Jump to content
Search In
  • More options...
Find results that contain...
Find results in...

ObeliskAG

Member
  • Content Count

    63
  • Joined

  • Last visited

Awards


About ObeliskAG

  • Title
    Member

System

  • CPU
    AMD Ryzen 7 3800X
  • Motherboard
    Asus PRIME X570 PRO
  • RAM
    64GB G.SKILL Trident Z Neo
    DDR4-3600 (16-16-16-36)
  • GPU
    Gigabyte 2080 Super Windforce OC
  • Case
    Fractal Design Meshify S2
    3x stock 140mm fans, 2x Corsair ML140 fans
  • Storage
    256GB Samsung 950 Pro
    1TB Samsung 970 EVO Plus
  • PSU
    Corsair RM850
  • Display(s)
    NEC MultiSync EA244UHD
  • Cooling
    Corsair H115i Pro RGB upgraded to 2,000 rpm fans
  • Keyboard
    Cooler Master Trigger (MX Brown)
  • Mouse
    Logitech G502
  • Sound
    Motherboard audio
    Rotel RA-960BX Integrated Amplifier, Image Concept 200 speakers, Sennheiser HD 600 headphones
  • Operating System
    Windows 10 Pro
  • Laptop
    Lenovo T480 with 16GB HyperX Impact DD4-2400 CL14

Recent Profile Visitors

The recent visitors block is disabled and is not being shown to other users.

  1. For all practical purposes, due to Folding, I haven't done any serious gaming for 65 days. It is now time to shut down and vacuum out the PC, then play! 2000 WUs! Due to the event, I have contributed 2,004 WUs and 138,302,610 points to the Linus Tech Tips Folding Team: https://folding.extremeoverclocking.com/user_summary.php?s=&u=867950 It's been fun and the Gold forum badge looks good Looking forward to the next Folding event. Regards, ObeliskAG
  2. You know, saying that you built a new system without sharing the specifications (parts), is just being a tease.
  3. True - but as Nvidia hasn't been clear about any architectural changes between the TU104 and the TU106, we just have the stated specifications to go by. I'm going to guess that the TU104 (2060 KO) has more cache at the SM level than the TU106 (2060).
  4. Fixed - edited previous post. (As the 2060 KO Ultra really only has a higher factory boost clock than a regular 2060.)
  5. Yeah, you will keep the front filter until you want the 5°C drop by removing it.
  6. Looks like some people are contemplating getting some more folding power. As ongoing folding is going to use a lot of electricity, I've updated my table. Assuming the P106-100 as the reference - the higher Relative Efficiency the better: It looks like the RX 5600 XT is excellent value. Attached the spreadsheet for your enjoyment. GPU Specs and Timings.xlsx
  7. PCIe 2.0 x 8 is 4GB/s * 0.60% = 2.4GB/s. PCIe 3.0 x 4 is 3.9GB/s. The P106-100 is a PCIe 3 card, so four cards on a PCIe 3 motherboard? With a 2080 Super, GPU-Z indicates a Bus Interface Load of 35% while folding.
  8. I have the Meshify S2 in white and love it - 4 x 140mm intake fans, 2x 140mm (100cfm) radiator exhaust fans, 1x 140mm case exhaust fan. Lots of space to work. I did remove the foam filter from the front intake grill and top exhaust grill.
  9. Nope. Each GPU gets its own folding slot allocated, so a crossfire bridge isn't used or required.
  10. Yes, but only under very demanding load conditions like multi-user servers, like a Storage Area Network server with lots of attached SSD, 10 or 40 gigabit Ethernet and/or fast fibre channel, etc. It's nice having lots of PCIe lanes (more than 16), but they cannot usually be well utilized on a desktop PC. Notice how PCI3 4.0 NVMe storage doesn't necessarily improve games loading times, etc. (Windows NTFS leaves something to be desired in this context.) PCIe 3.0 is made of ~1GB bi-directional lanes that can be shared between devices. It's conceivable that loading a billion vertices onto a graphics card will keep a 16-lane PCI 3.0 connection busy for a few moments, but this traffic load on the PCIe bus is generally not sustained for very long. You could use PCIe to compensate for a lack of texture memory on a video card, but the results are not great. Not sure how much CPU/GPU data exchange is required for folding, but I would suspect 4-lanes would be more than enough. PCIe 4.0 16-lane bandwidth (31.5GB/s) looks like dual-channel DDR4-2166 RAM (34.7GB/s).
  11. Short answer - floating point math specialization and massive parallelization. Folding uses floating point math. But if I can provide a more technical example: An AMD ZEN 2 core has 2 floating point adders and two floating point multipliers. In a perfect world where you could keep these 4 FP units functional all the time, using a Ryzen 3700x the math would look like: 4 floating point units per core x 8 CPU cores x 4,000,000,000 Hz (4 GHz) = 128,000,000,000 Floating Point Operations per Second - 128 GFLOPS. In reality, it is unlikely that the FPUs can be fed with a suitable workload 100% of the time, so this GFLOP number will be much lower. Also, as the FPUs is part of a general purpose CPU, they are constrained by available resources within the CPU. (Other functional unit requirements, memory bandwidth, power consumption, temperature, etc) It should be noted what most of what CPUs do is executing instructions involving integer math, not floating point. Now looking at one of Den-Fi's RX580 graphics cards, this GPU has 2,304 unified shaders. Each shader is capable of 2 single-precision floating point operations per clock cycle (measured in Hz). So: 2 floating point operations x 2,304 unified shaders x 1,340,000,000 Hz (1,340 MHz) = 6,174,720,000,000 Floating Point Operations per Second - 6,175 GFLOPS. As much less general purpose than a CPU, a GPU is optimized to perform floating-point math calculations (heavily used in 3D graphics) very efficiently. GPUs can execute a lot of integer instructions too, but their raison detre is floating point power. Does this explanation help?
  12. Is it just me, or is the supply of WUs slowing up again? In other news, hit a milestone on the LTT folding team:
×