This pretty much mirrored my experiences with a 6380 on this board. There were a couple things you missed out on too (If you did touch on them, I'm sorry!). - The ram is not dual channel at all. Even with 2 sticks it remains in single channel, additionally the reason most software cannot identify the ram timings is because of the assigned node. Both sticks are handled by only one of the four IMC's (two IMC's per node) and software generally tries to read it from the unused node, failing to read it spectacularly. Reported it to Martin to correct this in HWinfo a while back. - The board maxes out at 1600mhz as you've noticed, I found no way to override that. Handles ECC/Reg ram just fine however. Unfortunately the RAM latency tanks because half the CPU needs to cross nodes to access it. Was reading upwards of 140ns latency and higher compared to a consumer AM3+ board. - FPS takes a 50% loss in comparison to the Opteron 3380 in my other rig at the same clocks during testing under OpenGL and DX11. I never could figure out why however but doom did run just fine under vulkan and 'close' but still under the 3380 FPS results. Tested with a GTX Titan X between both setups during comparison. - Usb3... Usb3... It's not real, it's very much fake. There is no USB3 controller on the board and contrary to the claims of USB3 support (the blue header) it is simply a recoloured USB2 port with a higher power capacity. - Using CorePrio by Bitsum helps a bit when it comes to thread shuffling to increase performance but it only works so much. The G34 Opteron suffers from similar issues with it like the earlier days of Ryzen Threadripper.
Yeah if going piledriver then the fx 83xx or 63xx are the only real choice the 8350 scores almost the same in cinibench with its 8 cores(?) cos of higher clocks and they do run the tested games no isues 2060 would perhaps be bottlenecking lol but fps would be playable both doom and csgo get over a 100 fps with 8350 and a furry nano know this as I played both these games on with 8350 and r9 nano
Hey Everyone! Before we shart on Socket G34 and everything else related to the Piledriver and Bulldozer server platform I want to mention a few things! I've dealt with these chips for a while now, more specifically in Dell servers and I can tell you straight off the bat, Jeff's x89 is ruining everything. These chips, I will not lie aren't great at single thread ESPECIALLY NOT GAMING!!!! (making that very clear, these aren't like older, bang for buck Xeons, who are good at gaming). But, I can tell you first hand, on Dell, Supermicro and HP motherboards, these things are meant for VM's and lots of them. Running on a proper motherboard, these things can multitask and multi thread like nothing else. The 6300 series (Abu Dhabi) (6380, 6378 and 6366 HE especially are compute champs for their value (6386 SE is a power hog) (Two 6386's can about touch one first gen EPYC) I would not recommend buying into this platform, if you have to buy a motherboard with it. Good motherboards for these chips are very very expensive, as well as coolers and are hard to find.. I think the only case you should be buying to this platform, is if you've found a good deal on a server chassis (which very much exist for this platform.) Servers like the Dell R715 (better R710 in my opinion) R815 and others in the Dell 11th gen "Rxx5" family are all, AMAZING servers and run this chips very much as intended, unlike this "X89" junky motherboard. The HP DLxx5 servers are good as well! But in my experience the Dell servers and Supermicro motherboards are amazing. My Dell R715, loaded with 15K SAS hard drives, dual Opteron 6380's uses about 200 watts for what I do What I'm trying to convey is that, if you are looking into getting into homelabbery and you need cores, on a budget and single thread isn't a HUGE deal, don't be scared off by this video, as these chips are really good, in certain use cases! This video is NOT a accurate representation of Socket G34 as a whole! Now, what I would not recommend is really going for anything but the 16 core Abu Dhabi chips, as really at that point, just buy an Ivy Bridge chip or something else, you're better off with Intel. I know these chips have BAD single thread, but they aren't as bad as the ones in this video. My R715 actually used to host a BungeeCord network (Minecraft) and performed pretty admirably with upwards of 150 people spread off across my network on 12 Abu Dhabi cores! (Mind you, you won't be getting that on a single server... Single core still isn't great). They've run Plex transcoding beautifully and Code Compile is wonderful on one of my 24 core VM's. It is a pretty snazzy render box and i've been happy. The chips are cheap (80 CDN for 32 cores {2 Opteron 6380's off ebay }) but motherboards and coolers are very very costly and not worth your time. This platform is only worthwhile if you have a preexisting high quality server chassis, and power is cheapish ( At 0.21$ cdn per Kilowatt, my main R715 at home costs about 18$ cdn a month...) these are really fun to play around with, if you need lots of cores for VM's, Code Compile or a multicore render farm. That is the only reason i'd buy these chips. I really hopes this helps at least someone! This platform isn't complete garbo, but special and has its niche and this video is NOT REPRESENTATIVE OF REAL WORLD SERVER PERFORMANCE AT ALL! (if you have a good motherboard ;) ) Great Video as always though Jeff! I'd love to see you try again with a proper motherboard :) P.S, These things don't game..... I tried it with a 1070 and it was the funniest bottleneck i've ever seen.......
A few weeks late, but thank you for this. I'll be honest I bought one off amazon as they were selling for less then 10$ just to frame it and use as wall art.
It would be great to see the performance of this processor if it and all the other components seen in the video were plugged into a motherboard from a quality brand, such as the Supermicro boards that can be picked up second hand on Ebay.
The funny thing is: I have seen that exact processor tested on an actual proper motherboard(not tossed together garbage) and it handled doom just fine. Yes, Piledriver was a pile, but if you come across some of these in a junk store somewhere it might be an ok system;assuming the motherboard you find isn't equally shit.
I've ran a FX 6300 from, 2014 up until January. It couldn't handle DooM 2016 too well, but Doom Eternal 1.0 (and only the 1.0 release day, I don't know why, I bought the game but always ran the pirated version cause it gave me much better FPS) ran fine, it wasn't pretty, but I even managed to get ultra nightmare done. Nowadays i5 11400f + 2060 12gb (it was cheaper than 3050 8gb).
Wait. I didn't expect it to be properly fast, but it barely POSTs, seems like. Why would it lock up Doom for example? Then again 900cb in R15 seems kinda alright. I mean no not that it is alright for the class or the power budget, but the FX-8350 does like what, 660cb, and you get double the cores here at the price of a clock disadvantage, so 900 sounds like about fair. Bulldozer/Piledriver had a bit of an FPU deficit in workloads with high SSE/AVX utilisation, by design. I wonder if it's similarly bad in Linux or whether it does any bit better there.
I got around 1700 with 2 of them, and around 3700 with 2 of them on R20. I agree that it's not that poor. I think motherboard maker can't write BIOS or configure all the component good enough to make this work correctly.
You can try the new x99 motherboards and e5 2620v3 with some cheap ddr4 ram and a bios hack it's pretty good value Tech yes city made a video about (I believe that you've already seen it)
I have one with the E5-2678 v3. It's been pretty good so far. I just have 32GB of RAM, and a PALiT GTX 980, along with a noname 1TB NVMe, and a Toshiba disk. (All my data is on a remote share, so I don't really care about storage).
Since most of ur problems are in the BIOS.... .... what if you flashed with a BIOS with a BIOS from the original motherboard that the chipset came from?
