Overclocked Graphics Cards
Capsces
Posts: 465
I want to get a graphics card for an old computer in hopes of better Iray performance. I am unfamiliar with the overclocked cards. Do they run overclocked out of the box? Are they less stable? If unstable overclocked, can they be run at a normal clock? When purchasing, how do you know if a card is not overclocked? I was initially going to go with EVGA as they usually work well for me, but with the heat issues EVGA’s 10 series had, I am not sure. Regardless, if overclocked cards are stable, I am interested in one of the following depending on funds:
https://www.amazon.com/MSI-GAMING-GTX-1070-8G/dp/B01GXOX3SW/
https://www.amazon.com/EVGA-GeForce-GAMING-Support-08G-P4-5173-KR/dp/B01IA9HLMM/
https://www.amazon.com/MSI-GAMING-GTX-1060-6G/dp/B01IEKYD5U/
https://www.amazon.com/EVGA-GeForce-Support-Graphics-06G-P4-6161-KR/dp/B01LZEZ1S7/
If anyone can answer the questions above and/or give opinions on the above cards or others you think would work as well or better, I’d be most interested.
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
Factory overclocked cards are overclocked at a speed that the manufacturer guarantees the cards to run at.
Overclocking any card BEYOND what the manufacturer says is the rated speed for that model has the potential to be unstable. This is why factory overclocked cards are more expensive than the base model. They're faster, and they are stable at that speed. Realize you CAN overclock the card BEYOND what the manufacturer states....and it may be stable up to a point. Overclockers (not the factory ones) experiment extensively on just how fast they can get the card (via extra cooling, voltage mods, and more) and stay stable.
tl;dr - Cards that are advertised as overclocked are stable and faster than base model cards.
As for how to know if a card is overclocked.....most manufacturers have some specific nomenclature to define it. EVGA uses "SC", "SSC", and "FTW" for their models (SuperClocked, SuperSuperClocked, and For The Win.) Other manufacturers use different ways to indicate the level of overclock. But the only real way to know exactly HOW much is to actually look at the specifications for the card and see what the Core Clock speed is, and then compare that to the others (and the Base clock speed of the base model for the card.)
I've just bought this one (OC means factory overclocked):
https://www.techpowerup.com/gpudb/b3654/msi-gtx-1070-armor-oc
GPU-Z reports that bot GPU and memory runs at about 1900 Mhz, and I haven't had any problems with stability or anything. It's running on 7-10 year old hardware though so I don't know if it's running at full capacity in all respects on that machine.
Its not going to make a huge difference for iray with the same model gpu clocked at different speeds. Well, it can, but you have to get a pretty big clock to even feel it. Clock speeds are not a substitute for more CUDA cores. Some of these cards are built with improved cooling, which is a nice thing to have if you are planning on running iray for hours at a time.
If you can afford one of the 1070s, go for it. That is a big upgrade over a 1060. Just make sure these cards will fit into your case! They are massive, and it is very easy to underestimate how much space these things take up, especially if you are rocking an older case.
I recently picked up a MSI 970, which has cooling like these MSI do. It works incredibly well, the temps stay quite cool. It hasn't even cracked 57C even when running for hours at 99%. Though I have a MSI card, I use EVGA's Precision X software to set the fan curve, which I make more aggressive than what it has out of the box. I'm pretty happy with it so far.
What's your current pc spec?
I wonder what's going on with my card, for the specs says this:
1557 MHz (+3%)
1746 MHz (+4%)
2002 MHz
8008 MHz effective
However, when I render, I get a much higher GPU clock but a lower memory clock:
Thanks for the replies all!
Nice to know at least one person is running on an old computer. The one I am upgarding is nine years old. It is an i7 965 and 6 gigs of DDR3 in an ASUS Rampage II Extreme. The case is a full tower Cooler Master HAF, so I'm hoping it is big enough, but I will measure just to be sure. It only has PCI-e 2.0, but it is my understanding that is not an issue, though it may keep the video card from performing optimally.
I wasn't particularly looking for an overclocked card, but most seem to be, so I just wanted to make sure they are stable. I may go ahead and get the 1070, as I can probably use it even if I build a new computer later. Of course, by then better gpus will be cheaper.
Wish I could help you Taozen, but I have no idea.
...wow that's newer than mine: i7 930 with 12 GB tri channel DDR3 a and 1 GB GT 460, on an ASUS PT 6. with X58 chipset.
As to overclocking I remember something Dr Tyrell' from the old Bladerunner film said:
"A light that burns twice as bright burns half as long".
Of course I don't do games either so boosting clock speed and frame rate means little.
