Implications of RTX3050 performance on PCI-E 2.0

Z

ZodaEX

Supreme [H]ardness
Joined
Sep 17, 2004
Messages
4,902
I'm looking at getting a RTX 3050 to upgrade my PC, however I just noticed that this card only supports an 8x PCI-E interface.
My Sandy Bridge Core i5 CPU limits my motherboard to run at PCI-E 2.0 speeds however, so in my situation I'd be limited to PCI-E 2.0 8x speeds which I believe would cut the card's bandwidth down to 1/4 of what would normally be available to it?
How much do you think this would nerf performance of the card for me? The GPU is still very much my bottleneck in Witcher 3 and Dying Light, so I'm curious what you guys think here about this paring of hardware. Thanks!
 
TurboGLH said:
PCIe 2.0 x8 is 3.0 x4 bandwidth. If the shitty 6500xt takes a hit when on 3.0 x4, it's hard to see how the 3050 won't also.
Click to expand...

The 6500xt's PCIe problem only exists because it hits the 4gb VRAM wall and has to pull from system RAM.

8gb of VRAM should make it way less of a concern on the RTX3050. It'll run into compute or mem bandwidth limitations long before VRAM is an issue.
 
I don't see a impact running at 2.0 speeds. If it was a higher end card I would think it would be a issue. Those cards were built for low end systems which don't need/use a lot of gpu power. I think the biggest problem you will see is it being a lower end card and not the pci-e being a bottleneck. Your I5 Sandy bridge would prob be more of a bottleneck. Run a few benches and.compare gou scores.
 
I wouldn't waste my money or time putting a modern card in a system with a Sandy Bridge 4 core/ 4 thread CPU. Even in games that you can get 60 FPS for an average you'll be lucky if they're not a stutterfest when it comes to 1% lows. There was a video that I watched a while back with a 4690, which is quite a bit better than what you have, and most modern games were not very playable due to the minimum frame rate drops and stuttering.
 
An I5 2400 will pair with something like a GTX 1050 TI or so... If you drop in a I7 2600 or a 2700 then you could almost fully utilize a GTX 1060 6gb

FWIW, PC Builfd bottleneck calculator shows a 26% bottleneck with an I7 2600 and a 1070 ( No 3050 data, but 3050 is relative 1070 performance ). A 2400 would be closer to 50% bottleneck.... Imagine spending $500 and getting 1/2 of what you bought.

A ~$200 GTX 1060 and a $50 I7 2700 will get you the most out of your setup and save you $250ish... Or just grab an RX 460/560 / 1050 TI in the $150 mark to get a still good bump over a 7750. Your 2400S will run it fine... I had a 7790, and it was a great card. I blew it up mining after it was retired from gaming
 
TurboGLH said:
PCIe 2.0 x8 is 3.0 x4 bandwidth. If the shitty 6500xt takes a hit when on 3.0 x4, it's hard to see how the 3050 won't also.
Click to expand...


But the question is, how much?


Here is the falloff measured by Techpowerup (equivalent to x16 2.0)

relative-performance_1920-1080.png



And here's the falloff for a card with almost 3x the performance (3080


relative-performance_1920-1080.png



If er use them both with the same relative starting point - x16 3.0 RTX 3080 (99%)approximates the same falloff on 3050 at x8 3.0 (98%), so the next drop down should be expected performance attained (slightly below 95%)!

the 6500 XT is inherently castrated in three different departments ( pcie, local bus, and vram ), so finding playable settings is already a pain at stock x4 4.0 bandwidth - it's impossible when you do something as stupid as run this crap on Z68

relative-performance_1920-1080.png
 
Last edited:
I think that spending $500 on a 3050 to upgrade a machine with a $15 CPU is a dubious use of money, regardless of what PCI-e version its on. There are a lot of GPUs faster than a 7750, you can get a 780 for $130 that, while not remotely modern, will be something like four times faster than your 7750.
 
whateverer said:
But the question is, how much?


