Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Nvidia RTX 50 series discussion
- Thread starter MacCheetah3
- Start date
- Sort by reaction score
Due to the lack of x86 CUDA and PhysX, it would be better to stick with 40 series as long as you can and DON'T SELL IT so you can play older games years from now. Its a shame some of my favorite titles have FPS tanking like Mirrors Edge and Batman Arkham Asylum.
Folks have successfully gotten an older GTX1660 (uses MB power) to work as a Physx accelerator.
Still surprised no one is talking about making a wrapper/translator for the old CUDA programs to work on the new hardware. Seems like there is some money to be had by someone filling in the gap.
Still surprised no one is talking about making a wrapper/translator for the old CUDA programs to work on the new hardware. Seems like there is some money to be had by someone filling in the gap.
My point is not to sell what you have. Can you buy a GTX 1660 20 years from now at a reasonable price when you want to play a classic like Mirrors Edge? The only reason I have put up with the horrible Windows is because it was the king of backwards compatibility. Now this has changed. I will NOT be upgrading from my 4090 now unless NVIDIA changes course.
NVIDIA's 32-Bit PhysX Waves Goodbye with GeForce RTX 50 Series Ending 32-Bit CUDA Software Support
The days of 32-bit software support in NVIDIA's drivers are coming to an end, and with that, so does the support for the once iconic PhysX real-time physics engine. According to NVIDIA's engineers on GeForce forums, the lack of PhysX support has been quietly acknowledged, as NVIDIA's latest...
NVIDIA Support
nvidia.custhelp.com
And GN indeed managed to drag out some actual figures:
Basically, I agree, “disaster” is a fitting description of the RTX 50 series.
Recoverable/reconstructable? Maybe. If not, PC enthusiasts will wait 1.75 years for the next likely horrific launch.
Basically, I agree, “disaster” is a fitting description of the RTX 50 series.
Recoverable/reconstructable? Maybe. If not, PC enthusiasts will wait 1.75 years for the next likely horrific launch.
So to be clear to those asking. Older games run, you just can't turn on all the features on newer hardware because 32bit CUDA support has been dropped from the driver.
I am curious if there is enough backlash for folks to switch to AMD GPUs and if Nvidia would add 32bit support back.
It's been a S* show. I think that it's just best to wait for things to settle out. Unfortunately, 4070s are mainly in price-gouge mode now.
I'm using a GTX 1660 right now and was considering a 4070. The GTX 1660 has a DVI, HDMI and DisplayPort outputs and I'd like to run 3 4k monitors off of it. But 4070s have asking prices at $1K to $1.2K. It may be cheaper to buy a prebuilt, take out the 4070, put in the 1660 and sell the desktop.
At any rate I can just wait until this s*show eventually resolves.
The GPU price beatings will continue until moral improves.
Especially since everyone is going to ignore AMD.
I've been considering a 4060 as well. I suspect the same thing with these will happen when the 5060s come out.
I have no familiarity with AMD's product lines outside of the Vega 56 in my iMac Pro. I would like a 4xxx for the efficiency compared to the 1660.
RTX 30 series isn’t any better. 3090s are selling for $900-950 on average, from what I’ve been seeing. 3080 (Ti) models are about $600 on average.
😊
I grabbed a 4070 Ti ProArt OC for MSRP (i.e., $830) from Amazon and a 4080 ProArt for MSRP (i.e., $1250) direct from Asus back in October 2023. In May 2024, a few months following the 40 Super launch, I snagged an open box 4080 ProArt OC for $910 from B&H — it really was in practically new condition, including the protective plastic.
I felt damned if I was going to pay scalper pricing. As of today, it appears, I could recoup all of that. Yet, it’s also ridiculous.
Someone did a test and come to find out, if you just have a 4090 doing Physx you are potentially leaving huge amounts of performance on the table.
Just add this adorable model. ☺️
ZOTAC GAMING GeForce RTX 4060 8GB SOLO | ZOTAC
<p>The ZOTAC GAMING GeForce RTX 4060 8GB SOLO is an ultra-compact graphics card, featuring the NVIDIA Ada Lovelace architecture. It is the smallest ZOTAC GAMING 40-series card to-date, and could fit in 99% of PC cases with ease. Despite its small size, it has all the advantages of a 40-series...
www.zotac.com
ZOTAC GAMING GeForce RTX 4060 8GB SOLO [Open Box]
2 years warranty (shipped in generic box with no accessories)
www.zotacstore.com
Is RMA series x Digital to be considered.
I could order Scan uk 265.k 5070 pre build.
FPS via you tube seem fine@1440.
Is 507012gb going to be faster than 4070ti 12gb.
Meanwhile EE store have Issued Xbox RMA code if I pursue return.
I could order Scan uk 265.k 5070 pre build.
FPS via you tube seem fine@1440.
Is 507012gb going to be faster than 4070ti 12gb.
Meanwhile EE store have Issued Xbox RMA code if I pursue return.
Series X 1tb white joins Omen Raptor lake in refund space…
Hopefully decide which 265k/5070 option to choose.
Hopefully decide which 265k/5070 option to choose.
So i realized something with respect to the 5080. I wonder if they would have had to make a D variant if it was as fast/faster than a 4090.
What I find odd is the 5090 D seems to be as fast a a 5090 accord to TPU, but with no reviews of any D variants on the list it is hard to say if this is actually true.
What I find odd is the 5090 D seems to be as fast a a 5090 accord to TPU, but with no reviews of any D variants on the list it is hard to say if this is actually true.
Interesting… In “AI” performance, it is… Or so says Nvidia.
RTX 4090:
4th Generation Tensor Cores
1321 AI TOPS
RTX 5080:
5th Generation Tensor Cores
1801 AI TOPS
Again, directly from Nvidia — you’ll need to use the previously linked page for the non-D 5090 specs
RTX 5090 D:
5th Generation Tensor Cores
2375 AI TOPS
RTX 5090:
5th Generation Tensor Cores
3352 AI TOPS
How they are limiting it is unknown. Maybe Tensor core clock frequency. And even though it isn’t an apparent hardware difference, Nvidia does label them different physical dies: GB202-250 vs. GB202-300.
For gamers, the Dragon variant would be seemingly only crippled in DLSS and frame generation performance.
By the way, the RTX 4090 D:
4th Generation Tensor Cores
1177 AI TOPS
So, I don’t know/recall how the restriction(s) are determined.
Last edited:
Physx and Flow are now open source.
So now some enterprising individuals should be able to map the 32bit calls to 64bit to run on 50series GPU's and charge a kings ransom.
So now some enterprising individuals should be able to map the 32bit calls to 64bit to run on 50series GPU's and charge a kings ransom.
Well…
I wouldn’t care much, however, I use FanControl to adjust case fans according to GPU temp.
By the way, it affects both 40 and 50 series — at least in my situation.
🤦♂️😡
And they apparently didn’t fix this yet:
I wouldn’t care much, however, I use FanControl to adjust case fans according to GPU temp.
By the way, it affects both 40 and 50 series — at least in my situation.
🤦♂️😡
And they apparently didn’t fix this yet:
Last edited:
I used Fan Control for a while. When I swapped loop tubing I went with Aqua Suite instead because it allowed me to have the fans ramp when the loop water temp gets above 35C and have the pump speed ramp when the water temp difference (GPU rad temp and res temp) are greater than 8C. My case fans stay running at a constant speed.
Yeah those bugs didn't appear in the release notes for this latest driver, I didn't hear as much hubbub about it as I did the black screen one.