It _MIGHT_ be able to RayTrace but Can it Fold?

[Gorgon]

TLDR: Yes it can, very nicely

 

AnandTech is one of the few sites that posts Compute Benchmarks in their product reviews. It would be nice for the folding community if more sites did this (hint hint LMG On their Review of the RTX 2070 I was intrigued by the single-precision FaHBench results which showed the stock RTX 2070 with 2304 Shaders achieving a score almost identical to a GTX 1080Ti.

As the 2070 was released I kept an eye on the prices and waited for a few reviews on sites I trust. Tech Jesus Steve at Gamers Nexus and others pointed out that there were two variants of the RTX 2070, both with the same number of shader cores but the lower-end units are apparently locked to a lower maximum boost clock.

 

With that in mind I kept an eye out on sales and picked up a EVGA GeForce RTX 2070 XC GAMING, 08G-P4-2172-KR, 8GB GDDR6, Dual HDB Fans & RGB LED for $750Cdn from a local bricks-and-mortar retailer not wanting to wait over a week for delivery from the US.

 

I pulled my test folding rig out of production:

• Acer iTX AX3400 Motherboard, nForce MCP78PV Chipset and a PCIe 2.0 x16 Slot
• AMD Athlon 64 XII 220 2.8GHz CPU
• 3GB DDR2 1333 RAM (2GB, 1GB)
• EVGA GTX1060 06G-P4-6163-KR
• Corsair TX650M Power Supply
• Fractal Designs Define S4 Case
• Noctua NA-FC1 Fan Controller
• 2 x 140mm Noctua iPPC 3000 PWM Fans
• 1 x 140mm Fractal Design Silent R2 Fan
• 2 x 140mm Coolmaster SP Fans

And installed the 2070:

I'm running Ubuntu 18.04. (Bionic Beaver) on this system and had to upgrade the NVidia binary drivers to version 4.10.73 to support the new card.

 

Preliminary results show with a +125/800MHz GPU/Memory overclock it is yielding 1.2-1.3 MPPD and drawing about 250W total from the wall with 50-60 Watts from the CPU, motherboard and fans.

 

So, for the price somewhere between what a GTX 1070Ti and a GTX 1080 is currently retailing for in Canada the RTX is a better value.

 

And I did check, it does support Hairworks so it will be fully backwards compatible.

[Dissitesuxba11s]

Do you know if there are any issues with bottlenecking with older hardware when it comes to folding?

[Gorgon]

I’m using 10 year old hardware with the latest GPU. Dual core 2.8GHz processor has shown no measurable change in utilization.

 

The PCIe2 x16 Slot in the mobo is equivalent to a PCIe3 x8 and the PCIe utilization as reported by the GPU has gone from about 34MBps received and transmitted to 47MBps received and 65MBps transmitted which is not surprising as this beast would need to get fed more often than the 1060.

 

I don’t fully trust the Bus Utilization numbers as reported by the GPU as they are instantaneous values rather than a byte counter so it depends on sampling but should provide a good indication after a few days.

 

The rule of thumb is you should have at least four lanes of PCIe3 for folding (x4) and some say x8 but empirical evidence suggests that Cards fold faster under Linux rather than Windows and one of the main reasons appear to be that Linux has less contention for the PCIe bus.

 

So, if your hitting a wall then verify that the slot your card is in has 8 lanes available and, if not, try moving it to a slot that does.

 

Finding a motherboard that provides 2 slots of PCIe3 at x8 with Intel processors is difficult in the 100-300 chipset series as most socket 1151 CPUs only have 16 PCIe3 lanes and the Chipset has an additional lanes but they are choked by the interconnect between the chipset and CPU only having a DMI3 connection (x4).

 

So a lot of motherboards with 2 physical PCIe3x16 slots the second slot may only be running at x4 depending on what other system components need PCIe lanes (NVMe M.2 slots, for example).

 

Without going to the x99-series motherboards the most you can expect is to be able to run 2 PCIex16 slots at x8/x8 but if, and only if, the motherboard has a couple of PCIe Multiplexor (PLX) chips which are pricey.