Here is the falloff measured by Techpowerup (equivalent to x16 2.0)

View attachment 447027


And here's the falloff for a card with almost 3x the performance (3080


View attachment 447028


If er use them both with the same relative starting point - x16 3.0 RTX 3080 (99%)approximates the same falloff on 3050 at x8 3.0 (98%), so the next drop down should be expected performance attained (slightly below 95%)!

the 6500 XT is inherently castrated in three different departments ( pcie, local bus, and vram ), so finding playable settings is already a pain at stock x4 4.0 bandwidth - it's impossible when you do something as stupid as run this crap on Z68

View attachment 447029
Click to expand...

Good research. The only thing that sticks out, PCIe 2.0 x8 is equivalent to PCIe 1.1 x16.

If we're using the extra data from the 3080 to try and extrapolate the performance drop on a 3050, then the performance will be in the mid 80% range.

I don't know that we can make the comparison, a 3080 will be used in situations where vram usage will be higher, and is more likely to have to pull from system ram. But, without actual data.....
 
TurboGLH said:
Good research. The only thing that sticks out, PCIe 2.0 x8 is equivalent to PCIe 1.1 x16.

If we're using the extra data from the 3080 to try and extrapolate the performance drop on a 3050, then the performance will be in the mid 80% range.
.
Click to expand...


No it wont. different level of performance cards have nearly the same levels of falloff (assuming you don't castrate them from the beginning, like the 6500 XT); the base bandwidth requirements will be different for each card (but the graphs are almost identical once you saturate the bus)

I'm using percentage falloff, so we can find the right point in the graph (that missing data point from the Techpowerup review)

Nearly every video card without a castrated setup has a similar falloff pattern to that 3080:

See GTX 1080:

perfrel_1920_1080.png


and GTX 980


perfrel_3840.gif


My point is, if you have two points on that graph, then you can easily interpolate the third (because we see none of the insane falloff like that broken 6500 XT); relative percentages allow you to do that!

So, the numbers Techpowerup has are 100% (8x 4.0), and 99 % (8x 3.0), so the anticipated performance for the next cut would be 94-95%

I do agree that it's an overpriced card right now, but when you're stuck handling all the unanticipated "6500 XT sucks worse than anyone could have thought" traffic from AMD, its going to take a few months to fall below $300 (and it will, because it mines like shit)
 
Last edited:
jobert said:
I wouldn't waste my money or time putting a modern card in a system with a Sandy Bridge 4 core/ 4 thread CPU. Even in games that you can get 60 FPS for an average you'll be lucky if they're not a stutterfest when it comes to 1% lows. There was a video that I watched a while back with a 4690, which is quite a bit better than what you have, and most modern games were not very playable due to the minimum frame rate drops and stuttering.
Click to expand...

I agree, a 3050 on a Sandy Bridge i5 is going to be extremely CPU limited
 
WilyKit said:
I agree, a 3050 on a Sandy Bridge i5 is going to be extremely CPU limited
Click to expand...

Do you think the AMD 6500XT or 6600 would be a better choice to make Witcher 3 run better? According to Afterburner, my GPU by far is my main bottleneck.
 
Last edited:
Interesting findings here. Looks like the 3050 is taking a pretty big hit as you move down to pcie 2.0. The lows are bad and stuttering galore.

Edit: If you are dead set on keeping the platform I agree with others that the i7 and a 1650 or 1050 ti and at most a 1060 would be all I’d spend on it. Now if you are planning on a new system and want to splurge on a current gen GPU then you could just buy the 3050 or 6600 and live with it until you carry it to a more modern platform.

 
Last edited:
ZodaEX said:
Do you think the AMD 6500XT or 6600 would be a better choice to make Witcher 3 run better? According to Afterburner, my GPU by far is my main bottleneck by far.
Click to expand...

