Are there any "roundups" comparing and sumarizing peoples' experiences?
 
Notifications
Clear all

Are there any "roundups" comparing and sumarizing peoples' experiences?  

  RSS

zovc
 zovc
(@zovc)
Eminent Member
Joined: 9 months ago
 

Hey egpu.io,

 

I was wondering if there are any "roundup" pieces (articles, videos, charts, etc.) that attempt to compare and summarize egpu setups?

 

Before going to ask this, I glossed over the egpu enclosure's buyer's guide. I reread it more thoroughly just now and it's more helpful than I initially thought but it's still very general and leans away from showing numbers. (Not saying that as a criticism, just an observation. I think it was an intentional choice for ease of reading.) I do kind of feel like the Laptop buyer's guide is not especially helpful, though I trust its conclusions/recommendations are good ones. All I mean--if I sound harsh or critical--is I don't feel like it does a good job of explaining why a user would buy the ~$2K Razer Blade Stealth over a $800 Lenovo Yoga 700 Series, when using the chart provided it's 'worse.'

 

Basically, I think these pieces are fine for someone just wanting to receive a suggestion and go with it (and, again, I think that's their target) but I'm looking for something that's a bit more data driven and not necessarily focused on making a specific recommendation.

 

It seems like this site's catalog of builds and its search engine almost has enough data to automate or neatly organize a lot of the data I'd like to see, though I understand a lot of it is user-dependent and not uniformly documented/recorded. And user data is not necessarily controlled well.

 

Is there anything organized/written, comparing egpu performance where one can apply fixed filters like a specific CPU or series and a specific gpu/enclosure? For example, on the i7-1065G7, how does a 1660 TI compare to a 2060, to a 3060, etc..., and how do those numbers compare to desktop numbers?

 

I feel like this would be an extremely valuable reference for making a purchasing decision, and determining what's the right GPU, and also help people have a better idea of what kind of performance is fair to expect. "80% at best," does seem like a reasonably fair summary, based on my limited comparing, it's just not as thorough or specific as I'd like when considering the investment an eGPU/laptop setup would be.

 

If this doesn't exist, would anyone be interested in it getting made? It's something I don't mind putting the effort into creating, but I would appreciate input and help from more informed folks along the way if I am going to do it. And, it'd be nice to know other people think it'd be useful, too.

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

.

ReplyQuote
Eightarmedpet
(@eightarmedpet)
Noble Member Moderator
Joined: 4 years ago
 

I think I get what you are getting at, although you do start of complaining about articles being too numbers lead then go on to asking for more "data"... 😉 

misread your opener (I blame it on being late at night)

 

The problem is so much is subjective and on a case by case basis. For instance (my original use case) if I went on a "is it worth it by the numbers" I would have pretty much fitted into the "hell no" category. My original set up was probably the most expensive possible with the least performance per £. It did however suit me perfectly (and was all a business expense so not out of my pocket, kinda).

 

The other part of what you are asking is something you just have in your head through reading lots of stuff, it would be good to have some sort of amalgamation of experiences and data where you can select a multitude of options to determine your ideal/advised set up (one source I remover is itsage s testing of 2080ti vs 2080 and concluding the extra money isn't worth it due to TB3 bottlenecking) but that would be one hell of a job. The proper mods on here have day jobs so doubt they have time for that but you could give it a bash yourself?

This post was modified 2 months ago

2017 13" MacBook Pro Touch Bar
GTX1060 + AKiTiO Thunder3 + Win10
GTX1070 + Sonnet Breakaway Box + Win10
GTX1070 + Razer Core V1 + Win10
Vega 56 + Razer Core V1 + macOS + Win10
Vega 56 + Mantiz Venus + macOS + W10

---

LG 5K Ultrafine flickering issue fix

 
2017 13" MacBook Pro [7th,2C,U] + RX 5700 XT @ 32Gbps-TB3 (Mantiz Venus) + macOS 10.15.4 & Win10 2004 [build link]  


ReplyQuote
mac_editor
(@mac_editor)
Famed Member Moderator
Joined: 4 years ago
 

"eGPU performance" is difficult to round up because its behavior varies significantly across various setups even though it may be "expected" to have behaved similarly. On top of that, there exist multiple ways to interface with eGPUs, including proprietary solutions such as with Alienware. PC performance is much simpler, with GPU interface being a standard denominator - PCIe x16 - which is ubiquitous. Yet no one compares PC 'A' to PC 'B', instead typically comparing CPUs or GPUs with other factors constant. On the other hand, Thunderbolt alone is not implemented in the same ways across laptops, with implementation also varying due to Intel processor generations, so the ultrabook article mentions key features to ensure you are on the right track for a good eGPU-capable laptop.

