Support Board
Date/Time: Tue, 26 Nov 2024 19:44:06 +0000
[User Discussion] - The Ultimate PC of the year 2020 for Sierra Chart. CPU/GPU/CM/RAM/SSD
View Count: 15490
| [2019-11-19 11:32:07] |
| Dorian - Posts: 57 |
|
Instructions for building the PC and updates https://trading-order-flow.fr/workstation/ IMPORTANT: It is necessary to disable the hyper-threading of the processor in the BIOS. MB: ASUS PRO WS C422-ACE CPU: Intel Xeon W-2295 18 Cores or Intel Xeon W-2265 12 Cores. VENTIRAD: Noctua NH-U12A RAM: Kingston KSM29RS8/8MEI 8Go DDR4, 2933MHz, ECC, QVL certification, 1.2V, Registered, DIMM, 288-pin Buy x4 ! for Quad Channel 32 Go. SSD: Intel SSD 760p Series 512GB (M.2 PCIe 3.0 x4) for Windows 10 Pro for Workstations 64-bit and Intel Optane SSD DC P4801X M.2 NVMe SSD for Sierra Chart (The world's best latency and I/O performance). GPU: PNY NVIDIA Quadro P620 for NView Desktop Manager (appear in your Windows desktop “right-click” menu). PSU: Seasonic PRIME ULTRA TITANIUM 650W Ref: SSR-650TR CASE: be quiet! Dark Base 700 OS: Windows 10 Pro for Workstations 64-bit Monitor: iiyama 21.5″ LED – ProLite XUB2292HS-B1 Cable: STARTECH MDP2DPMM2M Quad Monitor Stand ≦24'': AAVARA DS400 Ref: MB: 90MB11Y0-M0EAY0 https://www.asus.com/Motherboards/Pro-WS-C422-ACE/overview/ CPU: CD8069504393000 https://www.intel.com/content/www/us/en/products/processors/xeon/w-processors/w-2295.html VENTIRAD: NH-U12A https://noctua.at/en/nh-u12a RAM: KSM29RS8/8MEI https://www.kingston.com/dataSheets/KSM29RS8_8MEI.pdf SSD SIERRA CHART: SSDPEL1K200GA01 https://www.intel.com/content/www/us/en/products/memory-storage/solid-state-drives/data-center-ssds/optane-dc-ssd-series/optane-dc-p4801x-series.html SSD WINDOWS: SSDPEKKW512G801 https://www.intel.com/content/www/us/en/products/memory-storage/solid-state-drives/consumer-ssds/7-series/ssd-760p-series.html GPU: VCQP620V2-PB https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/productspage/quadro/quadro-desktop/quadro-pascal-p620-data-sheet-a4-nv-704415-r1.pdf PSU: PRIME-TX-650 https://seasonic.com/prime-ultra-titanium#overview CASE: BGW23 https://www.bequiet.com/en/case/1203 OS: Windows 10 Pro for Workstations 64-bit https://www.microsoft.com/en-us/p/windows-10-pro-for-workstations/DG7GMGF0DW9S With love. Good trading with Sierra Chart. Date Time Of Last Edit: 2021-07-16 08:36:24
|
| [2019-11-19 14:19:01] |
| Critcendo - Posts: 11 |
|
Make sure you're using house money on this build before committing personal capital.
|
| [2019-12-06 00:09:49] |
| j4ytr4der_ - Posts: 938 |
|
I just found this post. I noticed your comment about hyperthreading. Why is it important to disable that for Sierra Chart? It's currently enabled (by default) on my BIOS. I never considered disabling it... should I? Does it help with performance on things like replay backtests?