Yes, it'll cost you half as much and perform about as well considering your platform. The better option would of course be the 3050 with a more modern system, if your budget allows for it.
 
whateverer said:
No it wont. different level of performance cards have nearly the same levels of falloff (assuming you don't castrate them from the beginning, like the 6500 XT); the base bandwidth requirements will be different for each card (but the graphs are almost identical once you saturate the bus)

I'm using percentage falloff, so we can find the right point in the graph (that missing data point from the Techpowerup review)

Nearly every video card without a castrated setup has a similar falloff pattern to that 3080:

See GTX 1080:

View attachment 447133

and GTX 980


View attachment 447134

My point is, if you have two points on that graph, then you can easily interpolate the third (because we see none of the insane falloff like that broken 6500 XT); relative percentages allow you to do that!

So, the numbers Techpowerup has are 100% (8x 4.0), and 99 % (8x 3.0), so the anticipated performance for the next cut would be 94-95%

I do agree that it's an overpriced card right now, but when you're stuck handling all the unanticipated "6500 XT sucks worse than anyone could have thought" traffic from AMD, its going to take a few months to fall below $300 (and it will, because it mines like shit)
Click to expand...

Again, we're trying to extrapolate how a 3050 would perform, based on performance of other cards in a similar situation. It's not cut and dry how it would all react.

But I have to reiterate, the first chart you used for a comparison (3080) shows performance at 87% when run at PCIe 2.0 8x equivalent speeds.

Someone did some actual testing. Avg fps was decent, not much of a drop at all.

1% low, Using 2.0 x8 gets you an average of 89.8% performance vs 3.0 x8.
 

Attachments

  • Screenshot_20220224-135124.jpg
    Screenshot_20220224-135124.jpg
    138.1 KB · Views: 0
  • Screenshot_20220224-135148.jpg
    Screenshot_20220224-135148.jpg
    127.9 KB · Views: 0
  • Screenshot_20220224-135203.jpg
    Screenshot_20220224-135203.jpg
    134.5 KB · Views: 0
  • Screenshot_20220224-135241.jpg
    Screenshot_20220224-135241.jpg
    136.9 KB · Views: 0
TurboGLH said:
Again, we're trying to extrapolate how a 3050 would perform, based on performance of other cards in a similar situation. It's not cut and dry how it would all react.

But I have to reiterate, the first chart you used for a comparison (3080) shows performance at 87% when run at PCIe 2.0 8x equivalent speeds.

Someone did some actual testing. Avg fps was decent, not much of a drop at all.

1% low, Using 2.0 x8 gets you an average of 89.8% performance vs 3.0 x8.
Click to expand...
from what ive seen while googling the gddr6x gets hit harder with the speed drop. since the 3050 is gddr6 and most cards with that show little drop, id assume the same.
 
Could always go [H]ard and try out the RTX3050 in your current system to discover the results. The only way to know for sure instead of only speculating.
 
StormNobleheart said:
Could always go [H]ard and try out the RTX3050 in your current system to discover the results. The only way to know for sure instead of only speculating.
Click to expand...
I did that when I had an w3690 system and a 1070. Not any faster than a 1060 6gb.
 
chameleoneel said:
Tech powerup did PCI-E scaling articles for both the RX 6600 XT and RX 6500 XT. Kinda surprised they didn't do one for the RTX 3050.

https://www.techpowerup.com/review/amd-radeon-rx-6600-xt-pci-express-scaling/

https://www.techpowerup.com/review/amd-radeon-rx-6500-xt-pci-express-scaling/
Click to expand...


They did, that was where I was saying we had to extrapolate the third point in their 26 game summary:

https://www.techpowerup.com/review/evga-geforce-rtx-3050-xc-black/31.html

Unfortunately, they have not provided it, so you're left with Russian sites if you want detailed frame-time analysis ( but you'll have to sift through each game, instead of just providing an average)
 
Last edited:
FWIW... Newegg has some RTX 30 series just selling without the shuffle now, but in bundles.. This one with an EVGA 3050 is one of the 'slightly' better versions at a $329 retail, with a $80 power supply.. Yeah who needs a power supply, but it might come in handy if you decide to build a system for the RTX 3050 to breathe with, which could be as simple as a Ryzen 2600 or so, sitting on an A320 board. https://www.newegg.com/Product/ComboDealDetails?ItemList=Combo.4475021&quicklink=true
 
You must log in or register to reply here.
Back
Top