32 or 64 GB Ram for 11GB VRAM?
j1039564
Posts: 19
I work on HUGE secenes.
How much ram memory should I get?
I'm thinking about 32 or 64GB.
My GPU's can only do max 11GB VRAM.
Is there a max RAM amount a scene can use before it will not fit in VRAM for the render? Or can a scene take 60GB RAM and still render on a 11GB card?
Ram is so expensive now I don't want to buy more than I really gonna use.
Will 32GB RAM be enough for all possible projects I could push into my 11GB VRAM for rendering?
Sorry, posted thread in wrong forum part, please delete this one if necessary. Thank you! :)
Post edited by j1039564 on
Daz 3D is part of 
Connect
DAZ Productions, Inc.
224 S 200 W, Suite #250
Salt Lake City, UT 84101
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
Generally I've found that the system RAM requirement is 2-3 times the VRAM requirement. My 11GB 1080ti often uses 30+ GB of system RAM, but I have 64GB so I'm not sure if 32GB wouldn't be enough. You might be at the ragged edge with 32GB.
...huge scenes. Yep 64, and as many CPU cores as you can afford in case the process dumps from the GPU.
I have a single 11 GB GTX 1080ti card and 32 GB RAM So far it has been adequate for me; but I have room to expand to 64 GB RAM if I need to sometime.
Maybe you could try starting wth 32 GB RAM and expand it if necessary. Just be sure your motherboard has space for the additional RAM
Your 11GB card will have issues with RAM before your system does with 32; I have 64, and don't regret buying it; of course I have a lot of cores for when the scene drops to CPU, which it does far too often. Although I don't have an 11 GB card.
If your card will take 32GB, and you can add another 32GB later then that might be the way to go; compare prices on two lots of 32 v one lot of 64.
If you have to ditch/sell the 32 to get 64 then you have a different issue.
Yeah, I'm generally not a fan of just assuming "more is better" and throwing more hardware at everything. There are other ways to manage big scenes (as I just mentioned in another thread), and ultimately I think using those methods provide some huge benefits. I'm talking about breaking your scene into smaller pieces/scenes and render in layers, and do post-production on those layers. It saves on hardware, it saves on render times, it gives you much more control over your product, it gives you realtime effects like depth of field, and so on, like the pros do.
Whilst I agree with Ebergerly, being forced to jump through loops all the time is annoying.
Options are better, but sometimes, load setup > render; leave it to it is a great option. Which is why I like the CPU option, which is resonable in time if enough threads, and not having to spend what can be a lot of time tweaking and optimising so it fits is also a benefit. Whilst a render is happening I could do something else, if it wont fit on the card and it needs to, then that time my be 'very expensive'.
I would estimate that 32 would work. But if you want to do other things with your PC while you have Studio open, you might want to go to 48 or 64 (depending on how your motherboard is setup for DIMM slots). For instance I often have GIMP or Hexagon or some other tool open at the same time I'm working in Studio and having the extra RAM helps avoid paging memory to the hard drive.
...more CPU cores are also important for multitasking as you can allocate cores to specific programmes you frequently use. Daz only uses a single core (thread) during scene assembly, but will use all (I believe up to 24 or 32) when rendering unless you are rendering on the GPU in Iray.
so I'm really handicapped with my 16gig ram and onboard GPU ..
More with the GPU than the RAM, but things would be faster with 32GB RAM and at least a Geforce 1070, better with 1070TI, which isn't that much more$, but then you might as well step to a 1080 (which isn't that much more money than a 1070TI). Those three have returned to reasonable costs. The 1080TI is more reasonable than it was, but still pretty $$.
You want long iray render times, try my setup.
I myself am running a 6 gig ram and an NVIDIA 1030 on a ASUS CM5571. State of the art 8 years ago. It is strait out of the box exept for GPU and Power Supply. Even then I bought the 1030 because I was not sure if a 10 series card would run on it. Now that I know it will I can get a real card.
Currently planning a massive upgrade but will buy components as I can afford them.
I bought the Vid card, a 1070Ti and an MSI Intel motherboard. It will arrive in 3-5 days then it will go in my existing PC until I can get the CPU, First 16 gig RAM, CASE, HDD, and DVD drive. Then I will assemble the new PC and add the rest of the components. I will Reuse the 620 Watt power supply until I get the 2nd GPU.
Eventually I will be buying a second HDD for backup purposes, 16 more gig ram (32 gig total but will upgrade to 64 when I can),and a 2nd GPU to NVlink to the 1st.
I will be spreading the upgrades out over the next 2 - 4 months. Will purchase the less expensive items first and fit the more expensive ones when Christmas sales allow.
Since DDR3 Memory is cheep I might upgrade my existing PC to 16 gig (Its max capacity) until then.
