TR Forums

Shinare wrote:

Unfortunately there is no leeway in which model I am allowed to purchase because of reasons... Its only the Optiplex SFF line, been moving up the 70X0 numbers these days..

Damn, I know work can be stupid but that's something special, requirement for screens to work being blocked by a refusal to buy the same make and model in a different chassis option???

Topinio wrote:

Shinare wrote:

Unfortunately there is no leeway in which model I am allowed to purchase because of reasons... Its only the Optiplex SFF line, been moving up the 70X0 numbers these days..

Damn, I know work can be stupid but that's something special, requirement for screens to work being blocked by a refusal to buy the same make and model in a different chassis option???

Welcome to life in a big bureaucratic corporation...

just brew it! wrote:

Welcome to life in a big bureaucratic corporation...

Still taking that raincheck

Back semi-OT, if anyone cares for an entry in the EVGA affilate prize draw enough to want to PM me their affiliate code, go fo it.

@Topinio this is wonderful information, thank you for sharing it with us! Were those tests run on the i5-2500K system in your sig?

Interesting to see the GT 1030 basically the performance of the R7 250, although I too expected the difference to be a bit bigger. I wonder how much the 64 bit memory interface is holding it back?
With the Kepler cards you could do a comparison between the GT 730 and GT 740 to see how much extra performance came from the memory (the GDDR5 GT 730s were 64 bit, where the GT 740s were 128 bit, but they both had 384 Keplers) and as I recall it was about a 30% gain.

deruberhanyok wrote:

@Topinio this is wonderful information, thank you for sharing it with us! Were those tests run on the i5-2500K system in your sig?

Thanks, and no problem, glad to; and yes.

deruberhanyok wrote:

Interesting to see the GT 1030 basically the performance of the R7 250, although I too expected the difference to be a bit bigger. I wonder how much the 64 bit memory interface is holding it back?

It seems heavily task-dependant, which is to be expected as the cards have strengths in different areas. While a better processor, the GT 1030 is -- typically? -- starved of weak on memory bandwidth, only 2/3 of the HD 7750 ...

• doesn't run within the Radeon's 1 GB will favour the GeForce obvs,
• runs within the 1 GB and is compute-bound will favour the GeForce by 20-35% (Star Swarm, F1 2010 no AA),
• runs within the 1 GB but needs any reasonable memory bandwidth will favour the Radeon (SteamVR),
• reasonably balanced will be relatively close (F1 2010 with AA, Metro 2033) but minimum framerates will be likely better on the GeForce due to its much higher clocks,
• Street Fighter IV no idea, suspiciously better on the GeForce ... but ofc, that's a thing ...

Ok i got it. I put it on pci 4x becouse one port does not fit the casing when i try to put on pcie 16x.
my old R7 250 e https://youtu.be/9DG8D0bBYB8
my new GTX 1050 lp 2GB https://youtu.be/rY_S2uZQuyY

Sorry for my english....

Ok, if someone tell you Put your GPU to pci-e 4x do not listen to him.
I put the same gpu to x16 and Works much better - smoother
The videos do not reflect that but believe me the game works much better, there is no strange trimming.

GTX 1050 on pci-e 16x https://youtu.be/u_9HGiGAyMs

Im a bit late in posting as ive had this rig for almost a couple of years now. i have dell 9020 ssf and I managed to squeeze in a Geforce 750 ti.

A bit more aesthetically pleasing than having the psu hanging loose

I did have to take the side off my Dell stock power supply and luckily the GPUs fans fit inside the PSU. haha, in summer i have to have side off the pc else it can get a bit toasty and I get artifacts or glitches, its been really hot this year but normally it runs just fine fully closed up.

crabpunt.com/wp-content/uploads/2017/09/IMG_6064.jpg Pooop it wont let me post an image/links yet. copy and paste if you wana see.

The GPU that fits in is a - [MSI GeForce GTX 750 Ti DirectX 11.2 N750ti-2GD5TLP 2GB 128-Bit GDDR5 PCI Express 3.0 x16 HDCP Ready Video Card

I had to a get SFF bracket for the graphics card off amazon, You lose the VGA when you replace it with a half height bracket that includes HDMI and DVI.

