Gadgetory

we've had a project computer for the last few months now actually arguably closer to a year that's been our behind-the-scenes obsession as a sort of restoration project think of it like restoring an old hot rod except it's a dual socket xeon from the x79 era with a case labs case by a company that's now dead and an EVGA SRX dark motherboard which has its own tragic history after a lot of recent trips to watch on Bay in China and some calls to manufacturers we've managed to get this dual EFI 26.97 v2 system up and running and compressing our video files daily but it's far from done clearly because as you can see in our be role it's not pretty it is functional however and that means it's time to run our dual Xeon system and all of its 48 threads through our CPU test bench one important note all of this cpu test data is going to be older as we ran these benchmarks back in march when we built the thing originally and it will be wiped and refreshed for the rise in reviews so this won't carry forward but we're gonna go through the build process some of the cooler parts in this machine give a bit of history on it and talk about the performance in the latter half of video but this is really more of a project PC so let's enjoy the process of trying to get the thing to work before that this video is brought to you by the EVGA gtx 1660 XC black video card EVGA gtx 1660 XC black uses a fat heatsink to allow the card to run at lower noise levels for longer capable of sinking more heat before the fan kicks in and also has a day one at game ready drivers for game launches the 1660 XC black is short and compact for installations a small form-factor pc builds but if you need one that isn't extra thick it's also accompanied by a dual fan counterpart with a thinner heatsink for standard dual slot support learn more at the links in the description below this content combines a lot of cool things we have an appearance of the tragic EVGA SRX aboard that was purpose-built for overclocking the Intel Xeon CPUs and would have been the only dual socket board really capable of it the board has seven PCIe slots it has a big chip set and vrm heatsink and a VR hammond bios that were going to be prepared for overclocking unfortunately the intel xeon cpus of that gen ended up being lakhs for overclocking and so EVGA was stuck with an over built at dual z on board that couldn't fulfill its purpose what was meant to be EVGA highest-end board was utterly sandbagged a luxury overclocking platform for CPUs that could not be overclocked we also have the appearance of to high-end wants high-end z/os a lot of the now very affordable ECC memory appearing as well and an off-brand motherboard from Shenzhen so ignoring the massive case that it's the systems on the table because here in the massive case the system overall is pretty cheap it's pretty affordable because you can get ECC memory ddr3 which is not expensive server companies are dumping it on mass on eBay right now and that's maybe we might have paid 200 bucks for 1866 64 gigabytes of memory on there and then we also got the two CPUs we end up with three of them we bought two and then we had to ask Intel for a third one because we were troubleshooting an issue more on that later and those are in the range of 200 to $300 each so you get two of them call it let's call it 600 bucks on the high end and then you can find a motherboard and these are also a couple hundred bucks max they might be cheaper now we bought this a long time ago and then the content just didn't work out but we're here today so you can build a whole dual socket system for relatively cheap especially compared to the original price and that is where this idea started it was Patrick's idea to get into he saw this motherboard and then nothing ever worked and so we spent several months just kind of in our spare time tinkering with it and getting it to work the motherboard next to me is what started it but it's not the one that we're using currently we wanted to pull this server together as a compression system to reduce our footprint for storage requirements all the video files so it does actually run actively and compress videos now it's working properly something that two older Zeon's are actually really good at despite the fact that today realistically you could buy a ninety nine hundred K or a 2700 X and both would be far cheaper than even the used parts that the two cpu is used with the board and the memory and all that it would probably be cheaper to buy a modern CPU and use that if you're gonna do something like gaming it'd be far better for gaming frankly at a 1997 hard K 9600 K we'll talk about this later would blow away these two CPUs so that's not really the purpose but we didn't gave inventor McHenry for fun the purpose is to see how they do with production workloads we'll throw a premiere in there as well just because why not at this point in one way this content piece began