A used 1080ti is only $100 more than a new 1070ti. or less, actually if you look. A used 1080ti will still have around 2 years of warranty on it, too.
PCIe is back compatible, so you can use any new GPU, even a new Turing in older PCIe motherboards. It may not be optimal, but it is possible. And for the record, there are people who have installed 1070's into Core 2 Quad machines even older, like from 2007. The machine was so old that it used DDR2 RAM, not even DDR3, and capped at either 6 or 8GB total RAM. Even with this spec the 1070 ran the SY benchmark test in nearly the exact same time as other 1070's in brand new high end machines. So speed will not be an issue here. You may experience issues with running out of RAM and limit what you can do, but at least speed will be fine if the scene fits.
I generally recommend trying things out smaller before going all in. RAM is the easiest thing to upgrade on a PC, so there is no harm is trying out a smaller stack of RAM first. If you find you need more, then picking up more is easy. Just make sure it is the same RAM in brand and spec.
In case you don't already own it, Scene Optimizer is a really good friend when it comes to huge scenes and Iray: https://www.daz3d.com/scene-optimizer
biggest key point in regards to memory and upgrading is you might as well upgrade your current ram capacity to the maximum you can afford / motherboard can handle...
On one hand, "you can never have too much computer memory" . Even if you get 64 gigs and find you only use 32 or less of it, you can entirely disable the windows swap file, and/or make use of the other 32gb as a "ram drive" for swap/cache/scratch disks in applications like adobe photoshop.
On the other hand,
A.) If you only upgrade to 32gb = 4x8gb sticks to populate all 4 ram slots on your motherboard and find you need 64gb later, you're spending even more money and lost money on the 32gb you already bought.
B.) If you only upgrade to 32gb = 2x16gb sticks with 2 slots unpopulated, you would want to try find as close of matching set of memory for the 2nd 32gb so the timings are compatable. Hence the emphasis of vendors selling you "Matched kits" of 16gb 32gb, 64gb since those are sticks of matching part numbers, and timing specs.
C.) Its generally good to populate / max out all your ram slots so the memory controller has the most flexibility to use optimal performance settings. 4 bank interleave is better than 2 bank interleave, and distributed across 4 slots vs 2 of 4 slots.
And hey, as others have posted above, you don't need to worry about having enough ram if/when the scene exceeds videocard capacity and jumps to system ram + cpu only rendering mode.
Additionally for those looking at Octane render engine as an alternative to Iray, it does feature using video card ram + system ram for the texture storage and processing, and the new 4.0 coming out expands the capability for both texture and geometry to be offloaded to system ram ... For the cost of a mnimal reduction in gpu render speed, the more system ram you have, your render scenes can be infinitly larger and more complex than you could with even a $5000 Nvidia Quadro 24gb capacity workstation card.
Put it this way:
$1000 for a 11gb 1080ti to render Iray
$1749 for a 16gb Nvidia Quadro P5000 to render Iray GPU mode Only
$2300 for a 16gb Nvidia Quadro RTX 5000 to render Iray GPU mode Only
$2399 for a 24gb Nvidia Quadro M6000 to render Iray GPU mode Only
$4473 for a 24gb Nvidia Quadro P6000 to render Iray GPU mode Only
$6000 for a 24gb Nvidia Quadro RTX 6000 to render Iray GPU mode Only
$700 for an 8gb 1070ti to render Octane GPU mode only + current System ram
OR A,) $700 for an 8gb 1070ti or better (or lesser card like a 4gb or 6gb 1060) to render Octane GPU mode only + $229 for 32gb System Ram
B,) $700 for an 8gb 1070ti to render Octane GPU mode only $569 for 64gb System Ram
With Octane, it matters less how much video card ram you have since it can fully utilize the system ram to keep on GPU rendering, even ad a 2nd gpu to double the rendring speed. Compared to Iray that drops to CPU + System Ram only mode once video card ram limits exceeded, nomatter if you have 1, 2, 3, or 4 video cards since all that money and gpu processing power is wasted once you render scene exceeds the hard mmeory limit of the graphics cards.
Even with scene optimizer, you can still max out and exceed the video card ram limit by adding too much content.
This is why I am upgrading the GPU first, I can use it as a temporary upgrade on my current PC and buy the other components over time. I am actually thinking of upgrading this PC's ram to 16 gig, cost $120.00.
The reason I am going all in on a new unit is CPU and the ability to support atleast 32 gig of RAM. I am looking at getting a good CPU now while its workable in my budget. GPU's and RAM can be upgraded over time.
The thought of going all in on a 2080 ti GPU makes my bank acount cringe in terror. I would rather get a second 1070 or 80 and NVLink it to the first GPU for the vid memory.
It found its way into my cart as soon as I found out of its existance back in early May.