 

In General boards identified as SLI-capable should be able to deliver x8/x8 but you have to read the manuals carefully to ensure your mix of other components requiring lanes is not forcing the second slot to run in x4 or worse x1

 

Sometime you need to make a adjustment in the BIOS to change how the lanes re distributed

[Dissitesuxba11s]

I have the Asus Crosshair Hero VI with a Ryzen 1700 and it does look like both my cards (980 and 1080Ti)  are on PCI-e 3.0 x8.

 

I would personally like to transfer to Linux but unfortunately, I have a lot of files and programs already installed that take advantage of Windows, not to mention I also paid for this license. I'm also fairly new to Linux and am currently learning it on one of my older laptops. Maybe when I decide to set up a second system I'll make it a standalone Linux machine so that I can compare between the two.

[Gorgon]

1 hour ago, Dissitesuxba11s said:

I have the Asus Crosshair Hero VI with a Ryzen 1700 and it does look like both my cards (980 and 1080Ti)  are on PCI-e 3.0 x8.

 

I would personally like to transfer to Linux but unfortunately, I have a lot of files and programs already installed that take advantage of Windows, not to mention I also paid for this license. I'm also fairly new to Linux and am currently learning it on one of my older laptops. Maybe when I decide to set up a second system I'll make it a standalone Linux machine so that I can compare between the two.

You could dual boot or run one OS in a VM on the other.

 

i.e. Linux on Hyper-V on Windows

or Windows in a docker container on Linux

 

You can also run Linux natively on Windows. Forget the exact mechanism but there an option in Windows 10 additional applications.

 

But if your willing to either use VNC or just the command line you can run a Linux box on really old hardware.

 

Using a VM has the benefit of being able to use pass-through to use hardware natively in a VM though.

[Dissitesuxba11s]

2 minutes ago, Gorgon said:

You can also run Linux natively on Windows. Forget the exact mechanism but there an option in Windows 10 additional applications.

Yup, you can download Ubuntu on the Windows store. I was originally going to do that when I wanted to dabble with Linux but decided to just have it on my old laptop.

 

10 minutes ago, Gorgon said:

Linux on Hyper-V on Windows

This has piqued my interest. I don't have any experience with virtual machines but I might try it out just to learn the ropes.

 

I know you said you run only Linux, but do you know if there is any performance impact of running folding on a Linux VM on Windows?

[Gorgon]

4 hours ago, Dissitesuxba11s said:

I know you said you run only Linux, but do you know if there is any performance impact of running folding on a Linux VM on Windows?

I run folding on Windows too.

 

my daily driver is Windows 10 and since I got the 2070 I pulled the 6GB 1060 out of my Linux "Dev" box and installed it in my Daily Driver replacing my FirePro W4100.

 

I also CPU fold on:

My DD Win 10 Pro 64 Haswell Xeon

Win 7 64 Pro VM on i7 NUC ESXi

Win 7 Pro 64 i5 Laptop

Win 10 Home 64 i7 2 in 1

Win 10 Home 64 i5 NUC

 

I don't know about Linux on Hyper V as everytime I've played with it I've found it wanting but in order to GPU fold you'd need to use "pass-thru" for the GPU to dedicate it to the Linux VM so if it works it should be close to raw performance assuming, of course, that Hyper V actually behaves as a "bare-metal" hypervisor should.

 

In theory you should be able to setup a small VM with a couple of threads and 2-4GB Memory and 64GB Disk and pass-through the GPU and see a bump in point production.

 

In practice, however, you still need to contend for access to the PCIe bus which is what is rumored to be the bottleneck in windows so the MS Hypervisor might have the same issue.

 

An interesting experiment but one I would not attempt until this month is over.

[Dissitesuxba11s]

6 minutes ago, Gorgon said:

An interesting experiment but one I would not attempt until this month is over.

Yea I might try this after the folding month as well. I'll already going to be down one 1080TI for maybe 10 days of this month so I won't be experimenting just yet.