Jump to content

Titan RTX NVLink SLI - can it run Crysis? We'll find out soon...

Kalm_Traveler
9 hours ago, JoostinOnline said:

How does the CPU hold up, and what do you have? That's still a bottleneck on Crysis 1, as it only uses one core and was designed with the idea that we'd have much higher single thread performance these days.

The CPU holds up fine.  7820x @ 4.8GHz.  

i9-9900k @ 5.1GHz || EVGA 3080 ti FTW3 EK Cooled || EVGA z390 Dark || G.Skill TridentZ 32gb 4000MHz C16

 970 Pro 1tb || 860 Evo 2tb || BeQuiet Dark Base Pro 900 || EVGA P2 1200w || AOC Agon AG352UCG

Cooled by: Heatkiller || Hardware Labs || Bitspower || Noctua || EKWB

Link to comment
Share on other sites

Link to post
Share on other sites

  • 2 weeks later...

Did some quick and dirty testing tonight to get an idea of what the net gain of doing the shunt resistor swap on these cards provides. Here's the same thing I posted on overclock dot net:

 

Did some testing tonight to get a ballpark of what my power mod is doing for these cards. Judging by the wattage being pulled through the UPS, these are the numbers I'm seeing:

idle on this rig is about 305 watts

when I run a CUDA load stress test on both cards together power usage sits about 1220 watts. Each card thinks it is pulling ~ 315 watts

since the rig is actually pulling 1220 watts, if we subtract what the cards think they're pulling so 315 watts x2 cards = 630 watts from 1220 we get 590w left over.

Given that the rig idling sits pretty stable at 305 watts, that leaves a mystery 285 watt discrepancy.

I'm sure that is actually being pulled by the cards given the shunt resistor swap, which rough math would then give me 142.5 watts per card more than what they think they're pulling.

so... 315 (what they see) + 142.5 (half of the discrepancy in wall load - granted the actual load at the card will be a little lower due to PSU efficiency) = ~ 457.5 watts per card

I don't have a graph showing how efficient a Corsair AX 1600i is at 1220 watts load, but I think the numbers are clear enough to know that with the 5MO resistors replaced with 3MO and the Nvidia OC Scanner curve my cards were auto set to they are definitely pulling north of 400 watts each - which means the next step is to keep them cool to let GPU Boost do it's thing.

HEDT: i9 10980XE @ 4.9 gHz, 64GB @ 3600mHz CL14 G.Skill Trident-Z DDR4, 2x Nvidia Titan RTX NVLink SLI, Corsair AX1600i, Samsung 960 Pro 2TB OS/apps, Samsung 850 EVO 4TB media, LG 38GL950G-B monitor, Drop CTRL keyboard, Decus Respec mouse

Laptop: Razer Blade Pro 2019 9750H model, 32GB @ 3200mHz CL18 G.Skill Ripjaws DDR4, 2x Samsung 960 Pro 1TB RAID0, repasted with Thermal Grizzly Kryonaut
Gaming Rig: i9 9900ks @ 5.2ghz, 32GB @ 4000mHz CL17 G.Skill Trident-Z DDR4, EVGA RTX 2080 Ti Kingpin, Corsair HX1200, Samsung 970 EVO Plus 2TB, Asus PG348Q monitor, Corsair K70 LUX RGB keyboard, Corsair Ironclaw mouse
HTPC: i7 7700 (delidded + LM), 16GB @ 2666mHz CL15 Corsair Vengeance LPX DDR4, MSI Geforce GTX 1070 Gaming X, Corsair SFX 600, Samsung 850 Pro 512gb, Samsung Q55R TV, Filco Majestouch Convertible 2 TKL keyboard, Logitech G403 wireless mouse

Link to comment
Share on other sites

Link to post
Share on other sites

also in case anyone wants to see exact math, I found an efficiency chart for the Corsair AX1600i PSU to get more exact number of what the shunt resistor mod is doing for these cards (again I posted this on overclock dot net):

 

Found an efficiency chart for the PSU here: https://www.anandtech.com/show/12645...-performance/5

So it is at ~ 95% efficiency at idle as well as at the peak 1220w wall pull I see with the CUDA stress testing.

Given that, the math breakdown is as follows:

1220w at the wall * 95% PSU efficiency is 1159w on the PC end

idle 305w at wall * 95% PSU efficiency = 289.75w on PC end

The cards seeing 315w load x2 cards = 630w perceived power use total for the Titan RTX cards

idle PC 289.75 + cards seen load 630w = 919.75 watts

this leaves a discrepancy of 239.25 watts on the PC end. Since the only other high-draw part in the system is idling during these tests (CPU sits at 1%), it is reasonable to take that discrepancy as primarily going to the cards.

239.25w / 2 cards = 119.625w per card over detected input

315 seen + 119.625 = 434.625 watts per card
-------------------------------------------------------------

Thus, with the card seeing only 315w but actually pulling 434.625 watts the 3MO resistors are giving a 37.97% TDP increase without tripping safe mode (434.625 divided by 315).

All in all that seems like a great return for a pair of $2 resistors and 5 minutes of my time (swapped them during waterblock install process).

HEDT: i9 10980XE @ 4.9 gHz, 64GB @ 3600mHz CL14 G.Skill Trident-Z DDR4, 2x Nvidia Titan RTX NVLink SLI, Corsair AX1600i, Samsung 960 Pro 2TB OS/apps, Samsung 850 EVO 4TB media, LG 38GL950G-B monitor, Drop CTRL keyboard, Decus Respec mouse

Laptop: Razer Blade Pro 2019 9750H model, 32GB @ 3200mHz CL18 G.Skill Ripjaws DDR4, 2x Samsung 960 Pro 1TB RAID0, repasted with Thermal Grizzly Kryonaut
Gaming Rig: i9 9900ks @ 5.2ghz, 32GB @ 4000mHz CL17 G.Skill Trident-Z DDR4, EVGA RTX 2080 Ti Kingpin, Corsair HX1200, Samsung 970 EVO Plus 2TB, Asus PG348Q monitor, Corsair K70 LUX RGB keyboard, Corsair Ironclaw mouse
HTPC: i7 7700 (delidded + LM), 16GB @ 2666mHz CL15 Corsair Vengeance LPX DDR4, MSI Geforce GTX 1070 Gaming X, Corsair SFX 600, Samsung 850 Pro 512gb, Samsung Q55R TV, Filco Majestouch Convertible 2 TKL keyboard, Logitech G403 wireless mouse

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×