My understanding is that, with multiple GPU cards, the whole scene has to fit into the memory of each card - you combine the cards' processing power but not their memory. So having two 11GB 1080ti cards it not equivalent to having a single 22GB card.
If your using Iray, then unfortunately doubling up on video cards does not translate in doubling up on video card memory. Whether you have 1x 8gb card or 4x 8gb cards, your still only have a total effective video card memory of 8gb that the scene has to fit into for each card, since the 8gb per card is isolated to per card. It doesn't pool or stack combined across all cards , otherwise yeah you'd be going up from 8 to 16 to 24 to 32gb in total usable video card ram for a scene.
The only thing you gain in adding more cards of same capacity is increasing your GPU rendering speed (again assuming the scene will fit into the 8gb of video card memory - per card) else your back to rendering on the cpu + sytem ram only.
So yeah, getting a good CPU + motherboard that has the capacity to upgrade up to 32 or more ram is a wise choice and fallback option if and when your scene goes to the cpu render mode. On the bright side, a good fast cpu also means you can enable GPU + CPU rendering for a bit of an extra bump to your rendering speed than pure GPU only alone...
I just wish Iray gets onto the bandwagon regarding allowing a scene to surpase video card memory, utilize system memory and still be rendered by the cpu (this is what octane render allows, alternative and direct competitor to iray. and in octane 4.0 theres even more bigger improvements that'll make it a very attractive alternative to iray's current limitations)
Only the Quadro RTX 5000 16gb to 8000 48gb card has this capability, for linking to a 2nd Quadro RTX for up to a combined massive 96gb pooled framebuffer memory via a proprietary new nv-link for those cards. Anandtech Quadro RTX Review :
Now, the other thing is assuming someone has deep enogh pockets to buy one, let alone a pair of those cards, would they be detected in Daz Studio, and would the daz studio iray engine detect and utilize that total combined memory pool between 2 cards?
Would be really nice if nvidia could implement at the driver level for the 10xx and 20xx rtx consumer cards an nvlink memory pooling featuree. And/or copy Octane and allow for off-card system memory storage while still rendering via cpu.
Whewrease right now you could subscribe to Octane 3.0, have a single 6gb 1060 or a 8gb 1070 card, and as long as the geometry data fit into that 6 or 8gb card and you have more than 16gb of system memory, you could render a scene that easily used 16gb worth of textures all on your card's gpu. With Octane 4.0 around the corner, they're making even larger improvements, and possibly including cpu rendering support. They do that and iray's going to suddenly look very limited tying people to the physical capacity of their video cards or force to the much slower cpu only iray rendering mode. 2019 is going to be a very interesting year I think...
Don't get me wrong, i love how easy iray is to use thanks to daz studio's native support for it and the store content newer products are all iray optimized. And when you can render a final finished scene in 5 minutes in iray on your video card vs waiting 1+ hours for the cpu mode to finish... its hard not to start looking at Octane that doesn't confine you to iray's current limitations.
;)
Which is why I plan on trying one card first. But I was looking at a second card for increased proccessing speed.
Thats the big reason for targeting 64 gig of ram, cpu render dump for larger scenes.
GPU + CPU is my default setup.
...and again consider a High Core Count CPU like a Threadripper with16 cores/32 threads. The more CPU threads you have, the better your render times will be should the scene dump from the GPU. True, it won't be as quick as on the GPU but still faster than with say 8 or 12 CPU threads. Threadripper also supports four memory channels instead of two which helps spread the workload more efficiently among the memory sticks which also translates to an in render speed as well.
One thing to keep in mind is that a large scene is probably going to take longer to render. So an overly huge scene can take a long time even if it does fit on the GPU. So there is still value in being able to take the time in optimizing things for a scene. In some cases you can cut hours off a render time simply by optimizing it better, even it already fit the GPU. And if that huge scene does not fit GPU...look out, it might be a while.
GPUs may not stack perfectly, but you can get faster renders cheaper than going nuts on a big monster GPU. Just keep in mind that to use all the GPUs, the scene has to fit each GPU's own VRAM. So if you have a 1080ti 11GB and a 1070 8GB, the scene must fit the 1070 in order for it to run. But when it does, you will see a decent speed boost.
If you are going multiGPU, then having a lot of CPU cores can help. One thing that has been observed is that multiple GPUs run faster when the CPU has more cores. There was a test that involved Titans and Xeons, and same 4 Titans performed faster with Xeon VS the same Titans paired with an i7, even though the Xeons were not actively rendering. It was not PCIe, because they tested for that and for Iray the GPUs ran equally in x16 and x8.
So a high core CPU can help both when out of VRAM and when rendering multiple GPU. If you plan on staying with just 1 GPU, the CPU core count does not seem to have any impact. It only comes into effect with multiple GPUs. I have no idea if Octane or other render engines behave the same way, but Iray does.
