TPU Page for the RTX 4050 OC @ 2.8 GHz

[SJZL 2.0]

I decided to edit the TPU 4050M page to make a more accurate representation of the way my 4050 is clocked on my laptop. Any thoughts?

 

[GerKNG]

specified boost clock and what you're achieving is not the same.

 

[SJZL 2.0]

specified boost clock and what you're achieving is not the same.

I know, using the relative & theoretical performance data to scale actual performance isn't always accurate. But the way I calculated & scaled the 4050 in both data charts are.

 

[P4-630]

Sorry but I don't see a 4050 Mobile surpassing the performance of a 2070 Super...

 

[Dr. Dro]

No way, the 2070 S will walk on the 30 and 40 gen 50-class cards

 

[P4-630]

The current 4050 Mobile specs as now in the TPU GPU database seems ok.

NVIDIA GeForce RTX 4050 Mobile Specs

NVIDIA AD107, 1755 MHz, 2560 Cores, 80 TMUs, 48 ROPs, 6144 MB GDDR6, 2000 MHz, 96 bit

www.techpowerup.com

 

[Dragam1337]

The current 4050 Mobile specs as now in the TPU GPU database seems ok.


View attachment 329993

NVIDIA GeForce RTX 4050 Mobile Specs

NVIDIA AD107, 1755 MHz, 2560 Cores, 80 TMUs, 48 ROPs, 6144 MB GDDR6, 2000 MHz, 96 bit

www.techpowerup.com

Does indeed seem accurate.

Smells alot like a 4050 owner wanting it to come off as better than it is...

 

[SJZL 2.0]

Sorry but I don't see a 4050 Mobile surpassing the performance of a 2070 Super...

Does indeed seem accurate.

Smells alot like a 4050 owner wanting it to come off as better than it is...

No way, the 2070 S will walk on the 30 and 40 gen 50-class cards

For those who think the 4050 surpasses the 2070S just by looking at the chart. Don't forget using relative performance chart to scale actual performance is not always accurate in general. I have said this beforehand.

Clocks in the 4050 are different for each laptop depending on wattage. Mine tops out at 2505 MHz non-OC and 2835 MHz OC. This is why I made this to show what a TPU page of a 4050M clocked at 2.8 GHz would look like.

 

[Fouquin]

Honestly when you're talking about stock versus overclocked, the 4050 Mobile at 2.7-2.8GHz is going to be in the same league as a 2070 Super in some scenarios. Here's a quick comparison using 3DMark's Time Spy test.

The 2070 Super in the 3DMark comparison is overclocked 13% over a stock 2070 Super, and exceeds the overclocked 4050 Mobile by 11-17% in the graphics tests where the GPU is the primary contributor. So it's not far outside the realm of possibility that a stock 2070 Super and a roughly 2.8GHz 4050 Mobile would be within 2-3% of each other, with a very slight chance of the 4050 Mobile being the faster GPU depending on what the test is. With the benefit of generally faster CPU options to pair with, as well as being able to reach 10-15% overclocks itself, the 2070 Super should always be faster in practice.

 

[Dragam1337]

Honestly when you're talking about stock versus overclocked, the 4050 Mobile at 2.7-2.8GHz is going to be in the same league as a 2070 Super in some scenarios. Here's a quick comparison using 3DMark's Time Spy test.

The 2070 Super in the 3DMark comparison is overclocked 13% over a stock 2070 Super, and exceeds the overclocked 4050 Mobile by 11-17% in the graphics tests where the GPU is the primary contributor. So it's not far outside the realm of possibility that a stock 2070 Super and a roughly 2.8GHz 4050 Mobile would be within 2-3% of each other, with a very slight chance of the 4050 Mobile being the faster GPU depending on what the test is. With the benefit of generally faster CPU options to pair with, as well as being able to reach 10-15% overclocks itself, the 2070 Super should always be faster in practice.

Surely the tpu pages are about stock gpus though, and not what some random guy was able to oc his gpu to...

 

[Fouquin]

Surely the tpu pages are about stock gpus though, and not what some random guy was able to oc his gpu to...

They are, and this post is about an edited page that reflects what some random guy was able to OC his GPU to. Everyone taking issue with the screenshot in the OP seems to be simply ignoring the entire post; the image is edited to reflect the overclocked status of this particular 4050.

So if the "Relative Performance" chart shows a stock 2070 Super and OP's now overclocked 4050 Mobile, it is not far off that they are only 2-3% apart. Though the 4050 Mobile is behind the 2070 Super, not above.

 

[tabascosauz]

For those who think the 4050 surpasses the 2070S just by looking at the chart. Don't forget using relative performance chart to scale actual performance is not always accurate in general. I have said this beforehand.

Clocks in the 4050 are different for each laptop depending on wattage. Mine tops out at 2505 MHz non-OC and 2835 MHz OC. This is why I made this to show what a TPU page of a 4050M clocked at 2.8 GHz would look like.