I also had to unscrew and slightly tilt the PSU

I would generally recommend against doing that, for the heat problems you're seeing sometimes and also because an open power supply is a recipe for a pop and blue smoke incident. But if it works for you, I guess I can't argue with the results.

I haven't tested anything in any of the SFF boxes newer than a 9010, which still had the x16 slot in the "correct" location to use a double-wide GPU. Did you try running the card in the x4 slot? I'm fairly certain it won't work as the slot isn't keyed to provide enough power, but over time a lot of people have asked here and it would be good to have confirmation one way or another.

single slot low profile?
GTX 750ti> 1030 GT.
Just follow this guide:
/geforce-profile-single-sujet_968419_1.htm

nicoG wrote:

single slot low profile?
GTX 750ti> 1030 GT.
Just follow this guide:
/geforce-profile-single-sujet_968419_1.htm

Heh, welcome, and we know GTX 750 Ti performs better than GT 1030, it's also twice the TDP... but isn't it also either only available as a non-LP single-slot card or a LP but two-slot card that throttles (so might not even beat the 1030)?

Or correct me please, where can I buy a single-slot, low-profile, brand-new-with-warranty GeForce GTX 750 Ti which doesn't throttle and so perform worse than the benchmarked ones?

Your best option would be finding a GM107-based Quadro K1200 if that's what you're looking for. Performance will be about the same as a GTX 750 (non-ti). You could also look at the GP107-based Quadro P1000, which is basically a single slot GTX 1050 with slightly lower GPU clock and much lower mem clock. But as I've mentioned in the past, workstation cards are far too expensive for the relative performance if you're just trying to play games.

I also don't know what that link is supposed to be, but I won't consider user-modified cards a real option for this segment.

If your primary purpose is gaming and you're going to go to all of the trouble of modifying a card for single slot operation (which is going to require you buy two cards and swap heatsinks, spending more and voiding warranties) or if you'd rather spend even more on a single slot Quadro / Radeon Pro card, you really would be better off buying a larger system and not wasting your time with trying to keep to low profile, single slot.

Thanks for guide,

I’m running a dell opti 9020 sff (i5 4690) with a 750 ti low profile double slot. It does fit in the x4 slot and works just as well when it was in the other computer (q9650) where it was in a pci 2.0 x16 slot. Its a comparetively low end graphics card so the x4 vs x16 doesn’t really come into play in my opinion. Also, I read that power is supplied the same no matter if it is x4 vs x16. Difference is its not a direct connection to cpu, uses shared pipelines with usb and lan. Might be an issue if your using usb 3.0 during gaming or something?

Still waiting for something better that can fit without spending $200+ lol but it can play gtA5 1080p 60 fps (consistantly thanks to the newer cpu) at low/normal settings so no rush...

deruberhanyok - Firstly, I have to thank you for consistently keeping this thread updated with recommendations. I was one of the first beneficiaries of this thread within a week or so of you having created it, and a week ago I used it again, because I replaced my R7 250E with a GT 1030. Here's results of the one benchmark that I ran.

Settings used for the Unigine Valley 1.0 benchmark are:

• Render: Direct 3D 11
• Mode: 1920x1200 2xAA fullscreen
• Quality: Ultra

Running in a:

• 2014 Dell OptiPlex 9020 SFF
• Intel Core i5 4570 at stock 3.2 GHz
• 16 GB DDR3 at 1600 MHz
• SanDisk Extreme II SSD

Nvidia GeForce GT 1030
MSI's GeForce GT 1030 2G LP OC, which is mildly overclocked from the factory.