Hey Jeff, a professional audio engineer here with a couple suggestions. First you should high pass your microphone up to around 150hz and possibly notch out a bit around 200, it will considerably clean your vocal up and get rid of that muddyness in the lows/low-mids. Also regarding the audio dropouts, not sure what that is but seems like maybe a buffer issue with your driver. You also really gotta get some treatment on the walls as well to kill all those reflections being picked up in your Lav. Cheers!
1. 3.2 ghz is really low clock for Pilrdriver, as they scale up to 4.6-5ghz. 6836's clocks are comparable mobile APU-s 2. i5-3317u boosts up to 2.6 ghz in ST loads. 3. 6386 SE is not 16 core CPU, it is 8 module CPU. Bulldizer line in particular can't be compered core to core against other architectures.
Re: gaming - Was the IOMMU disabled? Leaving it enabled can have pervasive, unpredictable stability effects when you push around lots of data on the PCIe bus. Also wondering how badly the dual channel memory configuration is holding back performance on a line of CPUs intended for quad channel operation and famously ravenous for memory bandwidth.
Well , my e5 2640 runs on dual channel and has performance almost on par with ones on quad channel , this may be thing that i got good bin or good 1333mhz 2x4gb sticks.
"how badly the dual channel memory configuration is holding back performance" very... specially since while it have 2 sockets it runs single channel only so it's not even dual-channel.
Børre Børresen That’s what I was afraid of - each NUMA node only has access to a single channel, which will starve performance. Anybody committed to playing with one of these old Opterons really needs to snag a used socket G34 motherboard from a name brand.
@@Halon1234, i think not only do they run single channel, they also are both in the same NUMA node. yeah... this board is really a big pile of garbage and the price is so high that you could add 10$ and get a second hand dual socket Supermicro, Tyan or Asus board (and most of the Asus boards will give you a tiny bit ability to overclock, or quite a lot if you get hold of some ES versions of the Opterons... just beware of cooling)
@@brrebrresen1367 Are you sure that it run only in Single channel just from one NUMA node not from both? If it's true, this will got performance hit from both Single channel and the same hit that 2990WX got. I don't want to imagine how bad it is. P.S. You must pay at least 40-50$ more than this board to get something like SuperMicro H8SGL on used market, at least in international platform.
A low clock server CPU doesn't do well with your desktop workloads? shocker. These CPUs were intended to be used in applicants like ESX, where you need a lot of cores to support VMs...
I had one, and it lived for two weeks before it started rebooting every few minutes, I tore the heatsinks off of the chips to find OK written on top of the northbridge chip not new..Salvage??? I have pics.
When I saw this notification my first thought was "WTF is X89?!". At the end, I just start thinking, once again, how much AMD has improved in the CPU department. It is really impressive.
6386SE for just 55$!? Over a year ago I bought 2 of them for my working PC and they cost me 300$ included shipping. I think all because of the garbage motherboard that can't work correctly (when I saw BIOS screen, I can't expect to be able to configure anything that should help solve the problem). I have never encounter such problem showed in this video. I can play games or edit video just fine and it's working right from the start without doing anything. However, I used SuperMicro dual-CPU motherboard (which is god damn expensive for such a platform so I can't recommend, Asus one is OK), so it's fairly stable that I haven't encounter any system crash or any real issue until now. The only problem I have is when first plugged-in and power-up the build for the first time, video card always run at x8, but it will be x16 if restart the PC or power up again without unplug it. Nevertheless, I still can't recommend anyone to build PC out of G34 platform. Even with 2 of this CPU and full Quad channel memory for each, they can just barely outperform Ryzen 7 1800X. Also, it's not good for gaming with lower clock speed than FX-8350. And if you use lower model than this such as Opteron 6380, despite 32 cores and threads, can't process footage from HD capture card correctly because it have too low clock speed. (When you open OBS/XSplit, if you are not recording or streaming the preview footage on OBS/XSplit will be freeze then move and freeze again for around half a minute before the footage completely freeze, but if it's during recording or streaming it work normally but it will be the same as soon as you stop recording. I prove that it happened because of CPU's clock speed when I tried the capture card on my FX-8350 build and the same problem appear only when I underclock it to 2.40GHz.) In addition, those Chinese boards are really garbage as showed in the Video, and even outside of those, all branded motherboard are meant for server-use so they have very limited I/O (2 rear USB ports, dual LAN, one serial and VGA port, and nothing else). You have to spend additional money on Sound Card and USB host controller card to be as usable as normal PC. All of these still not consider how hard to find the heatsink is. In other word, anyone must stay away from this platform, it will not value your money. I bought it just because I (irrationally) want to own something not many people owned.
Glad I found this review before buying 2 of this board. I has 2x 6276 in a Supermicro H8DGi until the board started randomly rebooting after 45-60 minutes of use. No BSOD, just as if someone pushed the reset button. Never had performance issues. Was looking into what to put the CPUs into and it definitely wont be what you tried. Maybe an Asus KGPE- D16 if the price is right (and if the price is right for a case as well).
The CPU was actually pretty decent back in its days for heavy multi-threaded workloads and running VMs, but not with a rubbish motherboard like this. The main problem with these "fake/scam" boards is not the BIOS, but power delivery, as they simply cannot deliver enough power at a high enough consistency or efficiency to the CPU and memory, and this might be partly the reason that Sleep mode never works on any of these boards (fake X58/79/99 included). Those fake X79s are actually fine, since they've been making them for years now, and there are BIOS hacks from Chinese and Russian websites. But this is nothing but a scam, baiting people with the high core count of cheap Opteron CPUs.
This doesn't surprise me. My old Vishera used to score 800 cb screaming its lungs out at 5.2ghz and consuming over 230w. If you account for the extra cores, yet huge losses of clock speed the 900 is about right. Those were actually big numbers back then. They put I5s and indeed I7s to shame, but weren't good for gaming or other tasks that requiired IPC. I'm glad you did this though, and that you have one of these up and running as there was *never* a cheap board for these so hardly any one has them. I see you like the same sort of things I do, re-purposed server gear so at least you have this for historical purposes. I still loved the video. Maybe next time go into it with less expectation. Thanks for another one that I found much more interesting than most of the tosh on here.
The consumer FX-8350s were really handicapped. If you run Opteron 63xx ES chips (piledriver cores) with 16 cores, you can downcore alternating cores and get Piledriver cores that are not sharing L2, decoder, L1 inst cache and other parts of the pipeline with the other core of the compute unit, netting a 15% boost in IPC.
I see where you are coming from, but I got the same board with no problems. Found an Opteron 6328 server in the trash, and bought this board a year ago. It is what it is and games well with 16gb of ram and 1050ti. You might have gotten a bad board.
The best combo in this video is Jeff and the booze, because hardware failed completely. TBH I wonder if the CPU or chipset weren't overheating with that little coolers. If I recall correctly older AMD platforms had some weird temperature reporting, where if you were starting to get close to 70C that meant your CPU is actually close to the melting point. I would like to see if swapping cooler for something bigger and checking chipset cooler mounting along with VRM heatsinks would net some better results, but just out of curiosity, because the platform itself is based on a failed architecture.