|
| [2019-12-06 11:22:23] |
| Dorian - Posts: 57 |
|
Hyper-Threading of the CPU will create 2 transfer lines per physical cores instead of 1 transfer line. Therefore this will increase the latency. Latency is the time between when you receive information from your internet network, and when it will be calculated by the CPU and finally display on the screen. Even if the CPU calculates quickly, between the moment the information was processed on the EUREX server, the CME, or the ICE ... you still received the information 200 ms later . Performance question: Watch the performance when you export a video with an Intel Core i5 processor (of 4 physical cores), and an Intel Core i7 (of 4 physical cores with Hyper-Threading). You will see that the difference is small. It's a marketing tool. And that will increase the latency. In trading the low latency is the key. It's not enough to disable in the BIOS, you must disable devices that you do not use, to gain latency. You also need to disable all tools to save CPU energy to allow it to focus only on your programs and not in parallel on the CPU temperature. The CPU's energy-saving and security algorithms also increase latency. To go further, you can disable the CPU security settings in the BIOS, some do, others prefer security. Remember that the more you tell your CPU to use technologies such as energy saving or security will require it to focus on parallel tasks and will therefore increase latency. This video is French but watch everything you need to disable to gain latency and you will be amazed. https://youtu.be/xSjLnO5VZEY?t=407 Disable options in the BIOS that create latency. https://youtu.be/_JdSDCWmT6M?t=477 Manage devices and power to gain latency. https://youtu.be/fUDE3MErS7c?t=105 Make Windows more responsive and gain latency. https://youtu.be/eCqgJAFryLo?t=274 Optimize application priorities. https://youtu.be/hOxmD88SCYQ?t=104 Disable unnecessary devices in Windows. People are racing for fiber optics for low latency, but the PC must also be well adjusted otherwise it is useless. Date Time Of Last Edit: 2020-09-21 07:07:07
|
| [2019-12-06 17:05:00] |
| j4ytr4der_ - Posts: 938 |
|
Good info thanks, I will watch those videos. I did try disabling HT in he BIOS but saw basically no difference. I tend to run with minimum security on my home machine already but I know MSFT has a habit of sneaking things in that you don't realize are running. Thanks again.
|
| [2019-12-06 20:39:15] |
| j4ytr4der_ - Posts: 938 |
|
I started watching the first video but unfortunately I couldn't follow him at all since it's not in English (and the automatic translation is terrible). I'm trying to find equivalent information in English now, but everything is almost exclusively geared around gaming performance (lower ping, higher FPS) which is not exactly the same as what we need to run SC as fast as possible.
|
| [2019-12-06 23:20:08] |
| Dorian - Posts: 57 |
|
If it works for games, it also works for Windows applications. I'm French and I saw all these videos, he is very good in the race for latency. We can trust him. I hope you find a channel in English.
|
| [2019-12-07 00:33:40] |
| j4ytr4der_ - Posts: 938 |
|
Thanks. Generally speaking I agree that what's good for games is good for trading, particularly in he areas of data transfer and I/O speeds. But mostly gamers are focused on graphics card performance (and I don't have a super high-end gaming card) and network ping. I don't really see either of those as being a primary concern for system development and back testing, which is what I am exclusively focused on at this time. Still, I'm making every tweak I can with regards to disk access, memory and CPU speeds (short of overclocking), power, etc. Unfortunately I don't have an actual way to test the performance differences in SC, and tests like Cinebench and unigine are, IMO, focused on things that aren't the major factors for this application. Will keep looking for more info tho... I'm sure it doesn't help that I'm running on 10 year old hardware (except for the graphics card which I upgraded maybe 3-4 years ago). |
| [2019-12-09 18:28:32] |
| j4ytr4der_ - Posts: 938 |
|
I've been thinking of starting a separate thread but this thread already seems like the topic I want to discuss, just not quite the same angle. What would be the fastest possible build, for a machine designed expressly for system development, back testing, and replay? It seems to me that this is not the same as a machine designed for live trading, since it would not need to have ultra-low network latency, for example. What factors are involved, specifically in back testing and replay? I understand SC is not really able to backtest over multiple CPUs in parallel, so as counter-intuitive as it is, a heavy-duty multiple processor system doesn't seem like it would make the biggest difference. What would? Obviously I'm thinking of things like disk access (so use SSDs) for sure. RAM also yes, but also SC doesn't seem to make heavy use of RAM so I'm not sure how much that helps. Would it help to load SC onto a RAM drive? My understanding has always been that RAM access will always be significantly faster than SSD access, it's just very volatile for the obvious reasons. I'd really like to get a discussion going on how to build the fastest possible backtest/replay machine. Anyone got any thoughts? |
| [2019-12-09 20:17:33] |
| j4ytr4der_ - Posts: 938 |
|
