RTX 4080 vs. RTX 4090


DH7777
 Share

Recommended Posts

1 hour ago, Chief_Trent said:

RayTrace is only run on the CPU so upgrading your graphics card won't impact performance in that rendering technique at all. For PBR, here is a graph comparing the performance of a bunch of different hardware including the 4080 and 4090.

https://www.chiefarchitect.com/blog/computer-for-design-and-gaming/

X16-Ray-Trace-Speeds.png

 

In case any Mac users want to know my benchmark...

 

I opened the saved camera "Kitchen1" in the Nashville kitchen plan which is already set to PBR.

 

On my 2023 MacBook Pro with 36GB memory, 12-core CPU, 18-core GPU and 16-core Neural Engine (apple marketing terms)

 

500 samples takes almost 11 minutes.

 

However!  this is what 50 samples looks like which only takes about 30 seconds...

 

image.thumb.png.822d6081022debfaf6c8bfdb5cc9c431.png

  • Upvote 2
Link to comment
Share on other sites

3 hours ago, Chief_Trent said:

RayTrace is only run on the CPU so upgrading your graphics card won't impact performance in that rendering technique at all. For PBR, here is a graph comparing the performance of a bunch of different hardware including the 4080 and 4090.

https://www.chiefarchitect.com/blog/computer-for-design-and-gaming/

X16-Ray-Trace-Speeds.png

You should add the 7900XTX right above that 4080

Link to comment
Share on other sites

3 hours ago, Paul454 said:

I love my Nvidia RTX 4090 Founders Edition!

I’m sure I’d love it as well. 
 

Could you take the time and perform the same 50 samples? 
 

I’m assuming you’d be less than 1 second perhaps?

Link to comment
Share on other sites

  • 2 weeks later...
On 8/23/2024 at 12:04 PM, Chief_Trent said:

RayTrace is only run on the CPU so upgrading your graphics card won't impact performance in that rendering technique at all. For PBR, here is a graph comparing the performance of a bunch of different hardware including the 4080 and 4090.

https://www.chiefarchitect.com/blog/computer-for-design-and-gaming/

X16-Ray-Trace-Speeds.png

So if I am reading this chart correctly. the 4080 doesn't look like its much different overall than the 4090. Or at least not something to brag about by sporting the 4090. Correct/

Link to comment
Share on other sites

The exact numbers for that benchmark between the 4080 and 4090 are that the 4080 got 120.4 samples per second and the 4090 had 156.8 samples per second. To render the view with 500 samples, that is about 3 seconds for the 4090 verses about 4 seconds for the 4080. Roughly a 25% performance increase for the 4090 above the 4080 in this case. There is a difference between the two, but it is a much smaller difference compared to other graphics card options. For example a 4070 in that benchmark got 70 samples per second.

 

(The testing benchmark was 500 samples in Chief Architect Premier X16 at the time of release using the “Nashville Kitchen”)

  • Like 1
Link to comment
Share on other sites

56 minutes ago, Chief_Trent said:

The exact numbers for that benchmark between the 4080 and 4090 are that the 4080 got 120.4 samples per second and the 4090 had 156.8 samples per second. To render the view with 500 samples, that is about 3 seconds for the 4090 verses about 4 seconds for the 4080. Roughly a 25% performance increase for the 4090 above the 4080 in this case. There is a difference between the two, but it is a much smaller difference compared to other graphics card options. For example a 4070 in that benchmark got 70 samples per second.

 

(The testing benchmark was 500 samples in Chief Architect Premier X16 at the time of release using the “Nashville Kitchen”)

It would be helpful to know if this was camera number 2 and what resolution for the rendering window. I have duplicate systems with one having a 7900XTX and the other a RTX 4090. I could give an apples to apples

Link to comment
Share on other sites

I believe the original benchmarks were Nashville, Kitchen 1 at 1400x860 (don't ask me why, we were profiling the Mac at the time and I'm pretty sure this was the resolution I got by default on an MBP).  Please bear in mind that the numbers Trent posted were from a specific build relatively early on in X16 and that exact samples/sec in the current version may have changed somewhat since then.