more than a year ago when we picked up the intel xeon e5 26 97 v2 freeze and our production machine and we do have an old video on that as well it was probably about $500 at the time they are two to three hundred dollars now it's a twelve core 24 thread part which for the era was very good this is circa 2011 213 range depending on the version of the CPU and it was two thousand six hundred eighteen dollars new so quite a discount you knock an entire digit off of that number for the modern pricing we didn't put much more thought than that until a few months ago when we well six eight six months ago at this point when we bought the dual socket board next to me which is by a Chinese brand this is Juan Angier and unfortunately I can only speak in listen I can't I'm illiterate for Chinese characters so you can help me out in the commets with what the characters say but I can read the opinion and one on Jerry I think is just South China made maybe or South China something depending on the tone of that jerk but anyway that's the board and it's two sockets it's only four dimm slots for a CPU that is quad channel so it's kind of beats it's gonna be dual channel for this board and it's compact so that is where we started technically it's not an X 79 board it's a C 600 series chipset because dual socket Zeon's mean C 600 X 79 is for enthusiasts single socket only chips like the 49 60 X but X 79 is the name that consumers are familiar with hence the branding we were interested because cheating the system by buying cheap old decommissioned Zeon's is a perennial II cool idea limited only by how expensive old server motherboards are with brand-new reasonably-priced dual socket boards on the market the idea might finally be practical moreover the 26 97 B 2 that we already owned is specifically capable of being run and dual CPU systems so with a board and one CPU already in our session we were already two-thirds of the way there for the bill or so we thought the Quan Andrew board arrived after a couple months then we purchased a second 26 97 v2 off of ebay and assembled the system it didn't boot everything worked with either CPU in the primary socket but nothing can persuade the system to boot with both sockets filled there's an LED display with post codes but since it's a Chinese board with no manual that was presumably assembled from repurposed server hardware and most of the instructions were in Chinese which again can't read the characters we didn't know where to look up the post code any code tables that we did manage to find would list that code as quote reserved for future use ominous normally we have more than enough hardware to replace one part at a time and figure out the problem but we only had two of the LGA 2011 Zeon's and one dual socket board to tasked with and a whole lot of question marks our original CPU the one we used for editing was an Intel engineering sample that we bought on eBay you aren't supposed to sell engineering samples but someone did normally that'd be fine but the S spec of the consumer and sample CPS are different and that could be causing some kind of problem in a system that's looking for two identical processors the solution here would be to buy yet another s or well en aspect SR 19h the Han answer board might not be compatible with Ivy bridge-e processors at all was another one of our thoughts something that we were able to validate later compatibility lists for the board we bought we're limited to whatever the reseller happened to test so even though the socket matches in the c6 no two chips that can't support the CPU we thought it might still need a BIOS update to do so again there was no place we could download a manual or BIOS update just a CD of drivers sent with the board the solution here would be to buy a new motherboard that explicitly supports dual Xeon 26 97 v 2zv is since the motherboard this one was the sketchiest the link in the chain the board we bought off Aliexpress with no real BIOS updates available online we ended up deciding to replace that one we got a new board first and we could have gone straight to server hardware but we wanted something that could be argued as a more gaming targeted and so we picked up a board with a lot more PCIe slots we also wanted some that might have some overclocking potential if we could break the Zeon lock and also something that looked a bit better than the WoW orange-and-black of this one so in pursuit of this we turned up one promising result the EVGA SRX motherboard the SRX was the follow up to EB J's popular SR 2 SR stands for super record and the older dual socket LGA 1366 board was a success amount overclockers especially EVGA resident overclocker kingpin we saw one in our recent tour of his lab something he ran with seven LNG pots in the day at a minimum sometimes more for the RAM and there was one for the chipset two for the CPUs four for the GPUs and a lot to manage it was the most complicated