Is it going to be worth going 3090 to 4090?

[VR Flight Guy in PJ Pants]

Do not want to rain on your parade but rumours have it the new AMD GPU may be delayed till early next year. Not that I hate AMD but fingers-crossed

[SkateZilla]

don't understand why they choose every launch to put out the card with the lowest yields first..

the Ada Titan is cancelled, the RTX4090Ti may be right behind it, 

the RTX4080 and Likely RTX4080 ti are power hogs, and only show decent numbers when using DLSS3.0.

RTX4080 12 GB, is being discontinued (as it never should have existed), likely being held for a while and re-launched as the RTX4070 Ti it should've been.

But you got 4070, 4060, 4050, the cards that have the higher chip yields due to smaller size, more per wafer, etc etc. 

$50 all their production wafers are being used for RTX4090 and 4080s they dont have any for 4070/4060s

[Willdass]

2 hours ago, YoYo said:

About this.... Im surprised but I know few people who bought RTX 4090 and they was used VR (but not overlocked versions and they dont use any overlock of CPU as well) and no any issues with 850W PSU + 4090 ! Im thinking to buy 4090 with 1250W PSU but this tests surprised me.

And ???

 

The store screwed up, had one less in stock than they said... So did not receive mine, they estimate that mine should arrive next week. I will post here when I get it

[skywalker22]

Just for example, how 4090 runs on a game which supports multithreading, Assetto Corsa Competizione (which I also drive a lot): 

 

[Tom Kazansky]

29 minutes ago, SkateZilla said:

don't understand why they choose every launch to put out the card with the lowest yields first..

Well... a lot of people that don't have a problem with money have a problem with patience. So nVidia would sell a lot of RTX 4080s and 4070s but most of those customers won't buy a 4090 one month later, imho.

[VR Flight Guy in PJ Pants]

1 hour ago, skywalker22 said:

Just for example, how 4090 runs on a game which supports multithreading, Assetto Corsa Competizione (which I also drive a lot): 

Absolute gold. Thanks. There is no point and pretend neutral by saying worth it or not is to the eye of the beholder.

As a VR player, I would respectfully ignore all the reviews calling 4090 "too fast". It is like telling starving people that fat is bad for health.

Edited October 15, 2022 by VR Flight Guy in PJ Pants

[speed-of-heat]

what is fascinating is the DerBauer review, want to save 1/3rd the watts and lose 10% perf using a 60% power target, which seems to be the undervolting alternative.... 

 

Edited October 15, 2022 by speed-of-heat

[Hammer1-1]

https://gpu.userbenchmark.com/Compare/Nvidia-RTX-3080-Ti-vs-Nvidia-RTX-4090/4115vs4136

[FoxTwo]

14 hours ago, Supmua said:

That's a pretty strong statement considering the 13900K isn't even out yet.

Not quite the right thread, but the 13900K is no slouch especially at 1% percentile fps besides avg fps.

 

 

The little orange line on these graphs that intel is trying to hide is the 5800X3D. I'm not the one making the strong statement, intel is.

[Supmua]

2 hours ago, skywalker22 said:

Just for example, how 4090 runs on a game which supports multithreading, Assetto Corsa Competizione (which I also drive a lot): 

 

Seeing GPU frametime down to 7 ms (from 13s) at high setting in ACC is kind of crazy.  Looking forward to trying this too.

[VR Flight Guy in PJ Pants]

15 minutes ago, Supmua said:

Seeing GPU frametime down to 7 ms (from 13s) at high setting in ACC is kind of crazy.  Looking forward to trying this too.

Then someone will accuse you going off-topic to ACC.

 

[Werewolf_fs]

 

 

[Supmua]

Setting up some stuff to get ready for benchmarking with my 4090, won't have the card until midweek.  The fpsVR grab below is what I get with the default Harrier Caucasus Free Flight right now (9900K HT off/3090), ALT around 1500-1600 ft.  CPU frametime is pretty good on this mission and I'm pretty sure the GPU frametime will drop below 11ms with the 4090 which means I get 90 fps.  This is sort of best case scenario with light flying mission in a VR friendly aircraft.  Varjo Aero OpenVR mode at 3144 Horizontal pixels per screen (Very High setting on the headset but reduced SS to 76%, frame locked at 45 fps, Reshade 5 for sharpening and color  correction).

null

Edited October 15, 2022 by Supmua

[Supmua]

49 minutes ago, Werewolf_fs said:

 

 

It's interesting that CPU frametime also dropped a few ms?  Wonder why...but in any case this is a good thing.

Edited October 15, 2022 by Supmua

[Slammin]

I noticed that as well and assume is due to reduction of contention.  More headroom now due to the faster GPU processing keeping flows moving.

[Supmua]

7 minutes ago, Slammin said:

I noticed that as well and assume is due to reduction of contention.  More headroom now due to the faster GPU processing keeping flows moving.

Seems like it might help a bit with bottlenecking.

[Harlikwin]

33 minutes ago, Supmua said:

It's interesting that CPU frametime also dropped a few ms?  Wonder why...but in any case this is a good thing.

 

Yeah its interesting but lord both the CPU and GPU frametimes are horrid. I assume its because of the high pixel count on the aero I guess. I wonder what these numbers would be like on a G2.