OK just to add to this, I decided to test a RAM disk and running SC from there with all data etc. being stored on it. I made a 3GB RAM disk using AMD's free Radeon RAM Disk software, and replay backtests altering 3 parameters per test, are running at essentially identical speeds as running from the SSD (I'm using Sierra Chart Back Tester and it reports the # of backtests/hour). Too bad, I was hopeful that there would be a meaningful bump in speed but it looks like at best it's maybe a 2% difference. EDIT: Interestingly, after running for a number of hours on the RAM disk, it is now cranking along at around a 6% improvement vs. using the SSD. That's starting to get significant. Date Time Of Last Edit: 2019-12-10 02:05:45
|
| [2019-12-10 23:13:31] |
| Dorian - Posts: 57 |
|
The graphics card is not used for Sierra Chart. Intel Xeon processors do not have an internal GPU, so professional motherboards do not have an HDMI or DP port. The graphics card is necessary. The important thing is a stable graphics card (nVIDIA Quadro) designed for professional applications and not a gaming card (nVIDIA GTX) that is not very stable and consumes a lot of power. The professional graphics card is offered here, the cheapest in the Quadro range. The advantage of Intel Xeon processors over Intel Core is that they have ECC correction and therefore the PC is still stable. For settings, just focus on the BIOS and Windows, nothing more. We are not talking about setting anything on the graphics card. |
| [2019-12-11 23:51:43] |
| j4ytr4der_ - Posts: 938 |
|
Right, and this was my point about gaming tweaks. They are primarily focused on graphics card and clock speed (overclockers), neither of which seem to really make a big difference for replay testing. I've done all the BIOS tweaks I can, and haven't seen any appreciable improvement. My system however doesn't have UEFI and some claim that will make a difference. At the moment I'm continuing experiments using a RAM disk. It seems to make a very marginal improvement on a physical Windows 10 PC already running on an SSD drive, but on a Windows 7 VM (running in Parallels Desktop on Mac) it appears to give a pretty decent gain in replay performance. Nothing mind-blowing, but enough to be worth doing if you have the RAM to spare. Other than these things though, I haven't found anything that makes a big difference. My multi-core i7 950 PC running Windows 10, barely uses any of the RAM or CPU in the machine when running a replay. It's such a shame that more can't be done to utilize more of the available resources. |
| [2020-06-14 14:18:53] |
| Gianmarco - Posts: 10 |
|
i am still using an i7-2600K and i am thinking about an upgrade. is here anyone using an i9? i am trying to understand the difference between an xeon W22XX and a 1090k? |
| [2020-06-15 09:42:18] |
| Dorian - Posts: 57 |
|
Hello, the only difference is the ECC technology which allows the PC to be stable 24/7. Intel Core may make errors and display a blue windows screen when windows crashes. Intel Xeon are therefore recommended for a trading PC. |
| [2020-06-15 14:02:17] |
| Gianmarco - Posts: 10 |
|
thanks...and how important is the amount oof cores? how much more can i expect from fe 10 cores instead of 6? sorry for asking. das sierra handel the cores? Date Time Of Last Edit: 2020-06-15 14:06:27
|
| [2020-06-25 18:08:29] |
| Gianmarco - Posts: 10 |
|
Does someone uses hier an AMD Threadripper?
Date Time Of Last Edit: 2020-06-25 19:35:37
|
| [2020-12-30 02:20:28] |
| User485461 - Posts: 9 |
|
Hi Did anyone tried to start the SC with dedicated graphics cards like Nvidia ? I forced to start it under the windows graphics processor settings, but looks like it doesn't force start with it. I also added the SC programs under the Manage 3 d settings in Nvidia control panel, but nothing looks to be working Any ideas appreciated. |
| [2021-02-06 14:51:17] |
| User308424 - Posts: 70 |
|
#Answering to post no 1. Hi! I have a two questions regarding your setup. Will be grateful for any feedback. 1. Do I understand correctly that this GPU was selected only because it offers 4 DisplayPort slots? As there is no need to invest in anything better because SC does not use GPU? 2. What is the reason for going into win 10 pro for workstations instead of standard win 10? Is it because of the Xeon processor which was selected or due to the fact that as a rule worstation systems are more 'clean' and not filled with unnecessary programs which affect performance? Thanks! |
| [2021-02-06 16:05:31] |
| Dorian - Posts: 57 |
|
Only Windows 10 Pro for Workstations 64-bit is compatible with Intel Xeon processors. The graphics card is chosen because it has ECC correction, the Quadro ranges are professional ranges, for critical environments. A simple P620 is enough, because Sierra Chart does not consume a GPU, only the CPU. |
| [2021-02-23 19:06:21] |
| GravisHTG - Posts: 306 |
|