I messed around with one of these X89 boards - got a little bit ahead with the BIOS-MODS forums for it (was Jingsha branded) - we unlocked voltage settings and RAM speed settings along with Hypertransport settings and a few other things. He was working on getting CPU overclocking unlocked, but I got impatient and force-flashed the wrong BIOS to my board. Only way to fix it is to get a chip programmer and reflash, but that project has been on hold forever. Sooooo if you want a project, check out the bios mods forums and search for Jingsha, carry on where I left off, get some overclocking going on that Opteron and let’s see what 4.5GHz of Piledriver ridiculous with only half the quad memory channels enabled can ACTUALLY do!
On certain Supermicro boards you can get a custom bios for overclocking. I've overclocked my quad socket 6378 rig up to 3.0 ghz on all cores. Couldn't keep them cool for obvious reasons. I've wondered how far a single socket could go.
Bought an x79 MB with nvme, E5-2650 v2, 32GB ECC DRR3 1600Mhz direct from AliExpress for £138 delivered. It's been running flawlessly for 24/7 3 months.
Most of these Opteron's on x89 have a TDP of 115. It wouldn't surprise me if they ran better on this economical motherboard. Not all motheboards/server boards seem to support the 140 tdp SEs. But I don't know. The first case you posted came with a power supply that was rated for 380 watt if I read that correctly which might be a bit low. 140 tdp for the processor + 160 tdp for the gpu alone.. Not sure what power supply you used for the last build. If you have no interest in the economical x89 motherboard,cpu fan, and ram, I have a interest in the platform if simply for the peculiarity and uniqueness of the platform. Thank you for the video of the build. It was entertaining to watch.
I'm curious what the performance would be like on a server board from a reputable brand. Too bad those boards are really more expensive than they should be on ebay.
@@CraftComputing Will look more into this These old opterons seemed way more interesting to me than they actually should be. Thanks for the great video sir.
I've seen others review this board and it's always the same thing, smack in a high-powered server-CPU in that socket, get some of that ECC-RAM, run Windows [Insert edition of choice], stuff in a modern Nvidia GPU, then complain that a server platform does not behave as a desktop board, no overclocking options etc... I like Craft Computing as much as I (and they) like beer, which is A LOT! But this review did not pack the punch and deep-diving that I hoped for. Modded BIOS:es should even exist for those Chinese boards. I happen to own the very same model of Jingsha motherboard that is the subject of this video. -Yes, there are others from more reputable brands like SuperMicro or even ASUS boards for the G34 platform, though the ASUS ones are very rare! Most of the times they are multi-socket boards, meaning you can use several of these CPU's in one motherboard for multi-parallell workloads. -Not suitable for desktop OS... Just sayin'... This board could be found for about $50-$55 (including the cooler) about 5-6 months ago, now the prices have gone up to almost the double from what I see on places like Ebay. And while this particular board has gone through some REALLY strange and even kind of incomprehensible design choices (-like 2 RAM-slots on a 4-channel memory server platform??) the G34-platform really isn't a gaming or desktop platform what-so-ever, for what it's worth, it actually does not perform that bad under Linux for the workloads that it is supposed to run: Spoiler alert! -Not games and desktop applications. But yes, you can run desktop applications on it under Linux, only difference is they actually work better than on Windows. Of course it cannot measure itself against the way more recent, more effective Zen-platforms. -Even though Zen and the older Bulldozer architecture are practically ugly cousins, favoring the focus on more cores and interlinking them. Zen only does this much better as it a lot more recent platform and a re-design to counter the shortcomings of the previous generation. Thus allowing for higher speed, better power-efficiency, higher IPC, granted more headroom and above all: Built on 12nm(and also 7nm w. chiplets) compared of the giant lithography for the Bulldozer at 28-45nm(!) for the oldest parts. Granted the G34 platform cannot measure itself with Intel CPU's from the same era but seeing all tests being performed by TH-camrs on Windows, doing desktop tasks and gaming is almost like them finding an already dead, rotting horse in a ditch and going through all the pain of dragging the carcass back onto the road again, only to then kick it back down into the ditch! :D It's almost like a "Just because you can, doesn't mean you should!" scenario with this platform. The easy explanation; Linux handles cores and threads (and modules, in this case) a lot different (better) than Windows has ever done. Linux rules the world in the server-space and is has ruled it for ages, this is a server CPU, it is meant to do "server:y" tasks, parallell stuff, databases, web requests, things that desktop Windows is not good for, as Opteron CPU's was never intended for that desktop market. Remember the AMD FX-platform that also did not compete very well with Intel's stuff? -Yeah, that was the CPU's intended for desktop tasks. They excel over the Opterons in every way when it comes to desktop tasks because of their higher clock-speeds and aim towards other kinds of workloads. This is a server CPU and a pure server platform, in server scenarios it works quite well, so if you want to build on an old server platform that does not have as many *known* flaws (as of current date) -compared to the Intel platform of the same era (-and even later platforms) good parts are often very cheap. We're talking server-grade parts, not consumer parts. Combine this motherboard with low-powered Opteron CPU (=not the CPU used in this video) e.g. the 6366HE: 16core@85W or a 6262HE: 16core@85W or even a 6128HE: 8cores@65W) an it will work pretty nicely. Again: Please use it as was intended, as a >server< with Linux, that is... Run Docker containers on it, run a file-server(sftp), use it for Syncthing, Webserver, a streaming server, as only a few examples. Multitasking and web-requests is what this platform does well, so let it do ONLY that! :D
Thanks for your feedback, but here's the deal as I mentioned at the end of the video. ANY task, server, workstation, or gaming, will perform better on a Ryzen 1600 for less money overall. You'll get more memory bandwidth overall running 2666 dual channel vs 1333 quad channel, and around 30% CPU time improvement. Even for server workloads, the Ryzen will perform better. That's why this received a negative review...let alone the freezes, stutters and hard locks experienced on the system.
@@CraftComputing Agreed, this old server platform cannot measure itself against the newer Ryzen 1600, of that I'm certain, and that is true for most of the single-socket server hardware released in that era. ;) But it is actually does not perform nearly as bad when it runs tasks on Linux, when compared hardware released in that era, that is... Just a question: The cooler that you got with that motherboard, did it have any rating written on it? I got mine in a plain no-name box and it seems to be rated for 115W TDP which most Opteron CPU's was limited to at the time, your CPU in this clip (I think it is a 6386) has a rating of 140W TDP, could that be the reason for the constant hangs you experienced? Given this this system is indeed a server system so it usually throttles at around 50-60C to run stable for server operations. Props for the VERY fast response! Love your channel!
"Just because you can, doesn't mean you should!" scenario with this platform. --- WELL SAID. The way these CPUs are described on the TUBE is just confounding and makes little sense... You have brought reason to my madness. "Use as it was originally intended". "Don't assume". "Don't test with a multitude of unknowns". Monday morning (decade!) quarterbacking is always EZ. Thank you for you input :O)
The opteron, while an awful CPU, shouldn't have been hard locking and doing that nasty shit. That motherboard is definitely a pile of crap. PhilsComputerLab did some opteron content a while ago, and while it sucked utter balls, it didn't do that janky nonsense.
Are you sure the GPU was fine? I've gotten similar GPU issues where it hangs for a short period before recovering too. Also, have you tried CPU-Z for memory stuff?
I've run Doom 2016 with a 2.4Ghz Celeron before. And it never stuttered. Granted it was paired with 1070, but it never had issues, just slow load screens.
