Hi Guys,
I have 4 1070 that i would like to dedicate to rendering purpose with octane. I would need some advise to avoid doing anything stupid.
1 - what would give the best perf beetween : 1 PC with 4 * 1070 on one Motherboard vs 4 PC with 1 * 1070 + network rendering ?
From octane bench it looks that 4 PC with 1 *1070 gives slightly better perf :
1 pc + 4 * 1070 -> 512
(1 pc + 1 *1070) * 4 -> 526
2 - if the best road is 1 PC with 4 GPU i would need the highest number of PCI-E lanes at the best price, does threadripper 1920X and socket x399 sounds wise, are there any better options ?
3 - i think all calculation is done under octane vram so is there any difference beetween having 32GB of RAM vs 128GB RAM in term of perfs ?
Thanks for your lights
Cheers
E
Advise to build a simple Octane Render PC.
Forum rules
Please add your OS and Hardware Configuration in your signature, it makes it easier for us to help you analyze problems. Example: Win 7 64 | Geforce GTX680 | i7 3770 | 16GB
Please add your OS and Hardware Configuration in your signature, it makes it easier for us to help you analyze problems. Example: Win 7 64 | Geforce GTX680 | i7 3770 | 16GB
good day,
Single computer, with enough airflow and fast CPU will always give a better value and also be faster. Let me explain why:
Let's start from value. in order to have four computers running You would have a need for for sets of everything (4 CPUs, 4 sets of RAM, etc). That's a lot of wasted resources.. put half of those parts into single system and it would be twice more capable..
So what about speed? Quad GPU machine might suffer in few aspects: Temperature (that thermal throttles and clock down certain elements including GPUs and CPU) or stability issues due to complexity of the system (however I would not worry about that on quad GPU rigs, as guys are building 7,8 ..11 or 17 card builds)
The speed would get eroded much more in case of four systems, since Your LAN is going to be the biggest bottleneck and depending on it's actual speed Your GPUs will have gaps (with 0% activity) in utilization. Having 10G or faster is not practical from cost perspective (that's 100$ for 10G card + network gear that is easily 500+) ..
So all in all, just because You can have performance of 4x systems nearly matching 1 system on paper, that's only theoretical maximum - in real world I would say You can loose up to 30% if You would end up with lower end hardware, slower lan and such.
So to answer Your questions:
1. it could be faster or slower, but that depends on airflow and single core CPU speed and network.
2. Lanes matter little, I would advice not to go under 8x per GPU, as You will start to notice negative effects - however more important here is fast CPU (quad core is more than enough, just take the one that performs - latest SKUs could go to 5GHz+ )
3. as for RAM, that matters, but in Your case 32 at most 64GB would be more than enough on a build with 8GB VRAM cards. The thing with RAM, is that it is heavily used in preparation stage and then image is actually being constructed in RAM as You render it, so if You plan rendering high resolution and also heavily pushing those GPUs to their limits + relying on out of core, stepping over 32GB is not unreasonable & You would better have a motherboard with some empty RAM slots to upgrade.
Last but not least..keep in mind that Out Of core will slow down render process, so if You plan to do heavier scenes and do not care about optimizations, then something like 2x 2080ti (with 11GB of VRAM) would not only be faster side by side (600 in OvtaneBench), but also suffer less with scenes that would go over 8 GB or even 11GB, since new cards have NVLINK (in a simple way, faster interface for out of core, so it would effect performance less).
In Your case, I would say that something like Z390WS PRO + 8700k or 9gen CPU + 32GB RAM (2x 16GB up-gradable to 64GB) and good cooler + case with like 1000W PSU would be wise idea with good value and maximum performance. You can go X299 and more cores on Intel or X399 + AMD..but for extra price You would gain nothing. Parts I just mentioned sit in my personal system with quad GPU setup and I have AMD system nearby + other builds to compare against as well. From value perspective, I do not see any better deal. The only worry I have is that PLX chip equipped motherboards had some issues (that was the main reason I bought it - to test =) with Nvidia drivers on X99 platform a while ago, but ..so far, from testing I could not see this problem present (and head that it's also not evident on X299 systems with PLX chips), so maybe it's solved (finally).