Here are some examples of why Thunderbolt can be so messy:

https://egpu.io/forums/builds/2015-15-dell-precision-7510-q-m1000m-6th4ch-gtx-1080-ti-32gbps-m2-adt-link-r43sg-win10-1803-nando4-compared-to-tb3-performance/

https://egpu.io/forums/thunderbolt-enclosures/a-call-for-measurements-isolating-the-thunderbolt-effect/paged/17/#post-14987

purge-wranglertbt-flashpurge-nvdaset-eGPU
Insights Into macOS Video Editing Performance

Master Threads:
2014 15-inch MacBook Pro 750M
2018 15-inch MacBook Pro

 
2019 13" MacBook Pro [8th,4C,U] + RX Vega 64 @ 32Gbps-TB3 (Mantiz Venus) + macOS 10.14.6 & Win10 [build link]  


itsage and zovc liked
ReplyQuote
zovc
 zovc
(@zovc)
Eminent Member
Joined: 9 months ago
 
Posted by: @eightarmedpet

I think I get what you are getting at, although you do start of complaining about articles being too numbers lead then go on to asking for more "data"... 😉

I don't recall complaining about things being "too numbers." I did mention I think that some of the numbers chosen to be included  seem to work against the conclusions.

Posted by: @eightarmedpet

The problem is so much is subjective and on a case by case basis. For instance (my original use case) if I went on a "is it worth it by the numbers" I would have pretty much fitted into the "hell no" category. My original set up was probably the most expensive possible with the least performance per £. It did however suit me perfectly (and was all a business expense so not out of my pocket, kinda).

This is exactly my point, actually. The Buyer's Guides here work well enough as guides for making a choice if you've decided to buy something and don't want to look at numbers.

I'm sure I'm not the only person who wants to look at numbers before deciding if I want to buy something. For what it's worth, I'm considering budgeting as much as is necessary to get the best eGPU configuration, if that's a $2000 Razer laptop with a $500 eGPU enclosure and a $1000+ GPU... but from what I can tell there's not a lot of consistent documentation on if that actually makes sense to do, compared to spending ~300 on a Chromebook with a decent build and a $2K desktop and just using Parsec.

 

I appreciate that there is a difference between these things, and I'm glad they both exist as options. I just want (and believe it'd be valuable for) there to be a piece that is more thorough than a "buyer's guide," that does its best to compare all of the options and show the numbers. Not make a suggestion or declare what "is" and "isn't" worth it.

 

Just, give me a chart that demonstrates "a 1660Ti with this laptop gets you this. A 3080 with that same laptop gets you like 30% more performance" with actual data. (FPS, benchmarks...) If I decide that 30% extra performance is worth a bajillion more dollars, more power to me. That's basically what I'm getting at.

 

Posted by: @mac_editor

"eGPU performance" is difficult to round up because its behavior varies significantly across various setups even though it may be "expected" to have behaved similarly. On top of that, there exist multiple ways to interface with eGPUs, including proprietary solutions such as with Alienware. PC performance is much simpler, with GPU interface being a standard denominator - PCIe x16 - which is ubiquitous.

This is one reason I feel like I couldn't do a good job going in on something like this blind and without support. I didn't even know Alienware had their proprietary thing, which is both interesting and frustrating that something outside of TB that's similar in simplicity exists and is proprietary.

 

Thanks for linking those examples! Those are both pretty insightful and interesting.

 

I'm guessing this isn't nodded to or mentioned anywhere, but is there any hope that TB4's implementation will be 'better' and/or more uniform?

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

.

ReplyQuote
Eightarmedpet
(@eightarmedpet)
Noble Member Moderator
Joined: 4 years ago
 
Posted by: @mac_editor
Posted by: @mac_editor

"eGPU performance" is difficult to round up because its behavior varies significantly across various setups even though it may be "expected" to have behaved similarly. On top of that, there exist multiple ways to interface with eGPUs, including proprietary solutions such as with Alienware. PC performance is much simpler, with GPU interface being a standard denominator - PCIe x16 - which is ubiquitous.

I think this is the crux of it, there can never be a simple chart showing A vs B because there are too many variables - host cpu, host connection, how connection is implemented (straight to cpu or not), OS, software, loop back vs non loop back and then GPU. 
If you have a couple of GPUs yourself you could start trying to do a comparison yourself which could go on to form a template others could follow?