• FPS: 23.7
• Score: 990
• Min FPS: 13.7
• Max FPS: 41.9

AMD Radeon R7 250E
(a.k.a. a rebranded Cape Verde 7750)

• FPS: 13.2
• Score: 552
• Min FPS: 8.1
• Max FPS: 22.1

Intel HD Graphics 4600

• FPS: 3.6
• Score: 149
• Min FPS: 2.1
• Max FPS: 5.9

The reason for the replacement was because the AMD card seemed to be acting up - it caused occasional crashes of the PC, despite cleaning it of dust, cleaning its contacts, reseating it, etc. Once again, this thread came to the rescue as on page 1 there was a recommended single-slot low-profile card, namely the GT 1030. What further convinced me was the fact that this was finally a card with current generation technology, as this lower-end of the market has been languishing with rebranded parts for quite some time.

Objectively, the card idles and runs much cooler (which is expected, given the process changes), and with a lower power draw.

Subjectively, it also means that the card is quieter, and is effectively inaudible under load given the other ambient noise in my room. Moving from the R7 250 to the GT 1030 has made no difference to the gaming demands I place on it, since they're all lightweight games (Banished, Rise of Nations, Age of Mythology).

The one gripe I have is that in terms of specifications, the card reads like it has been gimped slightly, with 64-bit memory bandwidth, and with the card running on an x4 PCI-E configuration (it is plugged in to the x16 slot which is electrically an x16 slot, just that the card is internally capped at x4). In the real world, I don't think these matter, but when comparing specs to the previous card, it does seem an odd decision. I'm not sure why Nvidia felt the need to do this, as the GT 1050 is way ahead, so there's no risk of the 1030 undercutting it.

Hi,

for a 4k desktop music composer Win10-based computer (should be as quiet as possible), is this VGA sufficient?
- Gigabyte GT 1030 Silent Low Profile 2G
(Sorry, I am still not allowed to post links here).

No gaming at all, just for work.

thanks,
Baci

bacizone wrote:

for a 4k desktop music composer Win10-based computer (should be as quiet as possible), is this VGA sufficient?

Yes, if the screen is HDMI 2.0

I'm glad folks are still finding this thread useful. It has taken me well over a month (not much spare time these days) but I have run a set of benchmarks on several different cards and posted the results here:

viewtopic.php?f=3&t=120455

I'll give the front page of this thread an update in early January. Meantime, hopefully that data is useful to anyone else considering an upgrade, as @arunphilip found there's definitely a performance gain to be had.

While it isn't useful to me now I dream of a day it might be. I thoroughly enjoy this thread. Thanks for the benchmarks and keeping it going.

I just wanted to say I've been following this thread for a couple years now and I find it to be a valuable source of information regarding single slot, low profile GPU options. Especially before the gt1030 released.

Earlier this year a user had posted a link to a low profile single slot gxt 1050ti but sadly no other info could be found about it. Until now.
Someone on reddit has found a Chinese website that sells them:
item.taobao.com/item.htm?spm=a230r.1.14.1.7fc419a1VdFwIG&id=550688199970&ns=1&abbucket=2#detail

This is the original conversation:
reddit.com/r/sffpc/comments/6lp8ib/is_anyone_interested_in_the_single_slotlow/

Unfortunately at 190mm its too long for my HTPC....but it might be a game changer for some people.

We have a Dell PowerEdge R420, and it has a Intel Xeon E5-2400 CPU. It's only a 2U blade, and one x16 PCIe slot with x16 bandwidth, 3.0, half-length, full-height. Will something like the RX 550 or GT 1030 work? It needs to be compute capable version 3.0 and up for it work work with TensorFlow.

The GT 1030 is Pascal based so it ought to support the standard you need, though "compute 3.0" is a bit generic. Assuming you're talking about CUDA there, which case you'll be fine - the GT 1030 supports version 6.1.

https://en.wikipedia.org/wiki/CUDA#GPUs_supported

If there's a different standard you're referencing it would help to know what it is specifically - OpenCL doesn't have a version 3.0 and DirectCompute I think is just referred to by the DX version number.

As for finding something that fits - you're saying half-length but full height. I've got some R420s at work and they're all 1u units. If they're the same as the ones I've worked with, your limitation is single slot and the PCB length can't be much longer than the slot.