If You would still have some questions, feel free to ask.
Single computer, with enough airflow and fast CPU will always give a better value and also be faster. Let me explain why:
Let's start from value. in order to have four computers running You would have a need for for sets of everything (4 CPUs, 4 sets of RAM, etc). That's a lot of wasted resources.. put half of those parts into single system and it would be twice more capable..
So what about speed? Quad GPU machine might suffer in few aspects: Temperature (that thermal throttles and clock down certain elements including GPUs and CPU) or stability issues due to complexity of the system (however I would not worry about that on quad GPU rigs, as guys are building 7,8 ..11 or 17 card builds)
The speed would get eroded much more in case of four systems, since Your LAN is going to be the biggest bottleneck and depending on it's actual speed Your GPUs will have gaps (with 0% activity) in utilization. Having 10G or faster is not practical from cost perspective (that's 100$ for 10G card + network gear that is easily 500+) ..
So all in all, just because You can have performance of 4x systems nearly matching 1 system on paper, that's only theoretical maximum - in real world I would say You can loose up to 30% if You would end up with lower end hardware, slower lan and such.
So to answer Your questions:
1. it could be faster or slower, but that depends on airflow and single core CPU speed and network.
2. Lanes matter little, I would advice not to go under 8x per GPU, as You will start to notice negative effects - however more important here is fast CPU (quad core is more than enough, just take the one that performs - latest SKUs could go to 5GHz+ )
3. as for RAM, that matters, but in Your case 32 at most 64GB would be more than enough on a build with 8GB VRAM cards. The thing with RAM, is that it is heavily used in preparation stage and then image is actually being constructed in RAM as You render it, so if You plan rendering high resolution and also heavily pushing those GPUs to their limits + relying on out of core, stepping over 32GB is not unreasonable & You would better have a motherboard with some empty RAM slots to upgrade.
Last but not least..keep in mind that Out Of core will slow down render process, so if You plan to do heavier scenes and do not care about optimizations, then something like 2x 2080ti (with 11GB of VRAM) would not only be faster side by side (600 in OvtaneBench), but also suffer less with scenes that would go over 8 GB or even 11GB, since new cards have NVLINK (in a simple way, faster interface for out of core, so it would effect performance less).
In Your case, I would say that something like Z390WS PRO + 8700k or 9gen CPU + 32GB RAM (2x 16GB up-gradable to 64GB) and good cooler + case with like 1000W PSU would be wise idea with good value and maximum performance. You can go X299 and more cores on Intel or X399 + AMD..but for extra price You would gain nothing. Parts I just mentioned sit in my personal system with quad GPU setup and I have AMD system nearby + other builds to compare against as well. From value perspective, I do not see any better deal. The only worry I have is that PLX chip equipped motherboards had some issues (that was the main reason I bought it - to test =) with Nvidia drivers on X99 platform a while ago, but ..so far, from testing I could not see this problem present (and head that it's also not evident on X299 systems with PLX chips), so maybe it's solved (finally).
If You would still have some questions, feel free to ask.
Wahou ! Thanks for your very detail answer glimpse !
Lot of great info here. So 1 Machine with 4 GPU and 64 GB of ram sound like a good starting point.
From what you say it looks that Octane like some other apps , run better with [ Higher High Clock / Lower Number of Core ] CPU than with [ Lower Clock / High Number of Core ].
So basically a 4.7 GHz with 6 Cores will perform better in octane than a 4.0 Ghz with 16 cores am i correct ?
The hardest part for me now that i know that i need one machine with a 4 slot Motherboard is to find the best chipset ...