Assuming that 2835 is sustainable and leads to actual performance gains (which isn't always true for OCs on Turing and later hardware), I don't see a problem with having a bit of fun, just struggling to see any value in doing so outside of pure theory. Nvidia GPUs generally don't hit their max clocks under meaningful load, core clock starts coming down with load.

6GB 96-bit memory subsystem on the 4050 might have something to say about performance, probably ensures that core clock doesn't scale so perfectly with perf, especially considering what the 2070 Super has. Last I checked, 2070 Super Time Spy graphics score should be 10k+, so far I've only seen some heavily undervolted 4050Ms doing 8k or so. Time Spy is a pretty optimistic measure of GPU perf, so I don't know if you'll find an easier yet reasonably practical benchmark.

 

[Fouquin]

Last I checked, 2070 Super Time Spy graphics score should be 10k+, so far I've only seen some heavily undervolted 4050Ms doing 8k or so.

You missed the link to a Time Spy comparison with a 4050 Mobile at 2843MHz pulling just shy of 10K.

 

[tabascosauz]

You missed the link to a Time Spy comparison with a 4050 Mobile at 2843MHz pulling just shy of 10K.

Hmmm pretty good. Original chart was a bit optimistic but in the right ballpark. Stock LOQ should already be achieving somewhere north of 2600MHz on the 4050M though, so not sure if it's that much of a jump, just TPU's 4050M page being off the mark.

 

[SJZL 2.0]

You missed the link to a Time Spy comparison with a 4050 Mobile at 2843MHz pulling just shy of 10K.

My LOQ is at 95 watts maximum which is a shy away from 105 watts which is the most power hungry the RTX 4050 could be. It gets 9000-9500 for time spy at OC unlike around 10k with a 4050 at 105 watts and OC.

6GB 96-bit memory subsystem on the 4050 might have something to say about performance, probably ensures that core clock doesn't scale so perfectly with perf, especially considering what the 2070 Super has. Last I checked, 2070 Super Time Spy graphics score should be 10k+, so far I've only seen some heavily undervolted 4050Ms doing 8k or so. Time Spy is a pretty optimistic measure of GPU perf, so I don't know if you'll find an easier yet reasonably practical benchmark.

Speaking of GPU undervolting, the voltage clock option in MSI Afterburner is greyed out on my end. And I am not sure about which curve editor config would overvolt my GPU.

 

[tabascosauz]

Speaking of GPU undervolting, the voltage clock option in MSI Afterburner is greyed out on my end. And I am not sure about which curve editor config would overvolt my GPU.

So you were maxing out clock on stock Vcore alone? slider is not very useful, all recent years' Geforce GPUs should be UVd in the curve editor (ctrl+F). Pretty straightforward, pick the point where your desired vcore and clock intersect, drag the curve up to that point, hit apply, and repeat if necessary until the curve after that point is completely flat.

Assuming ur stock Vcore limit is like 1.1V, it should be more than enough to work with. I can't remember the last time the vcore slider actually worked and was relevant

 

[SJZL 2.0]

So you were maxing out clock on stock Vcore alone? slider is not very useful, all recent years' Geforce GPUs should be UVd in the curve editor (ctrl+F). Pretty straightforward, pick the point where your desired vcore and clock intersect, drag the curve up to that point, hit apply, and repeat if necessary until the curve after that point is completely flat.

Assuming ur stock Vcore limit is like 1.1V, it should be more than enough to work with. I can't remember the last time the vcore slider actually worked and was relevant

Using the slider was effective for me but I will try that, also I meant overvolting rather than undervolting since the GPU is already extremely power saving, a little too much to be exact.

 

[tabascosauz]

Using the was effective for me but I will try that, also I meant overvolting rather than undercoating since the GPU is already extremely power saving, a little too much to be exact.

Curve editor isn't limited to undervolting. If power limit slider is greyed out or already max out there's not much more to do on software side. It's a 4050M like the other Time Spy guy's but VBIOS power limits and power delivery on the board won't be the same.

 

[SJZL 2.0]

Curve editor isn't limited to undervolting. If power limit slider is greyed out or already max out there's not much more to do on software side. It's a 4050M like the other Time Spy guy's but VBIOS power limits and power delivery on the board won't be the same.

So just flatten the curve & have the line at 2800 for overvolting I guess??

 

[tabascosauz]

So just flatten the curve & have the line at 2800 for overvolting I guess??

well it won't be overvolting since there's no access to voltage slider, no? Just operating within the stock Vcore limits, and yes just flatten the curve and test for stability and clock behaviour

 

[SJZL 2.0]

well it won't be overvolting since there's no access to voltage slider, no? Just operating within the stock Vcore limits, and yes just flatten the curve and test for stability and clock behaviour.

It crashes when I flatten the curve at 2.8 GHz, & slider works just fine so I'm sticking with it.