A good fix for the fan would be a physical fan control like the ones ZALMAN used back in the day. Thaiphoon Burner is a great program for reading ram and all of is IC's and speed data. Some 3rd party programs can OC some of those older chips in the OS. Could be worth a shot to try some of these suggestions. Good luck with the X89 build. Cool video though.
I would say the motherboard (and partly software setup). I have a dual 6344 (lower clocked 12 core) and it scores just as much single threaded than this faster 6386 in cinebench. And well games, without quad channel, pretty low IPC and single core performance, and the situation worsened due to 2 numa domains I am not surprised things don't go that well. My machine is an absolute beast though for what I built it for, which is compiling big software projects.
That type of AMD coolers are actually super common -- ALL of the Opterons used the same one. Thing is there is no market for them so companies generally just toss them. I have something like 250-300 of these just laying around, highend low profile coolers, 2 heatpipe mostly solid copper with a aluminium mounting plate. Has no problem dissipating heat from older 6cores at the very least.
That is... actually true, performance numbers nearly triples on Linux-OS compared to Windows-OS, judging from from my own tests of Solus 4.1 vs. Windows Server 2019 and Windows 10 1909. No joke. -Though triple numbers of that magnitude is not impressive in today's computing, Windows just makes this platform look even worse than it actually was! :D Also: Windows is trash when it comes to multithreading, that has been known for a long time, and as has been showing very clearly in the likes of the AMD Threadripper platform, just juggling threads around in Windows causing 100% CPU usage for tasks that does not even utilize 50-70% under Linux. It's getting better, but Windows is still far behind on that...
I totally understand the sentiment, of a anger being [somewhat] quelled, with a slow sip. Comedic effect could've been progressively draining your Speyburn, leaving the filled glass unattended. Now in seeing an abysmal blind date, of a CPU to that sorry excuse of a [piece mealed] motherboard. Would you consider testing that CPU, on an established motherboard? [say a Supermicro, ASUS, etc.]
I really liked this video. I was thinking about an Opteron for my Plex server, just for the sake of it (and because the current AMD server processors are way out of my reach. And because I though a 6378 can manage the transcoding in all devices at home). I don't think the CPU is just crap (but it is definitely bad), I think the motherboard plays a major role there. The thing is, by the time you try to buy a G34 compatible motherboard, you've already spent Ryzen money. That's how my "stupid" home server project with an Opteron died and got replaced by a Ryzen.
Right, because THAT was the limiting factor in performance... Nevermind that this is a cooler blowing down on the VRM with an open top, and CPU temps never exceeded the low 60s.
wow that things actually slower than my dual xeon from 2010.Given it's a 12c/24t and it's barely slower than my r5 2600 difference is i use her for a storage server and a virtual box.It's in a 2u rack mount so no audio and only 8mb matrox display but for what i use her for it's perfect.I was fascinated by the opteron back in the day.I built 2/3 of them in my life but it was proper server mobo's and it worked perfectly.These chines knock offs are just not good for amd.For cheap xeons etc they are great though
You could give some better chinese boards a try. Look at Jingsha X99 with DDR3 (for cheap memory) or DDR4 version (same board, maybe sold as KLLISRE). Paired with a 2678v3 for DDR3 support, unlocked turbo with a BIOS mod, and you have a 3.2Ghz all core 12C/24T rig.
I've been eyeballing it for a while now. I was really looking forward to his review on it at the end! Are the reviews not happening anymore or was it missed due to stress??
evidently a host of issues with this board, first off being opteron 6xxx is quad channel, second that cpu is CMT enabled so it should be 16/32. Yes I know about the whole "were they true 8 core modules." That cooler wouldn't cool a 95w cpu properly, let alone the 140w from this cpu
That was some abysmal performance Ha!. You should do a followup when you explore the modded bios files for that board and see if it makes a huge improvement or if its still a pile!
There are some audio issues with my wireless mic in the second half of this video. Sorry 'bout that.
sounds as if its desyncing or the cable is coming loose on the mic pack - Your Friendly audio engineer
Thanks for doing all this work, I love strange build's like this.
thought you edited the swearing hahaha
I would be interested to see this cpu benchmarked with a real server board. For whatever that would be worth. Lol.
@@soniclab-cnc send it to the long hair guys who pours liquid nitrogen on everything.
I love the philosophy of "I'm so angry i'm going to......enjoy a nice aged whiskey"
I was disapointed you filled the glass and then _didn't_ drink straight for the bottle.
I was frustrated, but I'm not a monster.
@@CraftComputing Would've been the better gag though.
@@thekrautist but damn, that's an expensive gag :)
"let's get this thing together......."
2 hours later:
Edit: 4 days later
"Let's throw this thing in the trash"
"But first, what am I drinking?"
Whiskey. Needs whiskey.
I like this cpu-motherboard combo. I will buy one for my mother-in-law.
As soon as Jeff pulled out the Jameson... I knew it was gonna be a train wreck lol.
This pretty much mirrored my experiences with a 6380 on this board. There were a couple things you missed out on too (If you did touch on them, I'm sorry!).
- The ram is not dual channel at all. Even with 2 sticks it remains in single channel, additionally the reason most software cannot identify the ram timings is because of the assigned node. Both sticks are handled by only one of the four IMC's (two IMC's per node) and software generally tries to read it from the unused node, failing to read it spectacularly. Reported it to Martin to correct this in HWinfo a while back.
- The board maxes out at 1600mhz as you've noticed, I found no way to override that. Handles ECC/Reg ram just fine however. Unfortunately the RAM latency tanks because half the CPU needs to cross nodes to access it. Was reading upwards of 140ns latency and higher compared to a consumer AM3+ board.
- FPS takes a 50% loss in comparison to the Opteron 3380 in my other rig at the same clocks during testing under OpenGL and DX11. I never could figure out why however but doom did run just fine under vulkan and 'close' but still under the 3380 FPS results. Tested with a GTX Titan X between both setups during comparison.
- Usb3... Usb3... It's not real, it's very much fake. There is no USB3 controller on the board and contrary to the claims of USB3 support (the blue header) it is simply a recoloured USB2 port with a higher power capacity.
- Using CorePrio by Bitsum helps a bit when it comes to thread shuffling to increase performance but it only works so much. The G34 Opteron suffers from similar issues with it like the earlier days of Ryzen Threadripper.
That Opteron is straight fire! All heat, no go.
WHEN THE WALLS FELL
Picard, when the bass dropped.
hi timmy joe im a big fan bro
im not sure if the cpu performance was its own falt, i say find a better board and revisit
Yeah if going piledriver then the fx 83xx or 63xx are the only real choice the 8350 scores almost the same in cinibench with its 8 cores(?) cos of higher clocks and they do run the tested games no isues 2060 would perhaps be bottlenecking lol but fps would be playable both doom and csgo get over a 100 fps with 8350 and a furry nano know this as I played both these games on with 8350 and r9 nano
agree. i have a opteron 6376 and a supermicro board and i don't have those jams and hard-locks....and definitely not in counterstrike!
@@rnssr71 would you recommend the supermicro and opteron for video editing? Streaming? Photo editing?
the name scheme is intersting cause it uses the same naming philosophy of AMD's ryzen chipsets. make it one number better so it seems better.