- x470 -> rysen 2700x -> this one doesn't look to handle more than 3 GPU so it is already disqualify
- x399 -> TR 1920X -> sounds like a candidate but it's 4Ghz max so less than the intel ones but prices is good at around 750euros it also gives you 128Gb ram max
- z390 -> 9700K -> with 4.9Ghz sound like a good options but price is really high, you have to spend 1000euros just for the CPU+MoBo you also have 64Gb ram max
- x299 -> 7820x -> sounds more expensive and run at lower clock but they offer 128 Gb ram max
- x99 -> xeon -> sound really not interesting because xeon cpu runs at a really lower clock and shine mostly for CPU rendering task
So i'm quite tear appart for choosing the right chipset, i would exclude x470 cause it looks it can't handle 4 GPU, i would exclude x99 because xeon cpu run at lower clock
but between x399 / z390 / x299 it's a hard choice. As i will also use the machine for heavy fluid calculation and the possibility to upgrade to 128 GB of ram with x399 and x299 sounds like a good point.
but performance / price wise x299 sounds like a bad option.
so i guess the choice will be between :
x399 + 1920X : lower clock , lower perf in octane but better price / perf ratio and possibility to upgrade to 128Gb
z390 + 9700K : higher clock , best perf in octane but higher price and 64Gb max
Does it sound correct to you ?
Thanks again for your time.
Cheers
E
Lot of great info here. So 1 Machine with 4 GPU and 64 GB of ram sound like a good starting point.
From what you say it looks that Octane like some other apps , run better with [ Higher High Clock / Lower Number of Core ] CPU than with [ Lower Clock / High Number of Core ].
So basically a 4.7 GHz with 6 Cores will perform better in octane than a 4.0 Ghz with 16 cores am i correct ?
The hardest part for me now that i know that i need one machine with a 4 slot Motherboard is to find the best chipset ...
- x470 -> rysen 2700x -> this one doesn't look to handle more than 3 GPU so it is already disqualify
- x399 -> TR 1920X -> sounds like a candidate but it's 4Ghz max so less than the intel ones but prices is good at around 750euros it also gives you 128Gb ram max
- z390 -> 9700K -> with 4.9Ghz sound like a good options but price is really high, you have to spend 1000euros just for the CPU+MoBo you also have 64Gb ram max
- x299 -> 7820x -> sounds more expensive and run at lower clock but they offer 128 Gb ram max
- x99 -> xeon -> sound really not interesting because xeon cpu runs at a really lower clock and shine mostly for CPU rendering task
So i'm quite tear appart for choosing the right chipset, i would exclude x470 cause it looks it can't handle 4 GPU, i would exclude x99 because xeon cpu run at lower clock
but between x399 / z390 / x299 it's a hard choice. As i will also use the machine for heavy fluid calculation and the possibility to upgrade to 128 GB of ram with x399 and x299 sounds like a good point.
but performance / price wise x299 sounds like a bad option.
so i guess the choice will be between :
x399 + 1920X : lower clock , lower perf in octane but better price / perf ratio and possibility to upgrade to 128Gb
z390 + 9700K : higher clock , best perf in octane but higher price and 64Gb max
Does it sound correct to you ?
Thanks again for your time.
Cheers
E
perfectly, You summed things up pretty clear. And that's the main reason why I own Z390 + 8700k and X399 + 2950X - both systems has slightly different usage.Does it sound correct to you ?
that's the whole puzzle, it's really personal based on ones needs and preferences. Depending on what You need more, You would end up leaning one or the other side..So i'm quite tear appart for choosing the right chipset,
Few interesting notes:
* core clock should not be compared side by side since there are IPC and other bits (like turbo stages), but in general the newer the better..