Most GT 1030s will fit in there just fine. The passively cooled ones will not - they are double width due to the size of the heatsink - but the PNY, Gigabyte, MSI and Zotac low profile cards with a fan are single slot.

Some of the full height cards will fit as well. EVGA has one with dual DVI outputs that is very retro and I love it so so much, but I think the Asus and Gigabyte full height cards might be slightly too tall. Also of note, the EVGA low profile card with a fan on it would also probably be too tall.

As for RX 550, if you need CUDA, well, no, it won't work.

Also of note - holy crap! Look at the prices on these GT 1030s! $110 from Newegg? Have the miners screwed everything up that badly?

Time to update the first post: "If you want to build a system with a low profile GPU right now, don't bother. Just spend your money on a Playstation or Xbox and some games. If you are already stuck with a system that needs a low profile GPU, well, that sucks."

deruberhanyok wrote:

The GT 1030 is Pascal based so it ought to support the standard you need, though "compute 3.0" is a bit generic. Assuming you're talking about CUDA there, which case you'll be fine - the GT 1030 supports version 6.1.

Ah yes CUDA, for use with TensorFlow and some teaching of neural net/deep learning.

deruberhanyok wrote:

As for finding something that fits - you're saying half-length but full height. I've got some R420s at work and they're all 1u units. If they're the same as the ones I've worked with, your limitation is single slot and the PCB length can't be much longer than the slot.

Sorry I believe you are correct, 1U. Does it have drivers for Fedora Linux?

NVIDIA has binary drivers for the Pascal-based cards which you can download and install yourself:

http://www.nvidia.com/download/driverRe ... 0646/en-us

Or, if your distro/version has a proprietary drivers repository, you can (and probably should) install from there.

For what you are doing I think you'll want the proprietary driver package, but there's also open source support in newer Linux kernel releases (I think they started working in 4.11? I can't recall.)

https://nouveau.freedesktop.org/wiki/FeatureMatrix/

These cards fall under the "NV130" column on that page. Compute capabilities are generally WIP status for all of the open source drivers, so I wouldn't expect them to work for what you need, but it at least means if you're running Fedora 26 or 27 you shouldn't have any issues at boot (if there isn't some basic open source driver support the system may have trouble booting; typically the only solution is to add "nomodeset" to your one-time boot options and then install the binary driver).

I can't post links yet but these don't look promising. The Geforce GT 1030 does not seem officially support CUDA, if you check their web site with the page name: geforce-gt-1030

And on medium dot com search for the page name "using-the-nvidia-gt-1030-for-cuda-workloads-on-ubuntu-16-04-4eee72d56791" and that implies that TensorFlow does not work with this card, at least not with Ubuntu.

Do you know anything about the GT710? That supports CUDA and from the wikipedia page it implies compute version 3.5.

Would the EVGA GeForce GT 730 1GB GDDR5 Low Profile Graphic Card work? It seems small enough even with the fan.

RobbieTheK wrote:

I can't post links yet but these don't look promising. The Geforce GT 1030 does not seem officially support CUDA, if you check their web site with the page name: geforce-gt-1030

And on medium dot com search for the page name "using-the-nvidia-gt-1030-for-cuda-workloads-on-ubuntu-16-04-4eee72d56791" and that implies that TensorFlow does not work with this card, at least not with Ubuntu.

Do you know anything about the GT710? That supports CUDA and from the wikipedia page it implies compute version 3.5.

Would the EVGA GeForce GT 730 1GB GDDR5 Low Profile Graphic Card work? It seems small enough even with the fan.

The GT 1030 supports CUDA - unambiguously, and in ways both more flexible and dramatically faster than the GT 710. The link you suggested I search for indicates the 2 GB framebuffer is the big stumbling block for TensorFlow. Any problems you have with the GT 1030 to those ends will be much worse with low end Kepler gear.

So any other recommendations? You mentioned the EVGA GeForce GT 730 1GB GDDR5 Low Profile Graphic Card but not with fan. Or is the Geforce GT 1030 worth a shot now that it's been out a while and updates may have fixed the issues?

