2018 17" Lenovo ThinkPad P72 [8th,6C,H] + GTX 1080 Ti @ 32Gbps-TB3 (Razer Core X...
 
Notifications
Clear all

2018 17" Lenovo ThinkPad P72 [8th,6C,H] + GTX 1080 Ti @ 32Gbps-TB3 (Razer Core X) + Win10 [bobbie424242] // Detailed 4K benchmarks comparison vs my i7-8700K gaming PC  

 of  2
  RSS

OliverB
(@oliverb)
Noble Member
Joined: 2 years ago
 
Posted by: @bobbie424242

@oliverb

I don't need to make anymore game benchmarks to know that the eGPU incurs a massive hit, even at 4K. I mean, look again at the results I posted: ACO: 49fps vs 34, FH4 demo: 80fps vs 51, KCD: 39fps vs 26.
I also monitored the laptop's TB3 PCH  temperature in case it would overheat and degrade performance but it stayed at a normal 70 degrees Celsius.

@bobbie424242
If you want to verify your claims, some benchmarks for the same games and the same settings would help a lot. I don't understand such abbreviations, I take my time to write out titles.
Of course, monitoring TB3 PCH temperature doesn't help a lot, in this matter I am talking about a notebook with a direct connection between TB3 and CPU like my 15-inch MBP 2018.

So, I may precise my statement: If you use a good notebook with a direct link between TB3 and CPU, you won't have a notable performance loss in 4k, or even better in 5k gaming. Notebooks where the TB3 data goes over PCH are not recommended, if you want to have good performance. 

EDIT: Of course you are probably correct, that in the setup/notebook from this thread a Desktop solution is notably better then a eGPU solution, but this may not be the case generally. It is my mistake, that I didn't considered it. I wasn't aware that the difference is this big.

This post was modified 12 months ago

2018 15" MBP & 2015 13" MBP connected to RTX2080Ti GTX1080Ti GTX1080 Vega56 RX580 R9-290 GTX680

 
2018 15" MacBook Pro (RP560X) [8th,6C,H] + RX 5700 XT @ 32Gbps-TB3 (ASUS XG Station Pro) + Win10 & macOS 10.15.4 [build link]  


ReplyQuote
bobbie424242
(@bobbie424242)
Active Member
Joined: 1 year ago
 
Posted by: @psonice

Some general thoughts on this (I'm a graphics programmer, not a gamer):

- The system difference might have more of an effect than you expect. There's going to be non-rendering work to do which will be similar on both systems, and the remaining time gets spent rendering. The slightly faster system can end up with a lot more time left, meaning more frames can be rendered.

- RAM will make a difference too - probably not with geek bench, but a game might use >16GB, and each time it has to wait for disk (even with a fast SSD) it's going to stall things. 

Basically, you want to plug the eGPU into the desktop for a proper comparison, or the results won't mean much.

- The reason why the performance drop is much lower at 4k or 5k: the only performance difference with an eGPU over an internal GPU is bandwidth. In a game, the CPU has to copy a bunch of stuff to the GPU for every frame, and depending on the game architecture the GPU might stall waiting for it or the game code might have to wait for the GPU work to complete, which takes longer if it's waiting for that data to transfer. More frames per second means more waiting on that slower thunderbolt link.

So if you crank the resolution up, you get a lower frame rate, and the lower frame rate means fewer transfers and less bandwidth hit. So less performance drop on the eGPU.

1. I don't think the system (CPU and RAM) made too much of a difference: most benchmarked games where using the GPU at close to 100% indicating that the GPU is the limiting factor. Also the laptop CPU did not throttle (I looked at that) as it is in a 17" chassis, well cooled.

2. RAM is nothing spectacular on the PC: it is DDR4 2133 (and not 2666 as I wrongly mentioned). Although there was 32GB on the PC but I really doubt any game went over 16GB

3. The results are still interesting I think even without comparing the eGPU connected to the PC. At least it is interesting to me in my decision to keep my gaming PC or not.

4. Yes, the performance loss is game dependent but I did not expect to be this important (again: for the games I tested).

In any case, I really wish there were other PC vs eGPU comparison at high resolutions (4k and eventually 5K).

@oliverb

Macbooks have notoriously the best TB3 connection possible. So it is very well possible it would be faster than my P72 (assuming same CPU). I don't think there is a PC laptop with a TB3 connection as efficient as MacBooks. Maybe Alienware laptops with the amplifier port but that is something else.

This post was modified 12 months ago
2018 17" Lenovo ThinkPad P72 [8th,6C,H] + GTX 1080 Ti @ 32Gbps-TB3 (Razer Core X) + Win10 [build link]  

ReplyQuote
OliverB
(@oliverb)
Noble Member
Joined: 2 years ago
 
Posted by: @bobbie424242

@oliverb

Macbooks have notoriously the best TB3 connection possible. So it is very well possible it would be faster than my P72 (assuming same CPU). I don't think there is a PC laptop with a TB3 connection as efficient as MacBooks. Maybe Alienware laptops with the amplifier port but that is something else.

