle our issues pretty quickly.
It's the chassis that ends up costing a huge amount, and I don't think you can ship a board with the CPU + RAM mounted to the board without completely repackaging. Though I suppose I could request they test all the board prior to shipping... though in my case, the vendor sources parts from multiple locations, so although we got the boards, ram and heatsinks shipped over from the EU, the CPUs came straight from China.
I've done the calculations. It's cheaper and more cost effective to just buy more than we need, and assemble everything in house. This results in cheaper and faster delivery times, and so long as the failure rate is within 10 - 20%, we normally only need to have a couple of extra boards on-hand to hold us over. Having the spare parts in house also means recovery / repair times are reduced, since even if paying for the same day replacement premium service will mean 6+ hours before they come and actually do anything.
I don't think there is anything we can do to make things better. This fault rate is just insane. I've NEVER seen a faulty CPU. Faulty RAM is less than 2%, and I've also never bought a consumer grade board that had any faults within the first 6 years of operation. I still have some old E6550s purchased back in 2007 that work well today. Back in 2007, we bought over 100 E6550 CPUs and cheap Asus desktop boards. Each server ended up costing about $400 or so. It wasn't until 2013 or so before we saw the very first one fail, and I don't think we've had more than 10 failures in the last 11 years. To be fair though, most of them were decommissioned years ago because they are just not very useful. But still.. very very low rate of failure for consumer grade crap.
Did you ever try Fujitsu Primergy servers? Very attractive pricing.
Importing IBM/Lenovo/Huawei/Fujitsu servers from japan or china is really interesting, you pay like 800-1000 euros for a Dual E5 2630v4, 32-64G DDR4, 120G ssd incl rails and dual psu.
--
I have had a couple of Supermicro X8 servers (Dual E5 2620v2), but 2 of em had IPMI issues within 2 months, from random unavailability to even non-booting anymore. From my experience, Supermicro is either a hit or miss. I have had both really good and bad experiences with them.
Then you've bought all your boards after https://en.wikipedia.org/wiki/Capacitor_plague
Nope. Need to look into vendors.
That kind of pricing sounds pretty good, especially if you're shipping to EU.
@randvegeta - A bit off topic but what sort of RAM you putting in the X11SSL-F's? I was looking into building one for personal/home usage but the 16GB sticks cost twice as much as the motherboard
Yeah, that's the pain... And they are the most sought for which brings their price even more.
https://www.theregister.co.uk/2018/04/30/dram_vendors_sued_again_for_price_fixing_again/
You can join the class action in your country
I heard suing, so I am here.
Samsung M391A2K43BB1-CRC (16GB 2400 UDIMM).
I only buy the largest sized sticks for the E3s. Mostly fully loaded, so 64GB each. Cost a small fortune!
For RDIMMs, I order 32GB sticks.
Solid capacitors ftw
Actually I don't think I've had any problems related to capacitors with SM. Not as far as I can tell any way.