Oh, I feel your pain, especially if you are trying to render Iray. This one currently has a GTX 480. I'm planning to get a memory kit for it too, at least up it to 12 gigs. May not last forever, but I think I can put it all in my husband's computer if something doesn't work in this one.
Well I got some answers to my questions here, the card is fine, it's just that DS didn't put enough load on it to get the memory to run at full speed:
https://www.techpowerup.com/forums/threads/correct-readings.230493/
My board has only PCI-E 1.1, plus a Q6600 2.4 Ghz quad core CPU and 8 GB RAM. Nevertheless Iray renders quite fast with the GTX 1070, SY's test render completes in less than 4 minutes. So it will probably work fine on your system as well. I don't think cooling will be a problem with your case, I have a half tower with the same number and size of fans, and even with top fans running at low speed the card rarely gets over 65 degrees C with IRay rendering.
...yeah the rendering itself is not impacted, I heard it affects how long it takes to initially load the scene into VRAM but even then it isn't really al that much of a factor.
...totally CPU rendering. I tried a very small, simple test on the GPU (monitoring with GPU_Z to make sure it didn't exceed the card's memory) and it actually was a bit slower than the CPU. Need about a 700$ windfall to upgrade to a 1070 as I would also need to increase the physical memory to support it (which, as I have a tri channel MB would mean 24 GB) as well as require getting an OEM of W7 Pro since the Home Edition only supports up to 16 GB (no way am I going to let MS take control over my system with W10).
Do the 1070 cards need more than 12 gigs of system memory? I can't find any memory requirements for those I am looking at.
It says 8 GB (16 recommended) on my package, but I think mostly that's because the games and 3D programs that uses the card may require a lot of memory. I have 8 GB but when the PC is idle and nothing is running there's about 5½ GB free RAM available, but that's not much for e.g. DS. A large scene may easily eat those 5.5 GB, it actually happened to me today and it started swapping to disk which totally locked up the PC so I couldn't do anything. With small to medium scenes there are no problems.
...from what I've read (don't remember which thread it was on) it was suggested to have twice the physical memory as the GPU has so for an 8 GB card you would want 16 GB. Something to do with "mirroring" the VRAM. Again as my system has 6 slots and is configured for tri channel memory, I would need the full 24 GB.
...for myself, 8 GB of VRAM is the minimum as I tend to create fairly "heavy" scenes. With only 6 GB render jobs would end up on the CPU about half of the time.
I found a You Tube review with the MSI Gaming X's box and it too requires 8 gb system memory (16 reccomended). Looks like they would put that somewhere highly visible on their site. My system also uses triple channel memory. Think it can use dual too, but should run faster with three or six modules. Anyway, I plan to get at least 12 gigs, but I will probably not be able to use it with my current RAM.
I'll see if my poor system will run the test, outrider42, assuming things go as planned.
just look up your motherboard and look at the specs to see how much RAM you can put in it.
If that is the same size as my Cooler Master Haf X then you will be fine for big video cards. I have two Zotac 980TI AMP Extremes in mine and have no space issues.
And speaking of those cards, they are overclocked out of the box. If you cannot put a 1070 in your machine due to RAM you may want to consider a Zotac 980TI AMP Extreme. I will vouch for its strength as I can run Rise of the Tomb Raider on just one card without dropping below 60FPS in DX12. It performs really well in IRay too. Though two are better then one.
Taozen,
I have the same specs in this machine. You're reduced memory clock speed may be due to the MB, not the card as PCI-E 1.1 cannot take full advantage of PCI-E 3 cards.
I'm pretty sure my system will take 24 gbs. I was a bit confused at first because the manual says 12 gb, but the ASUS website says 24. I finally found a thread that confirmed the 24.
Mattymanx, my case is the HAF 932. Comparing the two, hopefully all will fit.
Kyoto, it looks like you could get away with the 12 gigs of memory that you have, which would bring the windfall down to around $500.00, depending on the card you get. You could break it down a little more and get the OS, then the card later. Of course, I've been fairly content with 3Delight, but am anticipating the need to render with Iray, considering what I do.
OK. I also intend to build a new machine with more RAM like 32 or 64 GB, I just purchased the card now so I could use in my current render machine until then, to get some more speed.
After playing a bit with the 8GB VRAM I agree, I'd actually say 32 GB VRAM would be preferable. I just wonder why they don't put more RAM on the cards, it's not that RAM is expensive. When a $50 MB can handle 32 GB RAM why can't a $400 video card? You can easily build a complete PC with 32 GB RAM for what a GTX 1080 costs.