thing that kingpin has ever managed for overclocking he told us and although he seemed to look back on it fondly he also told us that he didn't miss the complexity EVGA even put together a video called quote fastest system on planet earth that looks more like it's from the 90s than from 2010 but it does give perspective of how old the sr2 is these days the SRX though came after that despite the SRX being locked now all these years later the SRX fit our needs perfectly and the most recent BIOS revision specifically lists support for our CPUs we contacted Jakob at EVGA and asked if he remembered the SRX which was met with a sigh and the drawn-out yeah for an idea of why that was the response check out this 2012 overclock net thread where he was trying to gently manage the expectations of excited overclockers before the launch they had one board left and it was an old RMA that had been sitting in the warehouse he updated one of the three BIOS is on the board to the one that we needed and sent it over to us luckily he didn't overwrite the BIOS that someone either the RMA customer or an EVGA attack had modded with a bad luck Brian macro top text successfully makes BIOS skin bottom text EVGA SRX that's a pretty good summary of the reception this board got if a bit depressing we connected everything up as before and had the same postcode again reserved for future years so again a couple potential culprits the same issue could have been true before with the engineering sample CPU or this was a dud board and that seems I'm likely given that it was failing exactly the same way as the other board it boots fine with one CPU alone and it breaks with two but still there must have been a reason it got are made so it was a possibility and we really wanted to make this work so we finally broke down and contacted Intel and asked if there could be a problem with running the engineering sample with a consumer sample CPU and the answer was maybe so after some back-and-forth with Intel having to find someone still who worked internally who could also remember these CPUs we did end up just getting a second sample from Intel that they happened to have lying around contrary to popular belief Intel does not have stacks of CPUs on their desks to send out so someone actually went and found this thing for us and it's an SSR 19h CPU it's a consumer CPU so that would resolve what we thought was our last variable we really wanted to make this work even though the 12 core Xeon is worse than cheaper Intel CPUs that are made today like the 9900 K the point is that we wanted to use two of them in a cool motherboard with the case from a dead company and build an interesting computer that's all we wanted when we started this we reached out to EVGA again and we're told that our post code related to memory memory on the SRX is tricky there are 12 slots for per CPU plus 4 extra on the primary CPU and the filled slots are supposed to be mirrored between the two CPUs which was complicated by the fact that we didn't have eight identical sticks of memory lying around we began googling solutions again and hit this forum thread that we can put on the screen of the EVGA forum from only a few months ago against all odds' two other people on the planet we're having exactly the same problem and hello to those of you who were and they solved it one of the accounts is seven years old and it's only made two posts both in this thread from a few months ago apparently even though these boards were fully intended to be compatible with non-ecc memory and even though one CPU will boot with non-ecc memory there is an issue that keeps pairs of a 526 97 v 2 CPUs or possibly just Ivy bridge-ep CPUs in general from booting with non-ecc memory this is speculation but it might be because the 26 and 7v2 launched in quarter three 2013 by which time it was well apparent that supporting the SRX was a waste of e BGA's time we bought two Ram kits first a cheap four stick kit have 1600 megahertz memory just to make sure the board would boot which it did and then a 64 gigabyte 8 stick 1866 megahertz kit the SRX needs at least eight sticks to enable quad channel memory or well quad channel the memory itself is is not channel dependent but quad channels on the motherboard is it needs eight sticks with the SRX so for per CPU at last after multiple months two motherboards three CPUs and herbal sticks of RAM to power supplies and at least three GPUs we had found a combination of hardware that would work let's go to the testing finally we ran our standard series of CPU tests from march on the board that it's old data keep in mind and first it was with four sticks of 1600 megahertz memory in dual channel on one board and using EVGA s default CL 10 timings for the s or X which are tighter than the sticks are rated for presumably this was intended for non-ecc memory but we tried it out anyway and then we tested with the 1866 megahertz quad configuration