@bobbie424242
It was my mistake, I didn't consider this difference.
Yes, I hear that the amplifier port is clearly better than TB3.

2018 15" MBP & 2015 13" MBP connected to RTX2080Ti GTX1080Ti GTX1080 Vega56 RX580 R9-290 GTX680

 
2018 15" MacBook Pro (RP560X) [8th,6C,H] + RX 5700 XT @ 32Gbps-TB3 (ASUS XG Station Pro) + Win10 & macOS 10.15.4 [build link]  


ReplyQuote
psonice
(@psonice)
Estimable Member
Joined: 3 years ago
 
Posted by: @bobbie424242

1. I don't think the system (CPU and RAM) made too much of a difference: most benchmarked games where using the GPU at close to 100% indicating that the GPU is the limiting factor. Also the laptop CPU did not throttle (I looked at that) as it is in a 17" chassis, well cooled.

You won't be able to check how busy the GPU is like that. Yes, it's 100% busy... but doing what? Transferring data over the bus, or rendering? Waiting for data or sync? Honestly, this really is a pretty meaningless number unless you have a proper GPU trace and can see what it's actually doing.

For some idea of what I mean, I recently made a rendering pipeline ~10% faster. It was "100% busy" according to basic tools at the start, and the same at the end, and the workload didn't change at all! But looking closely at a full GPU trace I could see that part of the time the GPU wasn't actually rendering, it was transferring data or waiting for other work. Just changing the way things are scheduled can have a big difference.

With an eGPU the added latency and lower bandwidth your GPU might be 100% 'busy', but only actually rendering stuff for 10% of the time! 

So trying to compare without the same CPU and memory isn't going to tell you much at all.

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote
bobbie424242
(@bobbie424242)
Active Member
Joined: 1 year ago
 
Posted by: @psonice
Posted by: @bobbie424242

1. I don't think the system (CPU and RAM) made too much of a difference: most benchmarked games where using the GPU at close to 100% indicating that the GPU is the limiting factor. Also the laptop CPU did not throttle (I looked at that) as it is in a 17" chassis, well cooled.

You won't be able to check how busy the GPU is like that. Yes, it's 100% busy... but doing what? Transferring data over the bus, or rendering? Waiting for data or sync? Honestly, this really is a pretty meaningless number unless you have a proper GPU trace and can see what it's actually doing.

For some idea of what I mean, I recently made a rendering pipeline ~10% faster. It was "100% busy" according to basic tools at the start, and the same at the end, and the workload didn't change at all! But looking closely at a full GPU trace I could see that part of the time the GPU wasn't actually rendering, it was transferring data or waiting for other work. Just changing the way things are scheduled can have a big difference.

With an eGPU the added latency and lower bandwidth your GPU might be 100% 'busy', but only actually rendering stuff for 10% of the time! 

So trying to compare without the same CPU and memory isn't going to tell you much at all.

If the GPU is at 100%, of which 50% are only used rendering, it doesn't matter much as it is at the maximum capacity on this setup, possibly bottlenecked by TB3: it cannot do more. At least this is how I understand it.

"So trying to compare without the same CPU and memory isn't going to tell you much at all."

Except I wanted to evaluate replacing my PC with my laptop+eGPU, not switching my PC to using an eGPU.
Sure, benchmarks of PC + eGPU would be interesting to tell the full story (to better understand possible bottlenecks) but not practically useful as nobody is going to ever use a PC + eGPU if the GPU fits inside the PC case.

This post was modified 12 months ago
2018 17" Lenovo ThinkPad P72 [8th,6C,H] + GTX 1080 Ti @ 32Gbps-TB3 (Razer Core X) + Win10 [build link]  

ReplyQuote
psonice
(@psonice)
Estimable Member
Joined: 3 years ago
 

What I mean: the GPU might be at 100% load on both machines, but doing 30fps on one machine and 60 on the other. It's clearly not "100% busy" Smile You need some pretty serious dev tools (and enough knowledge to interpret the graphs) to know what's actually going on, but it's not as simple as "the thunderbolt bus is slowing it down". That will be a factor, but different CPU speeds will have an effect too, and it's likely the different RAM amount will too.

Basically, it's complicated, and without comparing on the same system the numbers won't mean much Smile

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote
bobbie424242
(@bobbie424242)
Active Member
Joined: 1 year ago
 

@psonice

Ok, thanks for clarifying. Sure nothing is really simple on these complex topic.

Unfortunately my PC does not have TB3. Otherwise I would run PC + eGPU benchmark just for science :).

2018 17" Lenovo ThinkPad P72 [8th,6C,H] + GTX 1080 Ti @ 32Gbps-TB3 (Razer Core X) + Win10 [build link]  

ReplyQuote
ritterbutzke
(@ritterbutzke)
Eminent Member
Joined: 3 years ago
 

@bobbie424242

If you want near Desktop performance how about an m.2 egpu? You have a free m.2 pcie x4 slot in your P72.

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote
 of  2