GPU Dedicated Server

GPU Dedicated Server
Hey guys, I am looking for the following GPU dedicated server in US or Canada.
Can anyone point me in the right direction?
++++++++++++++
list of top cheapest host http://Listfreetop.pw

Top 200 best traffic exchange sites http://Listfreetop.pw

free link exchange sites list http://Listfreetop.pw
list of top ptc sites
list of top ptp sites
Listfreetop.pw
Listfreetop.pw
+++++++++++++++
Dual Xeon E5-2680 V2 | or passmark equivalent
128GB Ram (must have)
2 x 1TB SSD
Less than 100 TB
RTX 2080 or RTX 2080TI (must have)
Price Range $350 - $400 USD
This would be easier to see with some 1080Ti's.

Are you looking for month to month on these or able to commit to a longer term?

How much bandwidth do you require?
PUREVOLTAGE.COM Premium Colocation, Dedicated & VPS Hosting
? Custom built Dedicated Servers to fit your needs. AMD/Intel servers, 10G,20G+ High Bandwidth
I would prefer month to month but would consider a longer commit if the price was right.

I do not require a large amount of bandwidth, somewhere around 50tb / month on a gbit connection.
I would consider some 1080ti's.
You realistically could have a ton of potential options depending on what you are doing.

What kind of workload is this?
Hi Brandon, the workload is mainly video editing and content creation (Mudbox, Maya ect.).
So you'll be running Autodesk Maya and Mudbox on RHEL/CentOS then or on another platform?
No I will be running a Proxmox hypervisor with different vm's (Windows being one of them).
That could get really messy, generally VMs don't like sharing the VGA card, you'd probably be better off running on bare metal and RHEL/CentOS, though you could also go with windows.

I'd recommend 1 to 3 bare metal AMDs each with 1 or 2 2080s, or a 5700 or higher PCIe 4.0 and a 2080 PCIe 3.x, with 2 or more discrete VGAs you'd be able to assign a dedicated VGA to 2 VMs provided the virtualization used supports that.

Most AMD Ryzen 9 boards have 3 or 4 PCIe slots however the PCIe 4.0 lanes are limited. The Threadripper has 64 PCIe 4.0 lanes and the Epyc2 Rome has 128 lanes.
Would it be possible for me to email you? I have a few questions.
Let me review the forum rules.

Are you considering 2 PCIe 4.0 NVMe SSDs? Your original post indicated you needed 2 * 1TB SSDs. Those would occupy 2 PCIe slots, or 2 M.2 slots, unless you go with U.2 or SAS, more specifically SAS-4 or perhaps SATA SSDs.
Based on what I have seen being offered by dedicated server providers, most offer SATA SSDs with an expensive option to upgrade to NVMe. If the upgrade to NVMe was reasonably priced I would consider it.
NVMe performs much better, especially when used with PCIe 4.0. One of the PCIe 4.0 capable drives can read at up to 5000 MB/s can write at up to 4400 MB/s and offers up to 750K IOPS for reads and up to 700K IOPS for writes. Real world performance will vary.

share.payoneer-affiliates.com
x domain y
hosting forum romania
host ip
earnit.eu
gslink.co
simpleadprofitz.com
host x reader

I believe you're at 10 posts now and you should be able to use the WHT system once the cron runs.
| AMD Epyc | AMD Ryzen 3/5/7/9 | Intel i3/i5/i7/i9 | Intel Xeon | ARM devices |
This would be easier to see with some 1080Ti's.

Are you looking for month to month on these or able to commit to a longer term?

How much bandwidth do you require?
Really? I don't see this happening at all at the ~$350-$400 price point: at least not reliably.

If you factor in the requirements for 50TB of data on a Gigabit line plus the GPU needs, I haven't seen anybody offer such a deal.

Of course, a long term contract might change that; especially if it's an upfront contract versus a 2-year agreement with monthly installments.

Aside from that, running applications like Maya and Mudbox (or really anything Autodesk) would be better served on the Quadro line versus the RTX line. I'm not sure how many hosts even keep RTX cards on hand.
Well I can tell you I have already received multiple quotes within my requested price range. A few offered RTX 2080's while most offered GTX 1080ti's. I agree with your comment regarding the Quadro gpus but the huge premium being asked for them over the 1080ti's and 2080's is not worth it to me.

Should I be concerned about accepting offers in this price range or are you are confusing cloud pricing with dedicated server pricing?
Honestly 1080ti's have come down in price quite a lot. The cost you get quoted depends on how motivated the seller is to be honest. Like if they have a lot of GPUs left over from mining, and it's been slow finding customers who want to use them for video editing type purposes like you. Also, easily half the cost of running a GPU server like this is power. If they've got a good deal on power, or have some left over for some reason, it's easier to offer a competitive price.

One main reason you'll see mostly 1080ti's, is those were available and popular for mining, whereas the mining boom died out before the launch of the 2080 / 2080ti. So there's just a lot more low cost inventory floating around of the 1080ti's.

For your use case, that GPU is perfectly fine anyway. The 2000 series really only is much better if you need "tensor cores" for deep learning, which gives the 2000 series a 2x speed improvement in that use case. For every other use case, a 1080ti is basically the same as a 2080, and a 2080ti is only marginally better at twice the price or more. Really not worth the premium going with a 2080 or 2080ti in that situation.
IOFLOOD.com -- We Love Servers
Phoenix, AZ Dedicated Servers
1Gbps and 10Gbps Unmetered Servers In Stock!
Several RTX 2080 Ti models are in the $1,500 to almost $2,000 range. There's a couple of models that are cheaper but they're still over $1,000 each. The 1080Ti still retails for around $1,300.