 

[Supmua]

30 minutes ago, Harlikwin said:

Yeah its interesting but lord both the CPU and GPU frametimes are horrid. I assume its because of the high pixel count on the aero I guess. I wonder what these numbers would be like on a G2.

 

It was Syria + Apache though which is a bad combo for VR.  The frametimes are highly dependent on DCS settings also (as well as rendering pixels).  You can see my fpsVR grab above in much less stressful scenario and I rendered at slightly higher res than his, but used my own custom settings instead of High preset (which is very taxing for VR).  It's all about tuning and reducing things to get the frametime numbers that you want. 

With fine-tuning of DCS settings I bet I can make that scene much better in GPU performance (but the overall fps already at >45 fps so not much of a point doing it) at slightly reduced image quality.  The CPU is still going to drag the final fps down so it won't get to 90 in this map, but DCS on the Aero with locked 45 fps is very very good (except for this ongoing tearing issue which is probably related to the new NVIDIA driver).

You can run OXRTK with the G2 and check those frametimes in real-time also, it's in the advanced tab (the unit is in ns instead of ms with OXRTK).  I do it all the time when I run it in OpenXR mode, everybody should do this since it tells you exactly what's going on with the CPU and GPU and you can fine tune your setup and settings this way to get the performance you want.

Edited October 16, 2022 by Supmua

[Slick Twitchy]

Just be aware the 4090 is a BEAST. I have an alienware with a 3090 in it. NO way could I fit it in those cramped quarters. I ended up sending the 4090 back and ordered an entire new system. Just an FYI.

[Harlikwin]

2 hours ago, Slick Twitchy said:

Just be aware the 4090 is a BEAST. I have an alienware with a 3090 in it. NO way could I fit it in those cramped quarters. I ended up sending the 4090 back and ordered an entire new system. Just an FYI.

sir you make me sad, you could have sold it to someone here that would have appreciated it.

 

Also I thought the physical dimensions were pretty close to a 3090?

Edited October 16, 2022 by Harlikwin

[Hammer1-1]

48 minutes ago, Harlikwin said:

sir you make me sad, you could have sold it to someone here that would have appreciated it.

 

Also I thought the physical dimensions were pretty close to a 3090?

 

its thicc, boi

[speed-of-heat]

I think he is using high preset, it’s been a while but, that has a whole bunch of stuff on that we normally turn off…

2 hours ago, Harlikwin said:

Also I thought the physical dimensions were pretty close to a 3090?

 

Depends on the card, the FE is 302 mm (my 3090 is 287mm) but third party cards are 320mm and up, quite literally the only card I can fit in my case is the FE 4090.

[WipeUout]

5 hours ago, Supmua said:

It was Syria + Apache though which is a bad combo for VR.  The frametimes are highly dependent on DCS settings also (as well as rendering pixels).  You can see my fpsVR grab above in much less stressful scenario and I rendered at slightly higher res than his, but used my own custom settings instead of High preset (which is very taxing for VR).  It's all about tuning and reducing things to get the frametime numbers that you want. 

With fine-tuning of DCS settings I bet I can make that scene much better in GPU performance (but the overall fps already at >45 fps so not much of a point doing it) at slightly reduced image quality.  The CPU is still going to drag the final fps down so it won't get to 90 in this map, but DCS on the Aero with locked 45 fps is very very good (except for this ongoing tearing issue which is probably related to the new NVIDIA driver).

You can run OXRTK with the G2 and check those frametimes in real-time also, it's in the advanced tab (the unit is in ns instead of ms with OXRTK).  I do it all the time when I run it in OpenXR mode, everybody should do this since it tells you exactly what's going on with the CPU and GPU and you can fine tune your setup and settings this way to get the performance you want.

Having any kind of FPS lock or re-projection on will not give actual and accurate performance figures.  If you activate re-projection, the CPU frametime will be artificially high and not representative.  When you do benchmarking, it is better to let FPS free, set your HMD to the highest refresh rate and do not use re-projection.  Get your settings to high quality across the board and make sure your FPS stays below the refresh rate to ensure you get a valid result.  You can pump up the supersampling also if settings are not enough to keep your baseline FPS testing bellow the HMD refresh rate.  And last but not least, use a track file to have a constant and stable testing environment.

[Slick Twitchy]

I got it for a <profanity>ty price, and off of Amazon. It came pretty quick, so I think it's (somewhat) available-if you can accept the pain of paying over retail price by a little. I paid over 2,500 USD. I barely researched it, so the one that I got may have been larger than standard versions, but it wasn't even close to fitting. I'd say a good 30% larger than my 3090. I figured it would probably be "safer" to get a system built around it, and return it via Amazon. If someone wanted to buy it, I still have it, but am returning it shortly. I think it would take too much trust to coordinate the transaction. But I'd be willing to try for this community. 

[Werewolf_fs]

About Ismael Jorda test with the 4090, I already talked to him about the fact that the instant action of the chosen, Apache in Syria, is a bit limiting due to the CPU time, he has already told me that he will continue to do tests in more appropriate instant action missions to really see what brings this new 4090.

I personally have been surprised by the improvement about 3090Ti in a mission limited by the CPU, also improving the time of the cpu!!

Edited October 16, 2022 by Werewolf_fs