with CL 13 timings as listed on these sticks and then one last time with a 104 megahertz BC LK this was the best we could do for overclocking sadly so that put us at 19 40 megahertz memory BJ's table of CL 11 timings and we would normally use a pair of or while our standard 20 400 megahertz HyperX kit for this testing on old boards but it doesn't work because it's not ECC so then raising the base clock beyon stock unfortunately breaks i/o it actually breaks Ethernet and it breaks almost all the USB ports as well and also a lot of other things so BC LK overclocking on this not the best solution but we tried it anyway 104 megahertz for that let's get into the data finally and see if any of this was worth it our blender benchmarks with the GN in-house made monkeyhead render is up first giving us realistic workloads that leverage various effects within blender we haven't yet read benchmarked any Intel X 299 CPUs and so that 2990 WX has remained relatively uncontested at the top of the sharp until now that is the 26 97 V two-by-two configuration lands us at an impressive 13 point 2 min at render time with either of the two primary memory options blender really doesn't care much about memory latency and focuses entirely on core account which is what we're seeing here if we sacrifice our IO in favor of a higher B clock we end up with a twelve point six minute render time the 2990 WX runs with just a 13.5 percent render time reduction which considering the tremendous age difference between these processors and the cost of to use 26 97 s it's really not bad this also positions the 26.97 v2 by twos thirteen point two minute render time as improved over the 9900 case stock result by 36% and over the 9900 KOC 5.2 gigahertz results by 22 percent in this instance the 9900 K board and CPU would cost less than the dual xeon set up and so does well to maintain proximity with its lower overall thread count and cost still I used 26 97 B to system is looking good in these initial tests with the GN logo ray-traced render the 26.97 v2 by 2 has more difficulty keeping up with the 2990 WX now allowing the 2990 WX a massive time requirement reduction of 49 percent versus the 26 97 v2 with the 1866 megahertz memory configuration this has to do with the complexity of the scene rendered and is precisely why we use these longer render times and varied scene complexities for benchmarking this allows us to better find where processors have weak and strong points and for the 26.97 v2 it struggles in the GN logo render when using it more modern instruction sets in blender because while the 26 97 VT doesn't have all of them the 9900 K ends up at 26 minutes for the render time with the 26 97 v2 leading the modern eye 9 a family that didn't even exist when the 26 97 launched with the reduction of 31 percent given that we're running a total thread count three times higher than the 9900 K but with lower frequencies this all makes sense blender wouldn't be a bad use case for a system like this that you'd probably want to have a good lead on cheap parts or already have some of them in hand memory for these systems is low cost thankfully as we got 64 gigabytes for under $200 but it is slower ddr3 the potential upside is that it's also ECC which this processor supports and which has value add for some users for the right use case given these blender results we have a promising start for a used dual CPU hot rod system GCC code compiled with cygwin is next we have some other tests that will publish shortly with mingw but these looks at the cygwin compile tests of GCC for today's benchmark we explain this test more in our workstation benchmarking methodology update that we published recently on the channel and on gamers nexus net the twenty six ninety seven v 2x2 does well here positioning itself in a highly competitive spot with the twenty nine ninety WX modern era thirty core cpu although frequencies are much closer between these two then between the 26.97 v2 and the ninety nine hundred k the twenty six ninety seven v 2x2 ended up completing the test in four point six minutes allowing thread rippers four point three minute result a time reduction of just 6.5 percent or a 17 percent time reduction when using Core Praia the 26.97 be 2x2 results it's hard to be mad at though because it's ahead of every other modern processor on the benchmarks and that includes the 2700 at 4.2 gigahertz and the 9900 K for reference as for why AMD does so well here we'll talk about that more in our separate piece exploring compiled workload performance in our benchmarks this is a great start for the 26 97 V 2 by 2 system for a blender and GCC we clearly have some real use cases that aren't even forced like a lot of other old system builds might be chaos groups v-ray is next in this render worth of the 26 97 b2 fails to impress sadly it does reasonably well and for its time it would have been among the best systems by a longshot but age is