2017 13" MacBook Pro Touch Bar
GTX1060 + AKiTiO Thunder3 + Win10
GTX1070 + Sonnet Breakaway Box + Win10
GTX1070 + Razer Core V1 + Win10
Vega 56 + Razer Core V1 + macOS + Win10
Vega 56 + Mantiz Venus + macOS + W10

---

LG 5K Ultrafine flickering issue fix

 
2017 13" MacBook Pro [7th,2C,U] + RX 5700 XT @ 32Gbps-TB3 (Mantiz Venus) + macOS 10.15.4 & Win10 2004 [build link]  


zovc liked
ReplyQuote
zovc
 zovc
(@zovc)
Eminent Member
Joined: 9 months ago
 

I don't have access to any GPUs at the moment, and I don't have the kind of budget to blow on a project like this. I could try reaching out to reps in my country asking for samples, but having worked as a video editor for a tech YouTuber here, I can say it's very unlikely they would offer products for something like this. Here, it's basically, "You can tell everyone it's great and then send it back, thanks." Haha

 

But, I think I get the point you're trying to make and I was trying to allude to it in the OP with the difficulty of expecting user-submitted data to be controlled in any way. When most everyone here has different configurations, it's hard to make any conclusive or specific statements about things--I get that.

 

An ideal circumstance would be doing a bunch of tests while controlling everything except one variable, (Say, specific GPU inside of a fixed TB3 enclosure, same TB cable, same laptop) and doing a bunch of tests. Then, repeating that with a few different circumstances. For example, all of those same tests with a different laptop connected to the eGPU enclosure with all those different GPUs... I have done this kind of work before (benchmarking CPU/GPU combos on desktop) and it's tedious and takes a lot of time so I don't suggest it as a trivial task. Just trying to outline what I imagine would be best.

 

If people would be willing to loan hardware, I'd be open to doing something like this, but I suspect my location would make shipping impractical and expensive. (Never mind the trust involved.) Up until now, I never thought about it, but I can try networking locally but I don't suspect it'll go very far.

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

.

ReplyQuote
mac_editor
(@mac_editor)
Famed Member Moderator
Joined: 4 years ago
 
Posted by: @zovc

An ideal circumstance would be doing a bunch of tests while controlling everything except one variable, (Say, specific GPU inside of a fixed TB3 enclosure, same TB cable , same laptop) and doing a bunch of tests. Then, repeating that with a few different circumstances.

This is still harder to do on eGPUs because even one type of workload - games, can behave very differently on eGPUs vs. PC. Example: Forza Horizon 4. Changing just the GPU to see what the differences are honestly wouldn't help too much. You are more likely to find many comments across the web and especially on the eGPU subreddit, which mostly has misinformed people, saying stuff like "RTX 3080 is overkill for eGPU - something like a 2070S is a sweet spot" yada yada but that's not true. The answer is always "it depends" (on your specific workload). For example, if you were looking for Forza Horizon 4 performance specifically, you'll find a nice thread detailing info about that on this forum (its heavily bottlenecked on TB3). But you can't take that info as face value for other games. Benchmarks, etc. provide an approximation of performance to "expect" - its not safe to make this assumption with eGPUs.

I do understand your desire for more insight/data points though. Hope you can find a mix of things on this forum that can somewhat guide you in the right direction. Oh and if you have specific questions, I'm sure the forum members and I would be happy to help.

purge-wranglertbt-flashpurge-nvdaset-eGPU
Insights Into macOS Video Editing Performance

Master Threads:
2014 15-inch MacBook Pro 750M
2018 15-inch MacBook Pro

 
2019 13" MacBook Pro [8th,4C,U] + RX Vega 64 @ 32Gbps-TB3 (Mantiz Venus) + macOS 10.14.6 & Win10 [build link]  


zovc liked
ReplyQuote
zovc
 zovc
(@zovc)
Eminent Member
Joined: 9 months ago
 

That makes sense, and I'm not surprised to hear it. But that makes me feel like this makes the 'roundup' thing I'm talking about even more valuable. That post is a pretty good disclaimer to lead all of the information with, along with the admission that one person (or one team) probably can't feasibly test everything.

 

I still believe it would be valuable to find the exceptions like this, as well as the 'rules' among more common things (assuming the large majority of games play a certain way with eGPUs) and outline those. Forza could be use as an example of "Here's an instance where PCI-e lanes/bandwidth matter a lot, for some reason." The investigative journalist in me wonders if we could ask the studio if they have any theories or insight as to why this might be the case.

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

.

ReplyQuote