Link to comment
Share on other sites

14 minutes ago, Ryan-M said:

I believe the original benchmarks were Nashville, Kitchen 1 at 1400x860 (don't ask me why, we were profiling the Mac at the time and I'm pretty sure this was the resolution I got by default on an MBP).  Please bear in mind that the numbers Trent posted were from a specific build relatively early on in X16 and that exact samples/sec may in the current version may have changed somewhat since then.

Assuming this was 3/4 bounces and not "approximate.."
at that resolution i clocked 140 samples per second on the 7900 XTX and about the same as your reported value around 155 for the 4090 

Link to comment
Share on other sites

46 minutes ago, Renerabbitt said:

Assuming this was 3/4 bounces and not "approximate.."

 

Yes.

 

46 minutes ago, Renerabbitt said:

at that resolution i clocked 140 samples per second on the 7900 XTX and about the same as your reported value around 155 for the 4090 

 

This is good to know, we don't have that card in-house.

Link to comment
Share on other sites

  • 1 month later...

Is there anyone out there using a Mac Studio?

 

@Ryan-M or @Chief_Trent has anyone ran any tests with a big apple? It would be nice to see where it stacks up in the comparison chart for only a couple thousand more than a really good Mac laptop. I haven't ever done RT and am more interested in how much faster it will perform regular work in CA than RT rendering. It's a fair jump in price, but I'm guessing that it is a fair jump in performance as well.


Mac Studio M2 Ultra
24 Core CPU
76 Core GPU

32 Core Neural Engine

192 GB Unified Memory

2 TB SSD 

 

 

Link to comment
Share on other sites

19 minutes ago, ValleyGuy said:

Is there anyone out there using a Mac Studio?

 

@Ryan-M or @Chief_Trent has anyone ran any tests with a big apple? It would be nice to see where it stacks up in the comparison chart for only a couple thousand more than a really good Mac laptop. I haven't ever done RT and am more interested in how much faster it will perform regular work in CA than RT rendering. It's a fair jump in price, but I'm guessing that it is a fair jump in performance as well.


Mac Studio M2 Ultra
24 Core CPU
76 Core GPU

32 Core Neural Engine

192 GB Unified Memory

2 TB SSD 

 

 

You're not gonna see much in performance from an M3 to an M4 etc in anything but raytrace views. It will be  a jump, but nothing monumental. The rest is up to how you draw and resource management. Doesn't matter what machine you have, 4.million polys and a ton of vector lines will slow down any machine. I should know, I have two of the fastest machines one could build 

Link to comment
Share on other sites

2 hours ago, ValleyGuy said:

Is there anyone out there using a Mac Studio?

 

@Ryan-M or @Chief_Trent has anyone ran any tests with a big apple? It would be nice to see where it stacks up in the comparison chart for only a couple thousand more than a really good Mac laptop. I haven't ever done RT and am more interested in how much faster it will perform regular work in CA than RT rendering. It's a fair jump in price, but I'm guessing that it is a fair jump in performance as well.


Mac Studio M2 Ultra
24 Core CPU
76 Core GPU

32 Core Neural Engine

192 GB Unified Memory

2 TB SSD 

 

 

In the context of ray tracing, we have seen performance scale approximately linearly with core count when comparing the same tier of hardware.  A 12-core M2 will be about half as fast as a 24-core M2.  A 12-core M3 will be about half as fast as a 24-core M3.  There is a fairly significant jump from M2 to M3 (for example, a 12-core M3 is more than twice as fast as a 12-core M2) due to the introduction of dedicated ray tracing hardware.  The most powerful M2 Studio probably isn't going to perform better than the most powerful M3 MacBook Pro when it comes to ray tracing.

 

I can't speak as much to non-rendering performance as we don't really have similar benchmarks. It is comparatively simple to benchmark ray tracing performance in Chief. Outside of that, there are so many unique operations with different performance characteristics that it would be difficult to quantify in the same way.

Link to comment
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
 Share