* compatibility (You should be a bit more careful choosing components, like ram for instance, to AMD system)
* availability (seems that we have less options for higher end X399 boards for instance)
* new stuff..like, not so long ago 32GB DDR4 RAM modules were announced (but at least for now seems to be compatible only with very few boards)
If You're planing to do any kind of simulation, etc. Threadripper based system offers great value You can get CPU for 700-900 EU that would give You like 3000 in Cinebench - that's really good result (even if motherboards will probably be a bit more expensive, that would not hurt the value too much). Just tip from personal experience - get a solid board and do not cut any corners with X399.
Thanks again for your Help Glimpse.
In fact i will use this PC to also run big simulation in Houdini. Generally Lower Core but with higher frequency tend to be better. But the ram is the most crucial factor imo for my needs.
While 64GB sound good enough for 99% of users, fluids sims / heavy geo computation in Houdini tend to eat a lot of Ram.
So my choice would be more oriented on 1920X or 1950X + x399. BUT like you mention i did find
- very few mobo that get very positive feedback even MSI / Asus / Gygabite Mobo have bad user reviews
- lot of people face problems of incompatible ram
- lot of people talk about the need of updating bios
This sound like a too young new chipset , with half finish work by AMD, and a not so bulletproof reliability, unlike the good old 1151 socket.
But for the extra 128GB i think i will go that way.
One of the only Mobo i have found that get only positive feedback and good price perf is this one
https://www.asrock.com/MB/AMD/X399%20Taichi/index.asp
The asus looks great
https://www.asus.com/fr/Motherboards/RO ... H-EXTREME/
But the price is nearly 250 euros more expensive, for feature i will not use, as i won't do any OC.
I will not rush and study carefully the test and review before making my final choice.
Thanks again for the very detail info.
In fact i will use this PC to also run big simulation in Houdini. Generally Lower Core but with higher frequency tend to be better. But the ram is the most crucial factor imo for my needs.
While 64GB sound good enough for 99% of users, fluids sims / heavy geo computation in Houdini tend to eat a lot of Ram.
So my choice would be more oriented on 1920X or 1950X + x399. BUT like you mention i did find
- very few mobo that get very positive feedback even MSI / Asus / Gygabite Mobo have bad user reviews
- lot of people face problems of incompatible ram
- lot of people talk about the need of updating bios
This sound like a too young new chipset , with half finish work by AMD, and a not so bulletproof reliability, unlike the good old 1151 socket.
But for the extra 128GB i think i will go that way.
One of the only Mobo i have found that get only positive feedback and good price perf is this one
https://www.asrock.com/MB/AMD/X399%20Taichi/index.asp
The asus looks great
https://www.asus.com/fr/Motherboards/RO ... H-EXTREME/
But the price is nearly 250 euros more expensive, for feature i will not use, as i won't do any OC.
I will not rush and study carefully the test and review before making my final choice.
Thanks again for the very detail info.
hi glimpse sorry to ressurect this old topic,
but there is one thing i can't figure out , you tell me that i can safely use Z390WS PRO + 8700k for a good octane 4 GPU config.
it seems that i don't understand how pcie-lanes distribution work.
The 8700K only have 16PCIE lanes
https://ark.intel.com/content/www/fr/fr ... 0-ghz.html
Intel mention that you can have those distribution : 1x16, 2x8, 1x8+2x4
So except having 2*4 + 2*4 distribution it can't handle 4 GPU ? ( i'm sure something is wrong in my understanding of PCIE-lanes ...
)
I have also read that GPU could run on 16* or 8* lanes without performance reduction.
But when running on just *4 PCIE-lanes performance can be lowered from 10 to 25%.
So how asus succeed to make a 4 GPU mobo With a CPU that can only handle 16 lanes ?
https://www.asus.com/us/Commercial-Serv ... -Z390-PRO/
Thanks for your Lights !
Cheers
E
but there is one thing i can't figure out , you tell me that i can safely use Z390WS PRO + 8700k for a good octane 4 GPU config.
it seems that i don't understand how pcie-lanes distribution work.