Well I was told ( https://www.techpowerup.com/forums/threads/correct-readings.230493/ ) that the PCI-E version should not affect memory speed, it was rather because there wasn't enough load on the card. Seem to be correct - when I tried the Heaven benchmark program memory ran at full speed (2002 Mhz). Despite the limitations of the PC the card runs pretty fast anyway - SY's test render completed in 3½ minute last time I ran it.
VRAM is not the same as regular motherboard RAM. It actually is a LOT more expensive. Your typical DDR4 DRAM DIMM module is operating at 2.133GHz (DDR4-2133, PC4-17000), using 16 chips (8 on each side of the DIMM). The VRAM being used in a 1080 GTX is GDDR5X running at 10GHz (that's 5 times the speed) with only 8 chips. It's more densely packed memory, that runs MUCH faster. GDDR5X runs at a much lower voltage as well, and lower power consumption than DDR4.
They are two VERY different animals.
Good to know. Makes sense too
Capsces,
Found this
https://www.newegg.ca/Product/Product.aspx?Item=N82E16811119160
So before you buy your new card, double check its length.
OK. But then, technically, what would it take to double the amount of VRAM on say a GTX 1080 from 8 to 16 GB, apart from adding the extra 8 GB VRAM and perhaps some better cooling?
...so if GPU_Z is not a good indicator, how else would you know if it swaps to the CPU? The one thing I did notice after selecting the GPU only is it reported every single iteration in sequence to the progress pane whereas in CPU mode it jumps by ten or more.
Doubling the amount of VRAM on a GPU involves several things.
First, is that it adds at least one more bit of address space, so the board requires another circuit trace for all memory address lines. That's a major redesign. However, since most modern GPU boards are using 64 bit memory address space, that's not really a concern.
Second, it increases power consumption. Memory requires refresh cycles, and other continuous updates. The more memory, the more cycles you consume in maintaining it.
Third, it occupies board space. Keeping the boards the same size means a denser arrangement of chips, tighter trace layout (or more layers in the board) which increases cost of the PCB itself. The tighter density of chips also increases the heat density on the board, and increases the requirements for cooling.
Furthermore, depending on the memory chip design, it may require bank-switching. Using memory that is designed for the higher addressing space to avoid bank-switching (which results in processing slowdowns) is more expensive (the chips include more logic lines to handle the additional address space.)
Also, at the speeds the memory runs at, circuit trace lengths start to be an issue. They have to be basically the SAME length for each set, so that the inputs change at the same time at the output. Despite electrons travelling at near lightspeed in copper/silicon, it still takes time for current to travel down from one part to another. And when you are talking about 10 GHz, traces that differ in length more than a few millimeters can become a timing/sync issue. So even more board layout potential issues.
So yes, adding more memory to a card is a lot more complicated than the main DRAM on a motherboard. Furthermore, unlike the old days, we can really use external modules on VRAM anymore due to constraints in the board layout caused by the higher speeds used in VRAM.
OK, I can see the problems yes. And when circuit trace lengths starts becoming an issue we're really starting to get close to the limits of technology. Then what? Interesting to see what they'll come up with, if anything.
Thanks for the explanation!
...HBM is the next level where four stacks of memory chips are arranged on both sides of the GPU processor significantly shortening the path between the two as wel as between the memory chips themselves.
The first GPU out that used this configuration was AMD's Fury-X which had 4 GB of HBM memory (4 stacks of four 250 MB chips) and had a form factor that was about 2/3rds the length of a conventional card.
Last year when Nvidia unveiled it's new Pascal architecture cards, they also released the Tesla P100 compute card with 16 GB of HBM2 memory (four stacks of four 1GB chips ea).
This coming spring, Nvidia will be releasing the first Quadro workstation card with HBM2, the GP100 that will also have 16 GB of VRAM. The card includes has dual CUDA Core sets on each streaming multiprocessor: 64 FP32 cores totaling 3,584 cores and an additional 32 FP64 cores totaling 1,792 cores (allowing for improved double point precision primarily for scientific applications) as well as a whopping 4096 bit memory bus (compared to 384 bit for the P6000) and 720 GB/s memory bandwidth (about two thrids more than the P6000 and Titan-X P). It will also be the first GPU card to utilise the new NVLInk connection when pared with a second (replacing the SLI bridge) which will offer a "wider" pipeline between the cards that will be up to 5 times as fast. The GP100 still uses the normal PCIe 3.0 x 16 interface to allow it to be installed on current MBs. Price is projected to be somewhere "north of 5,000$" as one review put it so clean the drool of the keyboard unless you just won the Lotto.
When (or even if) HBM2 memory technology and NVLink will trickle down to Nvidia's consumer line is still anyone's guess.