Hey Everyone! Before we shart on Socket G34 and everything else related to the Piledriver and Bulldozer server platform I want to mention a few things!
I've dealt with these chips for a while now, more specifically in Dell servers and I can tell you straight off the bat, Jeff's x89 is ruining everything.
These chips, I will not lie aren't great at single thread ESPECIALLY NOT GAMING!!!! (making that very clear, these aren't like older, bang for buck Xeons, who are good at gaming).
But, I can tell you first hand, on Dell, Supermicro and HP motherboards, these things are meant for VM's and lots of them. Running on a proper motherboard, these things can multitask and multi thread like nothing else. The 6300 series (Abu Dhabi) (6380, 6378 and 6366 HE especially are compute champs for their value (6386 SE is a power hog) (Two 6386's can about touch one first gen EPYC)
I would not recommend buying into this platform, if you have to buy a motherboard with it. Good motherboards for these chips are very very expensive, as well as coolers and are hard to find..
I think the only case you should be buying to this platform, is if you've found a good deal on a server chassis (which very much exist for this platform.) Servers like the Dell R715 (better R710 in my opinion) R815 and others in the Dell 11th gen "Rxx5" family are all, AMAZING servers and run this chips very much as intended, unlike this "X89" junky motherboard. The HP DLxx5 servers are good as well! But in my experience the Dell servers and Supermicro motherboards are amazing. My Dell R715, loaded with 15K SAS hard drives, dual Opteron 6380's uses about 200 watts for what I do
What I'm trying to convey is that, if you are looking into getting into homelabbery and you need cores, on a budget and single thread isn't a HUGE deal, don't be scared off by this video, as these chips are really good, in certain use cases! This video is NOT a accurate representation of Socket G34 as a whole! Now, what I would not recommend is really going for anything but the 16 core Abu Dhabi chips, as really at that point, just buy an Ivy Bridge chip or something else, you're better off with Intel. I know these chips have BAD single thread, but they aren't as bad as the ones in this video. My R715 actually used to host a BungeeCord network (Minecraft) and performed pretty admirably with upwards of 150 people spread off across my network on 12 Abu Dhabi cores! (Mind you, you won't be getting that on a single server... Single core still isn't great). They've run Plex transcoding beautifully and Code Compile is wonderful on one of my 24 core VM's. It is a pretty snazzy render box and i've been happy.
The chips are cheap (80 CDN for 32 cores {2 Opteron 6380's off ebay }) but motherboards and coolers are very very costly and not worth your time. This platform is only worthwhile if you have a preexisting high quality server chassis, and power is cheapish ( At 0.21$ cdn per Kilowatt, my main R715 at home costs about 18$ cdn a month...) these are really fun to play around with, if you need lots of cores for VM's, Code Compile or a multicore render farm. That is the only reason i'd buy these chips.
I really hopes this helps at least someone! This platform isn't complete garbo, but special and has its niche and this video is NOT REPRESENTATIVE OF REAL WORLD SERVER PERFORMANCE AT ALL! (if you have a good motherboard ;) )
Great Video as always though Jeff! I'd love to see you try again with a proper motherboard :)
P.S, These things don't game..... I tried it with a 1070 and it was the funniest bottleneck i've ever seen.......
I totally agree 100%,I saw much much better performance on a Super micro dual socket.
Real G34 mobos aren't that expensive, exluding rare quad socket ones. I'm planning to get a dual one for new diy server.
A few weeks late, but thank you for this. I'll be honest I bought one off amazon as they were selling for less then 10$ just to frame it and use as wall art.
It would be great to see the performance of this processor if it and all the other components seen in the video were plugged into a motherboard from a quality brand, such as the Supermicro boards that can be picked up second hand on Ebay.
The funny thing is: I have seen that exact processor tested on an actual proper motherboard(not tossed together garbage) and it handled doom just fine. Yes, Piledriver was a pile, but if you come across some of these in a junk store somewhere it might be an ok system;assuming the motherboard you find isn't equally shit.
Personally couldnt find another board
I've ran a FX 6300 from, 2014 up until January. It couldn't handle DooM 2016 too well, but Doom Eternal 1.0 (and only the 1.0 release day, I don't know why, I bought the game but always ran the pirated version cause it gave me much better FPS) ran fine, it wasn't pretty, but I even managed to get ultra nightmare done.
Nowadays i5 11400f + 2060 12gb (it was cheaper than 3050 8gb).
Wait. I didn't expect it to be properly fast, but it barely POSTs, seems like. Why would it lock up Doom for example?
Then again 900cb in R15 seems kinda alright. I mean no not that it is alright for the class or the power budget, but the FX-8350 does like what, 660cb, and you get double the cores here at the price of a clock disadvantage, so 900 sounds like about fair. Bulldozer/Piledriver had a bit of an FPU deficit in workloads with high SSE/AVX utilisation, by design.
I wonder if it's similarly bad in Linux or whether it does any bit better there.
I got around 1700 with 2 of them, and around 3700 with 2 of them on R20. I agree that it's not that poor.
I think motherboard maker can't write BIOS or configure all the component good enough to make this work correctly.
There are modded bios for these motherboards. Phils Computer lab explored this, and found it fixed many issues.
I'll give him a ring and see if he has a link. Thanks.
@@CraftComputing how did BIOS this end up going? I uhh… i just had to know what they were like so i brought one 😂
You can try the new x99 motherboards and e5 2620v3 with some cheap ddr4 ram and a bios hack it's pretty good value
Tech yes city made a video about
(I believe that you've already seen it)
I have one with the E5-2678 v3. It's been pretty good so far. I just have 32GB of RAM, and a PALiT GTX 980, along with a noname 1TB NVMe, and a Toshiba disk. (All my data is on a remote share, so I don't really care about storage).
Since most of ur problems are in the BIOS....
.... what if you flashed with a BIOS with a BIOS from the original motherboard that the chipset came from?
Thank you for testing this combo, I was looking for this for some time. You really saves me
Hey Jeff, a professional audio engineer here with a couple suggestions. First you should high pass your microphone up to around 150hz and possibly notch out a bit around 200, it will considerably clean your vocal up and get rid of that muddyness in the lows/low-mids. Also regarding the audio dropouts, not sure what that is but seems like maybe a buffer issue with your driver. You also really gotta get some treatment on the walls as well to kill all those reflections being picked up in your Lav. Cheers!
1. 3.2 ghz is really low clock for Pilrdriver, as they scale up to 4.6-5ghz. 6836's clocks are comparable mobile APU-s
2. i5-3317u boosts up to 2.6 ghz in ST loads.
3. 6386 SE is not 16 core CPU, it is 8 module CPU. Bulldizer line in particular can't be compered core to core against other architectures.
Re: gaming - Was the IOMMU disabled? Leaving it enabled can have pervasive, unpredictable stability effects when you push around lots of data on the PCIe bus.
Also wondering how badly the dual channel memory configuration is holding back performance on a line of CPUs intended for quad channel operation and famously ravenous for memory bandwidth.
Well , my e5 2640 runs on dual channel and has performance almost on par with ones on quad channel , this may be thing that i got good bin or good 1333mhz 2x4gb sticks.