The 8700K only have 16PCIE lanes
https://ark.intel.com/content/www/fr/fr ... 0-ghz.html
Intel mention that you can have those distribution : 1x16, 2x8, 1x8+2x4
So except having 2*4 + 2*4 distribution it can't handle 4 GPU ? ( i'm sure something is wrong in my understanding of PCIE-lanes ...
)I have also read that GPU could run on 16* or 8* lanes without performance reduction.
But when running on just *4 PCIE-lanes performance can be lowered from 10 to 25%.
So how asus succeed to make a 4 GPU mobo With a CPU that can only handle 16 lanes ?
https://www.asus.com/us/Commercial-Serv ... -Z390-PRO/
Thanks for your Lights !
Cheers
E
no problem, Guys. I'm here to help You.
The thing that allows 4 GPUs to be hooked on x8 eletricaly when CPU do not have enough lanes is called, PLX chip. You feed it with x16 and it spits out quad x8 =) HereYou can find more info about it and there's a link in that page to pdf that explains how it works.
in general these motherboards with PLX under the hood do cost a bit more, 'cos chip alone is like 50$ + implementations + extra features that are added onboard. But in most case it does worth the investment.
I have couple of systems with 8700k (one is mITX & other ATX) - stability wise they are both rock solid and there's no side effect for having that PLX now (we used to have some issues with older drivers though). Is it safe choice on the long run? I don't know, 'cos I can not talk for nvidia and other companies whos decisions do influence systems like this, but at least at the moment I see no problems (haven't had a single bsod since I assembled the system a month or two ago).
The thing that allows 4 GPUs to be hooked on x8 eletricaly when CPU do not have enough lanes is called, PLX chip. You feed it with x16 and it spits out quad x8 =) HereYou can find more info about it and there's a link in that page to pdf that explains how it works.
in general these motherboards with PLX under the hood do cost a bit more, 'cos chip alone is like 50$ + implementations + extra features that are added onboard. But in most case it does worth the investment.
I have couple of systems with 8700k (one is mITX & other ATX) - stability wise they are both rock solid and there's no side effect for having that PLX now (we used to have some issues with older drivers though). Is it safe choice on the long run? I don't know, 'cos I can not talk for nvidia and other companies whos decisions do influence systems like this, but at least at the moment I see no problems (haven't had a single bsod since I assembled the system a month or two ago).
Many thanks for your help Glimpse !
My understanding of PCIE-lanes was then correct, it is just that asus use some voodoo tricks on the motherboard to bypass the restriction thanks to PLX.
Well thanks again for all the info, i have finally set my choice and buy the beast.
I have finally pick
- Threadripper 1920X for just 350 euros -> i think it will be hard to beat this price / perf ratio
- Asrock X399 Taichi -> While i have some bad feeling with this brands, it looks that it is one of the best and most stable board for X399 and the price is really cheaper than asus top horse.
- Corsair 4*16Go 2666Mhz CL16 -> i didn't take the risk to go with 3000 or 3200 because i plan to have 128Go in the futur and i read in the QVL that the safer bet for this was to stick to 2666
I have my 1070, and was wondering if combining 2 *1070 + 2* 2070 would restrict the perf.
I'm quite sure it won't as all GPU runs with 8Gb of vram, and i also read that octane was very good to dispatch the computation to each GPU and use everything it can on each one.
But i just wanted to be sure to check that with a hardware guru !
Cheers
E
My understanding of PCIE-lanes was then correct, it is just that asus use some voodoo tricks on the motherboard to bypass the restriction thanks to PLX.
Well thanks again for all the info, i have finally set my choice and buy the beast.
I have finally pick
- Threadripper 1920X for just 350 euros -> i think it will be hard to beat this price / perf ratio
- Asrock X399 Taichi -> While i have some bad feeling with this brands, it looks that it is one of the best and most stable board for X399 and the price is really cheaper than asus top horse.
- Corsair 4*16Go 2666Mhz CL16 -> i didn't take the risk to go with 3000 or 3200 because i plan to have 128Go in the futur and i read in the QVL that the safer bet for this was to stick to 2666
I have my 1070, and was wondering if combining 2 *1070 + 2* 2070 would restrict the perf.