showing as the 9900 K begins to match performance for cheaper and with better versatility and frequency bound applications like Photoshop or gaming the 26 97 v2 overclocked with no IO functional really in the back of the board takes a 0.9 8 minutes or about 59 seconds to completely render with a 5.8 gigahertz in 9900 K completing the same workload in 0.9 5 minutes the 2990 WX holds a strong lead here and it's probably time weary bench the X 299 CPUs think of it some competent the charts the 2690 7v2 is ahead of the 900k stock result but the slow frequency of intel's older processor is holding it back Adobe Premiere is another one of our in-house workloads we explain these tests in our CPU workstation methodology video we use a 1080p 60 HT 6 for RNG encode which is a show floor report comprised entirely of a roll and b-roll and we also use a 4k 60 a roll and b-roll review render you learn more about these settings in our methodology section and the previous article that was published with the video the 1080p 60 convention shot render has the 1900 K at 5.1 gigahertz leading at 3.6 minutes to render this would be further emboldened if enabling Hardware encoding with the IGP and option that from here added in the last two years because this actually does boost performance in a noteworthy fashion still even without it we see premieres love for both threads and frequency in this benchmark the 26 97 b2 does a lot better than we had expected frankly with @id results when overclocked unsustainably to 104 b CL k or a 3.7 eight-minute result under stock conditions with 1866 megahertz quad channel memory configured stepping down from quad channel to dual channel and 1600 megahertz actually did hurt performance with a render time increase of 6.6 percent things changed when we moved to the 4k 60 render and the 26 97 v2 system with both of its CPUs actually manages to pull ahead of the 99 hundred K both overclocked and stock with the 26 97 using 1860 megahertz memory in quad channel configuration on the board if we were to run a 99 ATX e through here as we did last year in that review it be the chart leader but the cost goes up significantly for such a build this shows that the 26 97 B 2 by 2 system could be reasonably used as a backup renderer for our office or other premier workloads Photoshop is next with Photoshop CC 2019 we use a mix of filters transform scales resizes and effects applied to large images through puget spent working suite photoshop very clearly has a strong bias towards high-frequency processors exhibited here in the 9900 K is a 5.8 gigahertz results right alongside the 9700 K is 5.1 gigahertz results the fact that the 9700 K can keep up clock for clock with a processor running twice the thread count helps illustrate propensity for leveraging frequency it should come as no surprise then that the 26 97 v 2 by 2 system performs truly horribly in this workload it's a letdown really as even at a 1,700 and the 80 600 K stock are outperforming it the cores and threads just don't get used here 7-zip is next and is measured in millions of instructions per second with higher being better the compression is the first test positioning the 2990 w acts and an incredible lead although again uncontested by X 299 at this time with nearly 200,000 M IPs the 26 97 V 2 is the first processor we've run through this bench to approach the 2990 W X although the AMD spreader per processor still holds a lead of about 51 percent that stated Andi does particularly well in decompression so this will change with compression the cnidaria I heard K in the decompression workload is far behind the 26 97 V 2 by 2 showing that there is potentially another use case for this configuration that you'd want to find the parts for cheap to be worth while for reference the 2700 at 4.2 gigahertz ends up around just past the halfway mark at 80 6700 myths and between the 2700 X and 90 100 K at 5.2 gigahertz with the stock 9900 K below the 2700 AK stock cpu the compression benchmark run really benefits the 26 97 V 2 which is perfect for us aside from transcoding old video files into smaller formats we can also use the 26 97 V 2 systems that compress archives test data or text files or images or anything else on the server into hopefully slightly smaller formats this allows us to recover the data if we ever need it especially those massive amounts of CSVs but it allows us to keep it small enough that the footprint is minimized with the 1866 megahertz quad channel configuration the 26 97 V 2 by 2 runs at 80 5500 mips and boost to 91 thousand when B clock is pushed to 104 on the EVGA SRX Intel I $9.99 hundred K is next closest with its 5.2 gigahertz run allowing the 26 97 V 2 by 2 is a lead of 29% thread Ripper and the twenty nine ninety WX end up toward the bottom of the charge struggling to keep up in this more latency intensive workload moving into games we'll start to see the real deficit of the 26 97 be 2 by 2 systems