"how badly the dual channel memory configuration is holding back performance"
very... specially since while it have 2 sockets it runs single channel only so it's not even dual-channel.
Børre Børresen That’s what I was afraid of - each NUMA node only has access to a single channel, which will starve performance. Anybody committed to playing with one of these old Opterons really needs to snag a used socket G34 motherboard from a name brand.
@@Halon1234, i think not only do they run single channel, they also are both in the same NUMA node.
yeah... this board is really a big pile of garbage and the price is so high that you could add 10$ and get a second hand dual socket Supermicro, Tyan or Asus board
(and most of the Asus boards will give you a tiny bit ability to overclock, or quite a lot if you get hold of some ES versions of the Opterons... just beware of cooling)
@@brrebrresen1367 Are you sure that it run only in Single channel just from one NUMA node not from both? If it's true, this will got performance hit from both Single channel and the same hit that 2990WX got. I don't want to imagine how bad it is.
P.S. You must pay at least 40-50$ more than this board to get something like SuperMicro H8SGL on used market, at least in international platform.
A low clock server CPU doesn't do well with your desktop workloads? shocker.
These CPUs were intended to be used in applicants like ESX, where you need a lot of cores to support VMs...
I had one, and it lived for two weeks before it started rebooting every few minutes, I tore the heatsinks off of the chips to find OK written on top of the northbridge chip not new..Salvage??? I have pics.
i asked about this MONTHS ago and finally we get to see the result, wooooo!
When I saw this notification my first thought was "WTF is X89?!".
At the end, I just start thinking, once again, how much AMD has improved in the CPU department. It is really impressive.
6386SE for just 55$!? Over a year ago I bought 2 of them for my working PC and they cost me 300$ included shipping.
I think all because of the garbage motherboard that can't work correctly (when I saw BIOS screen, I can't expect to be able to configure anything that should help solve the problem). I have never encounter such problem showed in this video. I can play games or edit video just fine and it's working right from the start without doing anything.
However, I used SuperMicro dual-CPU motherboard (which is god damn expensive for such a platform so I can't recommend, Asus one is OK), so it's fairly stable that I haven't encounter any system crash or any real issue until now. The only problem I have is when first plugged-in and power-up the build for the first time, video card always run at x8, but it will be x16 if restart the PC or power up again without unplug it.
Nevertheless, I still can't recommend anyone to build PC out of G34 platform. Even with 2 of this CPU and full Quad channel memory for each, they can just barely outperform Ryzen 7 1800X. Also, it's not good for gaming with lower clock speed than FX-8350. And if you use lower model than this such as Opteron 6380, despite 32 cores and threads, can't process footage from HD capture card correctly because it have too low clock speed. (When you open OBS/XSplit, if you are not recording or streaming the preview footage on OBS/XSplit will be freeze then move and freeze again for around half a minute before the footage completely freeze, but if it's during recording or streaming it work normally but it will be the same as soon as you stop recording. I prove that it happened because of CPU's clock speed when I tried the capture card on my FX-8350 build and the same problem appear only when I underclock it to 2.40GHz.) In addition, those Chinese boards are really garbage as showed in the Video, and even outside of those, all branded motherboard are meant for server-use so they have very limited I/O (2 rear USB ports, dual LAN, one serial and VGA port, and nothing else). You have to spend additional money on Sound Card and USB host controller card to be as usable as normal PC. All of these still not consider how hard to find the heatsink is.
In other word, anyone must stay away from this platform, it will not value your money. I bought it just because I (irrationally) want to own something not many people owned.
Glad I found this review before buying 2 of this board. I has 2x 6276 in a Supermicro H8DGi until the board started randomly rebooting after 45-60 minutes of use. No BSOD, just as if someone pushed the reset button.
Never had performance issues. Was looking into what to put the CPUs into and it definitely wont be what you tried. Maybe an Asus KGPE- D16 if the price is right (and if the price is right for a case as well).
oh wow i feel like i can smell that jameson from over here... thats some good stuff..
The CPU was actually pretty decent back in its days for heavy multi-threaded workloads and running VMs, but not with a rubbish motherboard like this. The main problem with these "fake/scam" boards is not the BIOS, but power delivery, as they simply cannot deliver enough power at a high enough consistency or efficiency to the CPU and memory, and this might be partly the reason that Sleep mode never works on any of these boards (fake X58/79/99 included).
Those fake X79s are actually fine, since they've been making them for years now, and there are BIOS hacks from Chinese and Russian websites. But this is nothing but a scam, baiting people with the high core count of cheap Opteron CPUs.
This doesn't surprise me. My old Vishera used to score 800 cb screaming its lungs out at 5.2ghz and consuming over 230w. If you account for the extra cores, yet huge losses of clock speed the 900 is about right. Those were actually big numbers back then. They put I5s and indeed I7s to shame, but weren't good for gaming or other tasks that requiired IPC. I'm glad you did this though, and that you have one of these up and running as there was *never* a cheap board for these so hardly any one has them. I see you like the same sort of things I do, re-purposed server gear so at least you have this for historical purposes. I still loved the video. Maybe next time go into it with less expectation. Thanks for another one that I found much more interesting than most of the tosh on here.
FX 8320 Vishera, so 8 "cores"
The consumer FX-8350s were really handicapped. If you run Opteron 63xx ES chips (piledriver cores) with 16 cores, you can downcore alternating cores and get Piledriver cores that are not sharing L2, decoder, L1 inst cache and other parts of the pipeline with the other core of the compute unit, netting a 15% boost in IPC.
Loved that build montage! A nice change from dance or metal music backed montage's!
I see where you are coming from, but I got the same board with no problems. Found an Opteron 6328 server in the trash, and bought this board a year ago. It is what it is and games well with 16gb of ram and 1050ti. You might have gotten a bad board.
The best combo in this video is Jeff and the booze, because hardware failed completely.
TBH I wonder if the CPU or chipset weren't overheating with that little coolers. If I recall correctly older AMD platforms had some weird temperature reporting, where if you were starting to get close to 70C that meant your CPU is actually close to the melting point. I would like to see if swapping cooler for something bigger and checking chipset cooler mounting along with VRM heatsinks would net some better results, but just out of curiosity, because the platform itself is based on a failed architecture.
The E7 xeons on the intel 1567 socket from the same time period totally thrashed the opterons with lower clock speeds and less cores.
There is a newer version of this board. Can you make a review on it please with an 6386se processor.
I messed around with one of these X89 boards - got a little bit ahead with the BIOS-MODS forums for it (was Jingsha branded) - we unlocked voltage settings and RAM speed settings along with Hypertransport settings and a few other things.
He was working on getting CPU overclocking unlocked, but I got impatient and force-flashed the wrong BIOS to my board. Only way to fix it is to get a chip programmer and reflash, but that project has been on hold forever.
Sooooo if you want a project, check out the bios mods forums and search for Jingsha, carry on where I left off, get some overclocking going on that Opteron and let’s see what 4.5GHz of Piledriver ridiculous with only half the quad memory channels enabled can ACTUALLY do!
On certain Supermicro boards you can get a custom bios for overclocking. I've overclocked my quad socket 6378 rig up to 3.0 ghz on all cores. Couldn't keep them cool for obvious reasons. I've wondered how far a single socket could go.