-Does populating the memory banks actually have a measureable impact on performce? Iray does not operate like many programs out there and I seriously doubt Iray would see any impact at all from RAM timing or population. As I already stated, a 1070 with a Core 2 Quad running ancient DDR2 can run the SY bench in the same time as a 1070 with DDR4. I don't think it needs to be stated that DDR4 is lightyears ahead of DDR2, so it looks to me that memory speed and timing has no effect, just like x16 and x8 which have been proven to have no impact. Perhaps if you are only CPU rendering then maybe? Even then I am not so sure.
...yes, it would only come into play should the scene dump to the CPU.
...
...
I don't have a 1030 so I don't know how well it handles or lags while working with content putting a scene together in daz studio. I only have a 1070ti and i'm happy with it.
If your happy with how the 1030 handles for just posing and putting a scene together, you could keep the 1030 in the machine in the top most PCIE slot, and add the 1070ti in one of the lower slots below. When it comes time to rendering, since the 1030 is onlt a 2gb card, you'd want to select only the 1070ti and/or the cpu in the iray render settings tab. By using the 1030 as your primary display card and the 1070ti as a secondary card for gpu rendering only, you get use of the full 8gb on the 1070ti since windows will use the 1030 2gb for displaying your desktop and daz studio scene. And also you'd be making good use and value for the oney spent on purchasing the 1030 (compared to removing it and shelving it or selling it.)
Nope. The number of graphics cards has nothing to do with needing how much system ram you have. There's no hard rule or formula here.
The OP is asking about whether to upgrade to 32gb or 64gb system ram and a single 1080ti 11gb card for large rendering scenes
Another person is also asking about a 32gb or 64gb upgrade with a 1070ti 8gb card.
The OP could also add another mix of 2 or 3 1080ti's or 2080ti's later and be content with the 32gb or 64gb of system ram upgrade.
The OP might also be better ahead also looking at Octane for doing large scene renders for using GPU + system ram for those renders.
The other 3 points here is:
32gb (preferred) and 64gb (ideal) are the good choices for system ram capacity, mainly due to when your trying to render a scene and it won't fit onto the 8gb (or 11gb for a 1080ti / 2080ti) graphics card, the entire scene gets dumped to the CPU + System ram for rendering.
When you know your putting a large scene that obviously won't fit onto the graphics card, then its a matter of "The sky is the limit" for both how much content and what size your rendering the image so your only limiting factor is how much system ram you have.
In example, a heavy loaded scene going to cpu+system ram for rendering could take 12gb for a 1920x1080 image. Now, if you take the same scene and decide to render at 2560x1440, or 3840x2160 or higher, the system ram needed is going to increase steadily as well as render time. Rendering that scene at 16,000 x 16,000 image resolution for a high-end commercial print project could easily surpase 64gb and require closer to 96gb+ of system ram.
When your talking about 4x 8gb graphics cards vs 1x 8gb graphics cards its more about gpu rendering speed, and ONLY if/when the scene fits on the graphics card. You could squeak by on 8gb ro 16gb ram for a 4 graphics card setup, be comfortable with 32gb or 64gb system ram *just in case*. And once that scene nolonger fits on your 4 graphics cards, those 4 graphics cards become 100% useless to you.
You could have a workstation/server desktop with dual or quad cpu's , 8-32 ram slots with 256 - 512gb Registered ECC system ram and a single 8gb 1070ti or 12gb Titan XP, or 8gb Quadro P4000 graphics card. Single graphics card for working in daz studio and the focus of the multi-cpu + 96gb+ system ram for cpu+system ram rendering.
Like I said, to answer your question, there's no hard rule or formula here. You put your system together based on what you can afford, what you plan to upgrade further down the road and what you want it to do.
And all of this is assuming we're just talking about the default Daz Studio / Iray rendering.
Octane is a different type of PBR renderer comparable to Iray in quality, and a very viable alternative considering its feature set.
Once you start factoring in Octane, it will use as many GPU's you have of whatever capacity, it can also make full use of your system ram and still render faster than Iray in cpu/system ram mode. 64gb honestly lets your options are open.
...keep in mind that even when rendering on the GPU, having to keep Daz open with the scene loaded during the process still takes up system memory. If the render process dumps from the GPU then it will require even more system memory to support the process. So say you have a 1070 Ti and the scene is 9 GB in VRAM, when it dumps to the CPU, that 9 GB (18?) is added to the current memory load.of having to hold the scene open. This is why my system would often go into swap mode on large scenes when it only had 12 GB. This is also one reason I am not thrilled that Iray is integrated as with Reality/Lux or Octane once the scene was sent to the render engine's queue, you could shut the scene and Daz programme down while rendering.
With 3DL you could use the free standalone and render to RIB if you didn't mind using 8 CPU threads or less. Unfortunately with Iray, you have to purchase a licence for the render engine to do that.