in the modern age and that's primarily going to be frequency shadow of the Tomb Raider at 1080p post the 26 97 be 2x2 at 90 FPS average with low as reasonably well spaced at 60fps and 50 fps 1% in 0.1 percent lows the downside though is that it's doing worse then even an r5 2600 so don't get us wrong the r5 2600 is a good CPU but it's also 154 dollars and 12 threads the 26 97 is older architecture slower frequencies and dated memory are holding the 2 CPUs back besides it's not like games play particularly well with high core count CPUs or dual socket systems to begin with as the 2990 WX high court count CPU both stock and with Corp Ryo illustrates clearly in this very benchmark GTA 5 at 1080p is next at 1080p the GTA 5 been apart positions the Intel i5 CPUs at the very bottom of the charge far far away from anything else at all they're bad really bad GTA 5 just does not like this combination of hardware to the extent that you'd be much better off buying probably an i3 at the high end but definitely an i5 or R 5 for this game like the 2600 or 86 out of K both which are also in the process of aging that said this is clearly not a build meant for gaming it's just not one that you'd necessarily want to even game on even on the side it may be better to go with a second system for that task if doing something like this or disabling one of the CPUs when gaming 1440p doesn't help things here just by pushing the load onto the GPU the 26 97 B 2 by 2 it still runs under 60 FPS average here with the r7 1700 positioned far ahead at just 75 FPS average and that's not even a great result on this chart as discussed in our CP methodology update for game civilization 6 has gotten enough updates that it has routinely wiped our data set for testing civis 6 testing uses turn times instead of FPS looking at the time requirement for AI to process a single turn note that with more AI players at longer turns would start to add up significantly so this is a useful metric to understand the 26 97 V 2 by 2 system does poorly here with a 56 second turn time requirement on the 1866 megahertz quad channel configuration there isn't even much test variants here it's just 55 seconds for all 15 of the turns completed and this means that with 5 AI players you'd be waiting nearly five full minutes for your next turn which is agonizing in a turn-based strategy game the 9900 K takes about 30 seconds per turn with the 9700 K at 5.1 gigahertz taking a similar amount of time there's a lot of value to frequency in this benchmark which is made clear by the firm division between AMD and Intel in the charge where higher frequency intel parts always take the lead the 26 97 b2 is a low frequency part though and also oddly configured for games in the 2 socket build making it overall a poor choice for this game for total war Warhammer - and the battle benchmarks the Intel a 526 97 V 2 by 2 sadly performed nearly the worst on the charts once again landing at under 60 FPS average this puts it far below the 1700 which was the previous slowest processor on the benchmark the 2990 WX does much worse to be fair at under 20 FPS average 1 stock this game really hates the 2990 WX but turning game mode on and disabling half the cores helped significantly we have more games that we've tested but we'll leave it here it's the same story repeated across the next four games with the 26 97 v2 by to build always performing among the worst on the charts this is obviously a workstation build anyway so that's ok for our purposes so that's the build for the gaming tasks obviously it's not the best for handbrake which we didn't include here for our transcoding it's really good so we just run handbrake in the background all the time it's a lot faster than the 8086 case over clocks we used to run it on because it actually leverages the CPUs and there's 24 threads on each CPU so we've got 48 we can run handbrake and compress the hell out of our videos that we we don't want taking up all of our space when it's just b-roll files and stuff and it works out very well for that actually but again you'd be better off with modern parts that are about the same price after you count all of the various things we had to buy or ask for to do this but it was still a lot of fun and we still want to do more with this so we have some ideas for a custom modifying the enclosure or at least doing a cool liquid cooling build in it and we'll talk about that more later also I was told while filming this by our Chinese teacher that the name means South China Gold or maybe South China Gold Medal depending on how you interpret it so anyway that's that's what that company is thank you for watching and let us know if you find this kind of I don't even know what it is restoration content interesting we might do more of it subscribe for more go to patreon.com/scishow sexist or stored I Kara's axis dotnet supports directly where we have the new toolkit I'll see you all next time