After my evaluation The power of a GTX 1650 4GB is perfect for this software. ( with openGL enable ) The core requirement per instance goes with the amount of different Symbol you are viewing at the same time. ( don't know perfectly here but I know that If I only have 1 Symbol on the chartbook it require way less CPU power than multiple symbol , and that is even more truth if one of your chart require some intense loading Studies ) Another point is that it looks like 1 core per Sub instance is good enough . When I put 3+ core , I always get 1 core max out and the other core a 40% max. 2 core seems just fine too. I put all my trading program on the last 8 core of my CPU , looks like windows love to utilize the first 4 core of the CPU. Edit I add another sub instance A full SPY chartbook with 2 TPO chart 1 Number Bar Chart full load and 4 other chart with lots of indicators and put it on 1 core , I have just found the way to adjust my Computer ! to see what utilize the core of the picture. core 15 - Interactive Broker TWS - This program take a lot of resources ! 7 X 1 core for Sierra chart. Trade Idea on all core. My computer have Hyperthreading off. Another EDIT. I just add another picture , this time with Hyperthreading on. I assign 2 core to each main program with above normal priority. core 15 and 31 - Interactive Broker TWS the next 7 core with their respective Hyperthread ( ex : 14 and 30 ) for Sierra chart main 1 X and 6 Sub instance. Trade Idea on all core and the rest of my program on all core. Like I said , I start on the last core of the CPU because windows seems to get his action the the first 4 core. The picture is when i am Downloading 1 new Big symbol on every sub instance except the main ( no chart ) and 1 sub that is on another Virtual desktop. I can say , that Hyperthreading on AMD is really useful for the CPU load. I don't know about latency but for my trading , it does not matter that much. And My fastest Refresh rate of 1 chart is 150MS on the 1s chart. Ill try to go faster but first I am looking for a stable setup. Just add another picture with no Trading program Open , just chrome, some small program and youtube on. ( by looking at the picture , seems like windows is using the first 4 and the Last 2 thread. If we go by deduction, the first 4 are Real core and the last 2 are HyperThread.) Edit again In live market data, I put 2 real core + 2 hyperthread to the main instance , it needed it as 1 core was at 80% Edit I have experiment a little bit and to put 2 core + 2 Hyperthread on each instance had the best result. From what I have check , on how windows behave and sierra chart behave. It always max out the first real core and the last HyperThread Core then its going to the next. So , 0-3 and 27-31 , I leave them open for windows, chrome, trade idea, youtube and all the other stuff im doing while trading. Then I assign 2 core + 2 hyperthread to each instance. The main have 2 dedicated core but my sub Share their second core 2 sub for 1 second core. Main instance core number 4-5 + 25-26 Sub instance 1 core 6-15 + 24-16 Sub instance 2 core 7-15 + 23-17 Sub instance 3 core 8-14 + 22-18 Ill try today if its good , but it seems like it perform very well at the moment. I am trying to figure out what is the best Processor , computer to get. As I want to upgrade to a laptop in the future. Date Time Of Last Edit: 2021-02-26 11:54:15
|
 cpu hyper.PNG / V - Attached On 2021-02-23 22:42:40 UTC - Size: 72.38 KB - 397 views cpu hyper no trading program.PNG / V - Attached On 2021-02-23 23:19:45 UTC - Size: 190.17 KB - 387 views |
| [2021-02-27 18:39:06] |
| User61576 - Posts: 443 |
|
how do i set SC with specific CPU's?
|
| [2021-02-27 19:25:41] |
| GravisHTG - Posts: 306 |
|
Check second post from Dorianstoll at the top.
Date Time Of Last Edit: 2021-02-27 23:43:36
|
| [2021-02-28 18:22:40] |
| User308424 - Posts: 70 |
|
Regarding OpenGL: My situation is: I have 2 instances with one chart in each of them. 3rd instance is just t get the data. I have market depth historical graph on both of the charts, some bid/ask volume, VBP study and number bars. So I don't run hundreds of charts with extremely complex studies... OpenGL not active. My problem was scrolling back and forth the charts - they were lagging a lot. I was surprised by that! I noticed one of my cores goes to 80-90% at that time and charts were still lagging! I thought having a relatively strong CPU (Ryzen 3700x) for such a small number of charts I do would be more than enough. Well I was wrong. I just activated OpenGL and I have to say this made for me a huge difference. Bigger than switching to dedicated NVMe disk for SC. Even with my cheap Nvidia NVS 510 card everything runs fast, smoothly and its very responsive. After reading first post of this thread I was more CPU oriented. After my tries and researches I have to say that not having OpenGL enabled its a mistake - at least in my situation. |
| [2021-12-24 09:15:57] |
| Yeager - Posts: 125 |
|
Just ordered a new build i910980xe and 2x2tb samsung MVNE with 64GB ram and using an older invidious geforce GTX 1070TI. Got new build for more cores and threads. Will let you know how it goes. For me I'm looking for the fastest DOMs and initial opening load up of 1 instance. I run 2x most days. |
| [2021-12-24 19:10:07] |
| GravisHTG - Posts: 306 |
|
After all my experimenting I end up with a 5800X and I am using process lasso to distribut the cpu cores. I always dedicated 5/5 ( core/thread ) to SC and its ok , as long as I have GPU power. my GPU is a 1070ti. ( did not try better than this but I think SC always take all of it ) so the faster the better. Hope this can help. ( my next machine will be a laptop with at least 8/8 cpu and a 3060+ gpu ) And I will get a dedicated monster Desktop ( like AMD EPYC CPU or INTEL variant of it ) - but that is not in the short term. |
To post a message in this thread, you need to log in with your Sierra Chart account: