Iray video card upgrade question - 2 x Quadro K620 (2GB) or single Quadro K2200 (4GB)
HALO creative
Posts: 96
I'm looking to upgrade my current twin Geforce GTX 750 1GB cards and have the choice of two options:
2 x Quadro K620 2GB cards (384 CUDA cores each, 768 total)
or
1 x Quadro K2200 4GB card (640 CUDA cores)
Purchasing the 2 K620's will actually cost about $70 less than the single K2200, but is there any performance benefits when it comes to Iray rendering when using a single K2200 card instead of dual K620 cards?
I trust the community here so if everyone can vote on the poll which option they would go for it would be helpful.
:)
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
You can only do a GPU render if the whole scene fits on a single card. So if you do somewhat to fairly complex work then your better off with the 4GB. If on the other hand you would not normally do a scene larger than 2GB your better of with the larger number of cores.
Thanks for the tip Khory.
The main focus of my work is on "toon style" as well as medium res model rendering on isolated/white backdrops with postwork/compositing in Photoshop. Not so much on photorealistic models or high object/polygon counts - I may do a rare full scene or high res Genesis figure render.
So I guess from your tip above my best bet would be to go with the dual cards with more cores?
Also, currently all my renders are stills, but I do plan to get into Animate 2 rendering, should this have an influence on my choice as well?
I will go with Quadro K2200 4GB
... I have one on my Laptop.
Is the Quadro specifically required for some other software?
You can get a couple of GTX 960's, each with 4GB and 1024 cores, for only slightly more than a K2200.
Poll removed - the feature is not intended for general use.
The 4GB 960 and 970 have been described on various forums as having 3.5GB DDR5 RAM and .5G of RAM that is 80% slower than the rest.
I don't know if the card will use the .5 of the slower RAM for texturing or if it's exclusive to the GUI buffer, in which case it's a 3.5GB rendering card with an additional 1/2GB of RAM that refuses to lift a finger when it comes time to render. Hopefully not.
http://www.pcgamer.com/nvidia-faces-false-advertising-lawsuit-over-gtx-970-specs/?_escaped_fragment_=#
at ~ $240 it's probably the card I'll end up buying but I'm waiting to see if Daz does a tie in with Nvidia (total speculation BTW) or if Nvidia drops prices in the wake of the Taipei trade show this week, not holding my breath but I am crossing my fingers
The GTX 980 has 2816 cuda cores and 6GB DDR5 and retails start at $749 and includes two year old free Batman game (Mavis Batman Teaches Typing) The Titan has 12GB RAM and 3072 cores or 5760 for the "Z" model. The GTX Titan X starts at about $1,100.00, Titan Z ~$2,900.00 (Sans-Batman) and then the Quatro at $4,200.00 - $5,000.00
then there's this
http://www.nvidia.com/object/visual-computing-appliance.html
96 GB of DDR5 RAM (that's not a typo)
24,576 CUDA cores (not a type either)
it's about $50,000.00 US
(Batman comes to your house and installs and does the ghost pepper challenge)
For our purposes, the slow 0.5GB is is not an issue.
We are not changing that 4GB of data constantly like a game would. So the speed taken to read it is not critical.
It's still a full 4GB available, and that "slow" 0.5GB is still 4 times the speed of your motherboard memory.
The 960 has a different architecture and its 4GB is not effected by this.
They do have 4GB of RAM; whilst for gaming it is not directly useful, the RAM can be accessed for the purposes of rendering; I seem to remember starting a thread that showed it. The controllers (for want of a better term) that access memory on the 970 and 960 are one less iirc than the 980; but there are links to that memory via another controller.
I had a similar situation earlier this year, but my choice was between a second GTX 660 or a single GTX 970. There was also a big price difference there. Although my upgrade reason was because of the then upcoming GTA V and trying to get more speed out of Reality. Iray wasn't around then so it didn't figure into my thoughts, that was a bonus later on.
Everything I read suggested that one card is always better than two. This is the way I looked at it but using your two choices and I'll assume speed increases are roughly the same, over the two cards, just to rule that out but obviously that may factor in to your final decision.
Pros of 2 cards:
It's significantly cheaper than one card.
Cons of 2 cards:
You can only access 2GB of VRAM.
Double the heat output.
Double the power requirements.
SLI doesn't work with all software. Mainly applicable if you do any gaming, I think.
Using old architecture.
Increased fan noise my be a factor to consider.
Pros of 1 card:
Full access to all 4GB VRAM
Less heat output.
Less power consumption.
Using newer architecture.
You still have room to add another card later if you want.
Cons of 1 card:
It's more expensive.
I've probably missed something out but this seems pretty clear to me that a single card is the better option if you can afford it. It's the way i went and I don't regret it for a second. Also, you say that you only do relatively small renders now, but whether it ends up being a toon render or a photorealistic one, the whole scene still has to occupy VRAM. And what about the future, 6 months or a year down the line? Things can change, wouldn't it be better to be more secure in that with 4GB VRAM now? You might regret not having it later on.
But it's up to you of course. For me a single card was the only logical choice.
The 960 and 970 come with 2 or 4 GB RAM broken into 3.5GB or DDR5 and .5GB of apparently slower (by a factor of 80%) but total 4GB usable for rendering
The 980 can be had with 4 or 6 GB
The Titans come with 6 to 12 and "high speed double precision" for the NVIDIA CUDA enthusiast
http://www.roylongbottom.org.uk/cuda2.htm.
the Quadro's from what I can tell are only beneficial to a handful of dedication applications that are optimized to use their chipset. I would assume IRay is one of those technologies optimized to take advantage of the card, otherwise in not optimized environments the card is similar to a GTX 780 (a 6GB 780 is less than $500) if you want to take it home and play lemmings in a browse window
http://www.techsupportalert.com/content/play-classic-lemmings-game-online-your-browser.htm
Thanks for all the tips & suggestions :)
I have followed the suggestions to stick with the GeForce cards and have decided on a single ASUS GeForce GTX 970 DirectCU Mini 4GB (https://www.asus.com/Graphics_Cards/GTX970DCMOC4GD5/) for the primary card.
The main reason being the single fan/compact size (to help with case airflow/cooling) and fitting well within my budget (with a small investment later on, I can possibly afford a second card)
I guess another bonus... this one DOES come with Batman: Arkham Knight (by redemption) when it's released later this month :)
I may keep one of the GTX 750's as a secondary card to run my second 24" monitor...
There's another discussion - does running two 24" monitors on the one GTX 970 card have any performance degradation, or is it better if I run the second monitor off a dedicated GTX 750 1GB card? And, would DAZ Studio Iray use all CUDA cores and VRAM or only the cores/ram from the higher or lower VRAM card?
The card that doesn't have enough memory drops from the render list...it's as if it wasn't there.
So, if your scene was less than 1 GB, it would use all of them (both cards), but to be under 1 GB it has to be a very simple scene.
Also, yes, if you run your monitor off the same card you are using to render, there is a performance hit...not sure exactly how much of one, but usually enough to be noticeable.
My advice - put both monitors on the 750 if possible and don't use it for Iray. Depending on screen resolution and size you can have nearly 300 MB tied up just running monitors. That's 300 MB not available for Iray. Also, when the card is real busy rendering the video updates take the crumbs left over. I have a 740 as my only card, driving two monitors. When I let Iray use it, solitaire becomes unplayable because of video lag.
Iray will use both cards if you tell it to - but if the scene doesn't fit in the vram on one of them it will get dropped from the render - and only the simplest of scenes will fit in the 1 GB of the 750.
Good advice, and it is definately possible for me to have both DELL 24" monitors (1920x1200) running off one GTX 750.
I was doing that previously and sourced an identical GTX 750 second-hand to run the monitors separately.
Thanks
I apologize if wrong but I was under the impression that the two cards in discussion K620 and K2200 are of the same architecture. Other than that all the pros and cons are more than valid.
I know this thread is a tad old but would the GTX 980 Ti 6GB GDDR5 be adequate to run a single monitor and Iray at the same time or would a secondary card also be recommened?
I put a 980 ti in my system last month. I'm still using the 740 for the monitors. When rendering, the 980 gets to 97% GPU load, so I don't know what effect that would have on monitor response.
Thanks naffauk,
Is the hit to your card due to scene size or image rendering resolution? I think I could live with the hit in resolution but if it is due to scene size, then I have a GTX 560 that I could swap in to run my monitor if I needed to.
Most of my scenes use about 1.8 GB of Vram; I've had a couple go to 3.6 GB. I usually render to a 16 X 9 aspect ratio with the large side at 1920 pixels. As soon as the scene finishes loading to the card the gpu and vram usage become a rock-solid constant (I'm using gpu-z to monitor). The card also goes to 82 degrees C pretty much instantly, with the fan staying below 50% speed - so I think I'm going to work out a custom fan speed curve pretty soon.
That's the other thing - my system is strictly air-cooled but it has a lot of fans. :-) I've had the cpu run 30 hours at 100% and never get over 59 degrees C.
I run Reality/Lux renders for days between 1600-1900 pixel resolutions and my cpu doesn't go over about 54 degrees C and that's in summer - in winter about 40 degrees C.
My only problem now is that when taking a look at my motherboard specs I only have 1 PCI Express X16 slot.
Maybe a GTX TITAN-6GD5 would be better in my situation since I only have the one PCIE slot. I know it has the same vram as the GTX 980 but it has more cuda cores 2688 vs 2048.
***Had to make an edit to this post because after more research my other slots are all PCI.**
I just want to verify before I hit "buy", this IS the Titan X card everyone is praising?
http://www.amazon.com/EVGA-GeForce-GAMING-Graphics-12G-P4-2990-KR/dp/B00UVN21RQ/ref=sr_1_1?ie=UTF8&qid=1446467786&sr=8-1&keywords=nvidia+titan+x
2x has no advantage for iray
I just want to verify before I hit "buy", this IS the Titan X card everyone is praising?
it appears to be a very well packaged card for Iray. However for $549.99 more you can get one with 2688 more cuda cores, thats nearly like sticking a 980 Ti into the rig with the card your looking at for about $100 less and without the 6GB RAM limit of the 980Ti
Your renders will be complete before get past the Daz Studio splash screen.
http://www.newegg.com/Product/Product.aspx?Item=N82E16814487038&nm_mc=KNC-GoogleAdwords-PC&cm_mmc=KNC-GoogleAdwords-PC-_-pla-_-Desktop+Graphics+Cards-_-N82E16814487038&gclid=COD_ot3-8cgCFYRCaQodACwKLQ&gclsrc=aw.ds
http://www.migenius.com/products/nvidia-iray/iray-benchmarks-2015
Iray Performance Benchmarks and Hardware Setup
Benchmark Results
These are updated benchmark results for Iray 2015. We have started retesting our previous benchmarks with this updated version of Iray as there are significant improvements on newer architectures (Kepler and Maxwell) over the previous generation Fermi architecture. We don’t have all of the same cards re-tested as yet but are working to test as many as possible. Note that we have now also added benchmarks for both Iray Photoreal and Iray Interactive modes (the later with two different profiles). Additionally the benchmark scene has been fully updated to utilise MDL (Material Definition Language) to ensure accurate results.
Iray Photoreal
These results are for Iray Photoreal. We use the batch scheduling mode to ensure maximal use of the GPU resources. For more details see the testing methodology below the results.
00.511.522.533.54GeForce GTX TITAN XQuadro M6000GeForce GTX 780 Ti SuperclockedGeForce GTX 980GeForce GTX TITANTesla K40Tesla K20GeForce GTX 580Tesla M2050Quadro 6000GeForce GTX 750 TiQuadro 50004.44.293.983.233.022.712.11.961.341.311.170.97
Iray Interactive
These results are for the Iray Interactive mode which focused on faster interaction. We use the same testing methodology as Iray Photoreal however Iray Interactive has significantly more options available to tune performance. As such we have split the results into two profiles, one using only point lights and Screen Space Ambient Occlusion (SSAO) with no global illumination. This mode is generally one of the fastest modes and utilising raytracing functionality less. The other is a much more complex mode using full Global Illumination, area light sources and ray traced environment lighting. This mode is generally slower though can still be classed as interactive.
Iray Interactive (SSAO)
01234567891011GeForce GTX TITAN XGeForce GTX 780 Ti SuperclockedQuadro M6000GeForce GTX TITANTesla K40GeForce GTX 980Tesla K20GeForce GTX 580Tesla M2050Quadro 6000GeForce GTX 750 TiQuadro 500011.8111.5911.269.68.478.146.85.894.264.114.043.34
Iray Interactive (Global Illumination)
00.511.522.533.54GeForce GTX TITAN XQuadro M6000GeForce GTX 780 Ti SuperclockedGeForce GTX 980GeForce GTX TITANTesla K40GeForce GTX580Tesla K20Tesla M2050Quadro 6000GeForece GTX 750 TiQuadro 50004.184.093.63.432.812.572.121.51.451.131.12
Nimbix
Nimbix is a HPC cloud provider offering high end NVIDIA GPUs. We ran our benchmark on the various node types that are offered by Nimbix. These nodes use container technology rather than virtualisation so typically perform very close to bare metal speeds.
Iray Photoreal
00.511.522.533.544.555.5NAE_16C64-2K40 (GPU + CPU)NAE_16C64-2K40 (GPU Only)NAE_12C24-2M2090 (GPU + CPU)NAE_12C24-2M2090 (GPU Only)5.815.183.673.23
Iray Interactive (SSAO)
02468101214NAE_16C64-2K40NAE_12C24-2M209015.079.62
Iray Interactive (Global Illumination)
00.511.522.533.544.5NAE_16C64-2K40NAE_12C24-2M20904.883.5
Amazon EC2
We also tested Iray on Amazon EC2 GPU instances both with and without the CPU enabled. Note that Iray Interactive tests do not utilise the CPU since that rendering mode is GPU only.
Iray Photoreal
00.511.522.533.544.555.5g2.8xlarge (GPU + CPU)g2.8xlarge (GPU Only)cg1.4xlarge (GPU + CPU)cg1.4xlarge (GPU Only)g2.2xlarge (GPU + CPU)g2.2xlarge (GPU Only)5.744.733.072.631.491.28
Iray Interactive (SSAO)
02468101214g2.8xlarge (4 x GRID K520)cg1.4xlarge (2 x Tesla M2050)g2.2xlarge (1 x GRID K520)14.918.515.24
Iray Interactive (Global Illumination)
00.511.522.533.544.55g2.8xlarge (4 x GRID K520)g2.2xlarge (1 x GRID K520)cg1.4xlarge (2 x Tesla M2050)5.193.332.91
Testing Methodology
All benchmarks have been performed under Linux (usually CentOS 6.5) with the latest available NVIDIA drivers (as of writing 340.32). Some tests were performed on older drivers where administrative access to the machines was not available. Where we have full control over the environment the following setup was utilised.
In order to ensure we are testing raw Iray performance we have developed a stand-alone benchmark tool based on Iray. Our tool renders a fixed scene multiple times and averages the results to ensure consistency. To ensure the results mean something for real-world use we utilise a non-trivial test scene, ensuring the GPUs have plenty of work to do. The image above is a fully converged version of our test scene.
Please note that results between Iray Photoreal and Iray Interactive should not be compared directly. Rather you should compare the relative performance of the various GPUs when performing the same type of rendering. Also note that these benchmarks are not performed in a way that they can be compared to the previous series of benchmarks migenius conducted which is why we are retesting even the older cards where possible. This is due to changes in both the scene data (to move to MDL) and Iray itself.