Thank you Craft Computing for another excellent video
Bought an x79 MB with nvme, E5-2650 v2, 32GB ECC DRR3 1600Mhz direct from AliExpress for £138 delivered. It's been running flawlessly for 24/7 3 months.
Most of these Opteron's on x89 have a TDP of 115. It wouldn't surprise me if they ran better on this economical motherboard. Not all motheboards/server boards seem to support the 140 tdp SEs. But I don't know. The first case you posted came with a power supply that was rated for 380 watt if I read that correctly which might be a bit low. 140 tdp for the processor + 160 tdp for the gpu alone.. Not sure what power supply you used for the last build. If you have no interest in the economical x89 motherboard,cpu fan, and ram, I have a interest in the platform if simply for the peculiarity and uniqueness of the platform. Thank you for the video of the build. It was entertaining to watch.
HT and NB oc options in bios? FX CPUs need them cranked to get performance.
I have one of those generation Opteron CPUs in my junk drawer. Looks like it is staying there. Makes a cute display piece, though.
Had a couple of 6128 Opterons in my storage server. Extreme power consumption, extreme issues with anything and everything.
Hi,Jeff !!
Glad to see a new video !!!!
Hey if you still have the system gave you tried turning off virtualization in bios?
I'm curious what the performance would be like on a server board from a reputable brand. Too bad those boards are really more expensive than they should be on ebay.
Could the behaviour in games be the cou turbo-ing up in the start and then clocking down and going from bad to garbage?
It wasn't doing that in Cinebench. Holding 3.2GHz on 8-cores, 2.8GHz on the rest. I'd test it... but you know... it crashes before I get to look.
@@CraftComputing Will look more into this
These old opterons seemed way more interesting to me than they actually should be.
Thanks for the great video sir.
Would it be possible to do some power usage measurements? This would be a pretty nice alternative to a homelab intel xeon servers.
X89 is named by the manufacturer for suiting new AMD ryzen boards，like b250/b350 and x299/x399
could you please check out the huananzhi dual x79-16D ?
I usually check cpu-z to check my ram speed if the bios doesn't show it.
I've seen others review this board and it's always the same thing, smack in a high-powered server-CPU in that socket, get some of that ECC-RAM, run Windows [Insert edition of choice], stuff in a modern Nvidia GPU, then complain that a server platform does not behave as a desktop board, no overclocking options etc...
I like Craft Computing as much as I (and they) like beer, which is A LOT! But this review did not pack the punch and deep-diving that I hoped for. Modded BIOS:es should even exist for those Chinese boards.
I happen to own the very same model of Jingsha motherboard that is the subject of this video.
-Yes, there are others from more reputable brands like SuperMicro or even ASUS boards for the G34 platform, though the ASUS ones are very rare!
Most of the times they are multi-socket boards, meaning you can use several of these CPU's in one motherboard for multi-parallell workloads. -Not suitable for desktop OS... Just sayin'...
This board could be found for about $50-$55 (including the cooler) about 5-6 months ago, now the prices have gone up to almost the double from what I see on places like Ebay.
And while this particular board has gone through some REALLY strange and even kind of incomprehensible design choices (-like 2 RAM-slots on a 4-channel memory server platform??) the G34-platform really isn't a gaming or desktop platform what-so-ever, for what it's worth, it actually does not perform that bad under Linux for the workloads that it is supposed to run: Spoiler alert! -Not games and desktop applications.
But yes, you can run desktop applications on it under Linux, only difference is they actually work better than on Windows.
Of course it cannot measure itself against the way more recent, more effective Zen-platforms. -Even though Zen and the older Bulldozer architecture are practically ugly cousins, favoring the focus on more cores and interlinking them.
Zen only does this much better as it a lot more recent platform and a re-design to counter the shortcomings of the previous generation. Thus allowing for higher speed, better power-efficiency, higher IPC, granted more headroom and above all: Built on 12nm(and also 7nm w. chiplets) compared of the giant lithography for the Bulldozer at 28-45nm(!) for the oldest parts.
Granted the G34 platform cannot measure itself with Intel CPU's from the same era but seeing all tests being performed by TH-camrs on Windows, doing desktop tasks and gaming is almost like them finding an already dead, rotting horse in a ditch and going through all the pain of dragging the carcass back onto the road again, only to then kick it back down into the ditch! :D
It's almost like a "Just because you can, doesn't mean you should!" scenario with this platform.
The easy explanation; Linux handles cores and threads (and modules, in this case) a lot different (better) than Windows has ever done.
Linux rules the world in the server-space and is has ruled it for ages, this is a server CPU, it is meant to do "server:y" tasks, parallell stuff, databases, web requests, things that desktop Windows is not good for, as Opteron CPU's was never intended for that desktop market.
Remember the AMD FX-platform that also did not compete very well with Intel's stuff? -Yeah, that was the CPU's intended for desktop tasks. They excel over the Opterons in every way when it comes to desktop tasks because of their higher clock-speeds and aim towards other kinds of workloads.
This is a server CPU and a pure server platform, in server scenarios it works quite well, so if you want to build on an old server platform that does not have as many *known* flaws (as of current date) -compared to the Intel platform of the same era (-and even later platforms) good parts are often very cheap. We're talking server-grade parts, not consumer parts.
Combine this motherboard with low-powered Opteron CPU (=not the CPU used in this video) e.g. the 6366HE: 16core@85W or a 6262HE: 16core@85W or even a 6128HE: 8cores@65W) an it will work pretty nicely.
Again: Please use it as was intended, as a >server< with Linux, that is...
Run Docker containers on it, run a file-server(sftp), use it for Syncthing, Webserver, a streaming server, as only a few examples.
Multitasking and web-requests is what this platform does well, so let it do ONLY that! :D
Thanks for your feedback, but here's the deal as I mentioned at the end of the video. ANY task, server, workstation, or gaming, will perform better on a Ryzen 1600 for less money overall. You'll get more memory bandwidth overall running 2666 dual channel vs 1333 quad channel, and around 30% CPU time improvement. Even for server workloads, the Ryzen will perform better. That's why this received a negative review...let alone the freezes, stutters and hard locks experienced on the system.
@@CraftComputing Agreed, this old server platform cannot measure itself against the newer Ryzen 1600, of that I'm certain, and that is true for most of the single-socket server hardware released in that era. ;)
But it is actually does not perform nearly as bad when it runs tasks on Linux, when compared hardware released in that era, that is...
Just a question: The cooler that you got with that motherboard, did it have any rating written on it?
I got mine in a plain no-name box and it seems to be rated for 115W TDP which most Opteron CPU's was limited to at the time, your CPU in this clip (I think it is a 6386) has a rating of 140W TDP, could that be the reason for the constant hangs you experienced?
Given this this system is indeed a server system so it usually throttles at around 50-60C to run stable for server operations.
Props for the VERY fast response!
Love your channel!
"Just because you can, doesn't mean you should!" scenario with this platform. --- WELL SAID. The way these CPUs are described on the TUBE is just confounding and makes little sense... You have brought reason to my madness. "Use as it was originally intended". "Don't assume". "Don't test with a multitude of unknowns". Monday morning (decade!) quarterbacking is always EZ. Thank you for you input :O)
Where did you get your RTX shirt???