I'm quite sure it won't as all GPU runs with 8Gb of vram, and i also read that octane was very good to dispatch the computation to each GPU and use everything it can on each one.
But i just wanted to be sure to check that with a hardware guru !
Cheers
E
Hi I am really interested to know if you have build your machine with the mentioned components and what is your experience till today.
thanks a lot.
Luca
thanks a lot.
Luca
i9-10900x, 96GB DDR4, 2xRTX 2080 TI, ASUS X299 SAGE, Windows 10
http://www.visual4d.it
http://www.visual4d.it
Hi Luca,
i have build my machine with
- AMD TR 1920X
- Asrock X399 Taichi
- HyperX - HX426C16FBK4/64 - FURY DDR4 64Go (Kit 4x16Go)
- Corsair Carbide Air 540
- Corsair Hydro 115i RGB Platinum
- Corsair RM1000i
i am very happy with this config it works smoothly , the price / perf ratio is very good.
several important point :
- TR CPU display 30 degre celcius additional temperature in the bios. so if you see 65c it' s 35 in reality which is perfectly fine
- X399 chipset is sensitive on Ram be sure to take a memory stick that is approved by your mobo manufacturer
- Water cooling is not luxury if you plan to use 3 cards
- Beware of the GPU you pick you need to extract the heat outside ( backward ) the case not downside.
- the corsair case i choose is stellar because power unit is separated from the rest so no extra heat.
- buy all the fans you can to cool the beast
- if you want to enjoy nvlink to combine your GPU ram no choice you have take 2080 / 2080 Ti ( shame on nvidia that didin' t put any SLI connector one the 2070 !
)
i recommand MSI card with default cooling system they are silent and run around 78 degre at full load :
GOOD
https://www.amazon.fr/MSI-GEFORCE-RTX-2 ... ers&sr=1-6
BAD
https://www.amazon.fr/Carte-Graphique-M ... ers&sr=1-2
The Asrock X399 Taichi allow only 3 GPU because the case panel plug make contact with the last GPU card. you will need to do some tweaking to modify that. I have try to find a 90 degre adaptator on the net, but don' t find any.
if you are good with electronics it must be easy to create your own.
i have build my machine with
- AMD TR 1920X
- Asrock X399 Taichi
- HyperX - HX426C16FBK4/64 - FURY DDR4 64Go (Kit 4x16Go)
- Corsair Carbide Air 540
- Corsair Hydro 115i RGB Platinum
- Corsair RM1000i
i am very happy with this config it works smoothly , the price / perf ratio is very good.
several important point :
- TR CPU display 30 degre celcius additional temperature in the bios. so if you see 65c it' s 35 in reality which is perfectly fine
- X399 chipset is sensitive on Ram be sure to take a memory stick that is approved by your mobo manufacturer
- Water cooling is not luxury if you plan to use 3 cards
- Beware of the GPU you pick you need to extract the heat outside ( backward ) the case not downside.
- the corsair case i choose is stellar because power unit is separated from the rest so no extra heat.
- buy all the fans you can to cool the beast
- if you want to enjoy nvlink to combine your GPU ram no choice you have take 2080 / 2080 Ti ( shame on nvidia that didin' t put any SLI connector one the 2070 !
)i recommand MSI card with default cooling system they are silent and run around 78 degre at full load :
GOOD
https://www.amazon.fr/MSI-GEFORCE-RTX-2 ... ers&sr=1-6
BAD
https://www.amazon.fr/Carte-Graphique-M ... ers&sr=1-2
The Asrock X399 Taichi allow only 3 GPU because the case panel plug make contact with the last GPU card. you will need to do some tweaking to modify that. I have try to find a 90 degre adaptator on the net, but don' t find any.
if you are good with electronics it must be easy to create your own.