RobbieTheK wrote:

So any other recommendations? You mentioned the EVGA GeForce GT 730 1GB GDDR5 Low Profile Graphic Card but not with fan. Or is the Geforce GT 1030 worth a shot now that it's been out a while and updates may have fixed the issues?

Your best bet's probably to hold out for a low profile GTX 1050 Ti. It's capable enough to be worth plunking down money for, and 4 GB of memory should allow for an array of GPGPU applications to run on it without choking. I know that isn't a cheap option, but that's where things stand.

I agree with @Concupiscence. If a 1030 with 2GB of RAM is going to be too slow due to framebuffer issues, a 730 or 710 with even less memory is going to be even slower.

I don't think there are any 1050ti cards that will fit in the server @RobbieTheK is using, though. The limitation is single slot, with a short PCB length. Those are nearly all double slot, and they are also usually longer than would fit.

Possible a Quadro would be an option. P1000 has 4GB, that will fit in your system. K1200 as well (actually a maxwell-based part, despite the name). The larger cards, like the P2000 or previous gen K2200 are going to be too long - the PCB is a good length but the fan hangs over the back of the card too far.

On a side note, Don't be surprised if NVIDIA doesn't list CUDA support on their product page. They also don't have their "compatible DX12 cards" page updated to include the GeForce GT 1030 or 1050 cards, even thought they support it just the same as the other GTX 1060, etc. It was probably updated during the initial Pascal launch and then, like any corporate website, no one ever went back to update it later. The 1030 isn't a card designed to really appeal to anyone doing compute, they're basically just trying to sell them to low-end/mainstream esports gamers, so marketing probably didn't bother to include it as a bullet point.

Just throwing another option out there for people... I stumbled across this while trying to build the smallest portable system (with a decent gpu) to put in my luggage on trips.

The Lenovo M910x Tiny:

7.05" X 7.20" X 1.36"
179mm X 182.9mm X 34.5mm

2x ddr4 sodimm, 2x m.2 nvme/sata ssd, rx460 (not the best, but not terrible)

Have to use a T cpu when using the rx460, but that's not the end of the world either.

Specs here: http://psref.lenovo.com/syspool/Sys/PDF ... 20Tiny.pdf

Some pictures here: https://forums.lenovo.com/t5/ThinkCentr ... 943/page/2

I am looking for a graphic card for my Dell Optiplex 9020 SFF with i7-4790, mainly to improve video editing speed.
This is part of a review in Notebookcheck (notebookcheck.net/MSI-GeForce-GT-1030-2GH-OC-Review.236720.0.html) of the MSI GeForce GT 1030.
It is the final sentence I find interesting/confusing. "All six cards lack additional power connectors and are thus powered through the PCIe slot alone, and have a TDP of 30 W. This is low enough, even for the passively cooled models that use up two slots in order to ensure proper cooling. Accordingly, the actively-cooled models are smaller and only use up a single slot. Despite the fact that the cards only use four PCIe lanes internally, the external connector requires a PCIe x16 slot."
Does this mean the MSI card only used the x4 slot (so there should be little performance difference in using a single slot) or does it say the card will not work without the full x16 slot? I note recently the Nvidia GT1030 was shown to work OK so might the MSI have a different construction?

Hi everybody. I am new here, I registered to give some hope to all those that are looking for single slot low profile GPU. I finally got my ASL G1504 GPU (Geforce GTX 1050 ti). As soon as it arrived I ran benchmark on my recently obtained Radeon R7 450 from Dell. I swapped the cards, installed necessary drivers and re-ran the benchmark....... What a difference....... Dell card needs a new home now
On Unigine Valley Benchmark GPUs scored:
Dell Radeon R7 450: 846 score, avg FPS 20.2
ASL G1504: 2578 score, avg FPS 61.6

Resolution: 1920:1080
Quality: High

Said all that, I truly believe ASL G1504 is the best option out there for a single slot LP GPU. Make sure you get it from a legit source. I got my unit through GearBest.
I hope it will come handy for all those that have been looking for slim card.