9:15 8400 i5 6c\12t? Wut? I cant find any of 12 threaded i5 8400 on intel site...
Typo
not sure why your videos stopped showing up in my feed. I missed watching them and had to come find your channel again
Thanks for keeping an eye out for me!
What about some Linux to see if can perform a bit tiny better? Or maybe if you can get memory speeds reading? Games under Proton?
The opteron, while an awful CPU, shouldn't have been hard locking and doing that nasty shit. That motherboard is definitely a pile of crap.
PhilsComputerLab did some opteron content a while ago, and while it sucked utter balls, it didn't do that janky nonsense.
Does the ACPI/PnP BIOS support "Enable 4G decoding" ?
The function "PC Power ON After AC fail" doesn t work!
Are you sure the GPU was fine? I've gotten similar GPU issues where it hangs for a short period before recovering too. Also, have you tried CPU-Z for memory stuff?
What cpu cooler did you use? Opteron coolers are so f*ing hard to find...
I've run Doom 2016 with a 2.4Ghz Celeron before. And it never stuttered. Granted it was paired with 1070, but it never had issues, just slow load screens.
The i5 8400 runs around 3.8 GHz all core turbo in cinebench.
A good fix for the fan would be a physical fan control like the ones ZALMAN used back in the day.
Thaiphoon Burner is a great program for reading ram and all of is IC's and speed data.
Some 3rd party programs can OC some of those older chips in the OS. Could be worth a shot to try some of these suggestions.
Good luck with the X89 build.
Cool video though.
I'd like to see the benchmarks on a proper motherboard with the same chip.
That build effect filter ruined this video for me, I don't know why but it made me feel little sick.
But Jeff, is the poor performance due to the Motherboard or is that CPU just garbage (does it match with results for that same CPU on other boards)?
I would say the motherboard (and partly software setup). I have a dual 6344 (lower clocked 12 core) and it scores just as much single threaded than this faster 6386 in cinebench. And well games, without quad channel, pretty low IPC and single core performance, and the situation worsened due to 2 numa domains I am not surprised things don't go that well. My machine is an absolute beast though for what I built it for, which is compiling big software projects.
That type of AMD coolers are actually super common -- ALL of the Opterons used the same one. Thing is there is no market for them so companies generally just toss them.
I have something like 250-300 of these just laying around, highend low profile coolers, 2 heatpipe mostly solid copper with a aluminium mounting plate. Has no problem dissipating heat from older 6cores at the very least.
Love your channel!
Thanks!
I promise these CPUs are a LOT faster from linux
is there any overclocking options on this board?
That is... actually true, performance numbers nearly triples on Linux-OS compared to Windows-OS, judging from from my own tests of Solus 4.1 vs. Windows Server 2019 and Windows 10 1909. No joke.
-Though triple numbers of that magnitude is not impressive in today's computing, Windows just makes this platform look even worse than it actually was! :D
Also: Windows is trash when it comes to multithreading, that has been known for a long time, and as has been showing very clearly in the likes of the AMD Threadripper platform, just juggling threads around in Windows causing 100% CPU usage for tasks that does not even utilize 50-70% under Linux. It's getting better, but Windows is still far behind on that...
It hardlocks because vrms/chipset overheats try putting fan directly blowing on vrms/chipset that solved it for me and my opteron
Love the TNG shirt!
Thanks for this insane review. Great content!
The single thread benchmark... Just bit higher than core 2 quad q8400
Can you bench RandomX on that system?
try the Supermicro H8SGL-F with the same cpu and for sure the results should be much better
I wonder if it was just the motherboard.
Good vid. Any chance of finding a better board ? Keep up the good work
that board is power starving the cores thats the issue once it hits turbo its getting a shit load of vdroop
I was about to ask in here about the audio skips but then I've read the pinned comment.
I totally understand the sentiment, of a anger being [somewhat] quelled, with a slow sip. Comedic effect could've been progressively draining your Speyburn, leaving the filled glass unattended.
Now in seeing an abysmal blind date, of a CPU to that sorry excuse of a [piece mealed] motherboard. Would you consider testing that CPU, on an established motherboard? [say a Supermicro, ASUS, etc.]
Dual channel (if that?!) memory is gonna cause at least a little bit of pain.
I really liked this video. I was thinking about an Opteron for my Plex server, just for the sake of it (and because the current AMD server processors are way out of my reach. And because I though a 6378 can manage the transcoding in all devices at home). I don't think the CPU is just crap (but it is definitely bad), I think the motherboard plays a major role there.
The thing is, by the time you try to buy a G34 compatible motherboard, you've already spent Ryzen money. That's how my "stupid" home server project with an Opteron died and got replaced by a Ryzen.
How Good/Bad would be a real G34 Mainboard with real Bios would be?
Still terrible.
@@CraftComputing Yes. But THAT terrible?
And nowadays AMD has EPYC CPUs for server applications, they have made a pretty impressive job since the beginning of their EPYC and RYZEN lineups...
It would have all been fine had you installed the rear fan.
Right, because THAT was the limiting factor in performance... Nevermind that this is a cooler blowing down on the VRM with an open top, and CPU temps never exceeded the low 60s.
I was just kidding, I like your videos 👍🏻
This would be fine if you used it differently. I would use it as a small workstation :3
so the only use for it is to change the 140W CPU into a grill rack...
Jeff...experimenting with questionable computer parts so you don't have to! Cheers!
Did you try cpu-z to see memory speed?
I don't think the 8400 has 6/12 ( 9:23 )
wow that things actually slower than my dual xeon from 2010.Given it's a 12c/24t and it's barely slower than my r5 2600 difference is i use her for a storage server and a virtual box.It's in a 2u rack mount so no audio and only 8mb matrox display but for what i use her for it's perfect.I was fascinated by the opteron back in the day.I built 2/3 of them in my life but it was proper server mobo's and it worked perfectly.These chines knock offs are just not good for amd.For cheap xeons etc they are great though
Has anyone tried the OCGN5 bios on these boards?
You could give some better chinese boards a try. Look at Jingsha X99 with DDR3 (for cheap memory) or DDR4 version (same board, maybe sold as KLLISRE). Paired with a 2678v3 for DDR3 support, unlocked turbo with a BIOS mod, and you have a 3.2Ghz all core 12C/24T rig.
There may have been mistakes made but at least the beer was a good decision
I've been eyeballing it for a while now.
I was really looking forward to his review on it at the end! Are the reviews not happening anymore or was it missed due to stress??
evidently a host of issues with this board, first off being opteron 6xxx is quad channel, second that cpu is CMT enabled so it should be 16/32. Yes I know about the whole "were they true 8 core modules." That cooler wouldn't cool a 95w cpu properly, let alone the 140w from this cpu
Nope 16 core, 16 threads : www.amd.com/en/products/cpu/6386-se
So, is the CPU doing so bad, because the motherboard is kind of janky, not optimized for the CPU, even though it "works"?
A glorious trainwreck. Hilarous.
But how about a Darkflash case review, Jeff? Not much coverage of their cases out there!
I will be reviewing that one ;-)
That was some abysmal performance Ha!. You should do a followup when you explore the modded bios files for that board and see if it makes a huge improvement or if its still a pile!