The cheapest 2080 Ti's sell for a bit less than the 1080Ti does. I wouldn't necessarily recommend used cards especially if they've been mining 24/7 for months or even years or if they've been abused.

The Titan RTX, which I probably wouldn't recommend, currently retails for right around $2,500 on amazon.

US Pricing as of Dec. 8th 2019 my source was google and major online stores like amazon, etc.
Several RTX 2080 Ti models are in the $1,500 to almost $2,000 range. There's a couple of models that are cheaper but they're still over $1,000 each. The 1080Ti still retails for around $1,300.

The cheapest 2080 Ti's sell for a bit less than the 1080Ti does. I wouldn't necessarily recommend used cards especially if they've been mining 24/7 for months or even years or if they've been abused.

The Titan RTX, which I probably wouldn't recommend, currently retails for right around $2,500 on amazon.

US Pricing as of Dec. 8th 2019 my source was google and major online stores like amazon, etc.
For 2080ti in a server, you'd want to use a "blower fan model". PNY has one that is MSRP of $1400, but commonly available for $1200. I've seen used ones going for around $1000.

We had a lot of 1080ti (still have some left), that we've been selling on ebay. Tend to go for around $450 for a typical card.

So you're looking at, at least twice the price for a 2080ti. For deep learning that's worth it as you get something like twice the performance for a similar amount of power use. Whereas, for any other use, it's what, 10-20% faster for twice the money? Not really worth it.

As to used cards, I mean, as soon as you start using it, it's used. Every dedicated server you rent is a "used" server. If you're getting a rental server, any failures are the responsibility of the hosting provider to correct. The worst common issue is overheating -- either due to a poor system design, gpu fans not spinning up properly (can be corrected in the driver settings), or one or more GPU fans failed. As long as the nvidia driver reports GPU temperatures below 75C, I wouldn't worry about it. If it does show above that, ask the host to address the overheating issue. The cards will throttle their speeds down above 80C.

It's a different story if you want to buy your own GPUs for this use case. In that situation sure, getting a GPU that's been abused can be a bad investment, and it's tricky to know if you're looking at a good card or a bad one. In terms of being a dedicated server customer, those issues are the hosting providers problem and not yours so I wouldn't worry about it.

Personally, we've done hardware mods so the fans run 100% all the time regardless what the driver tells the card to do. We've got driver settings to do the same thing, but it's not very user friendly, so for customer use it's easier to get a "mini gpu fan connector" to "normal 3 pin fan connector" adapter cable and use that.
IOFLOOD.com -- We Love Servers
Phoenix, AZ Dedicated Servers
@funkywizard pretty much nailed it, while buying a new system for this it's quite costly however for the 1080Ti's there are a lot of those kicking around that providers have from mining.

Budget is tight however it's doable more so for 1-2 cards being needed once you require more that requires a more customized solution, which in turn raises the costs quite a bit.

Power costs are another big cost involved with GPU systems which even with a surplus of parts it costs quite a bit from the power usage.
PUREVOLTAGE.COM Premium Colocation, Dedicated & VPS Hosting
@funkywizard pretty much nailed it, while buying a new system for this it's quite costly however for the 1080Ti's there are a lot of those kicking around that providers have from mining.

Budget is tight however it's doable more so for 1-2 cards being needed once you require more that requires a more customized solution, which in turn raises the costs quite a bit.

Power costs are another big cost involved with GPU systems which even with a surplus of parts it costs quite a bit from the power usage.
Yeah, I agree with that. With 1 or 2 cards, there are a lot of very "typical" servers out there that can accommodate that with little or no modification. Up to 3 cards as well is not too terrible, but harder to keep cooled properly. Beyond 3 or 4 GPUs in one server, you're looking at needing a GPU optimized server platform, which can be pretty expensive.

As Jake mentioned, the power is one of the biggest costs. This is especially true for deep learning where the server can be maxing out the power use for an extended period of time. In that situation, a host has to make sure that power is always available when needed, which can be rather substantial for 4 - 8 gpus in a server. For one or two cards, especially for a less intensive 24/7 use case, you can sometimes fit that in where you've got a bit of extra power and thereby bring down the cost a bit more.

Even so, not cheap -- easily half the cost of what you asked about is power.
IOFLOOD.com -- We Love Servers
Thanks guys for your feedback, I think dual GTX 1080ti's are my best option right now.

I have a question with regards to gpu power consumption and how it affects a dedicated server quote initially and moving forward . A quoted price takes into consideration the anticipated use / workload being run on the server because gpus can potentially use a lot of power if they are running hard frequently. If my needs change in the future, and for example I am running a lot of rendering jobs (and using more power than I was previously) how do providers typically deal with this kind of situation? What should I expect to happen from my provider?
I agree with @Purevoltage and @funkywizard very good points

The blower is good at pushing the hot air out the back which is usually desirable, but I've had blowers fail before and usually prefer a 3 fan design versus a single blower. Extra case fans are also a good thing to have as @funkywizard said.

As far as the power usage goes, if the server is rendering 24/7 or close to it you might start to trip the circuit breakers and a few components might fail prematurely. If there's any components that aren't fully up to spec or have defects, they're more likely to fail under heavy load.
| AMD Epyc | AMD Ryzen 3/5/7/9 | Intel i3/i5/i7/i9 | Intel Xeon | ARM devices |
| Custom Server Builds | Server repair | Server Upgrades | System and Network admin |
Thanks guys for your feedback, I think dual GTX 1080ti's are my best option right now.

I have a question with regards to gpu power consumption and how it affects a dedicated server quote initially and moving forward . A quoted price takes into consideration the anticipated use / workload being run on the server because gpus can potentially use a lot of power if they are running hard frequently. If my needs change in the future, and for example I am running a lot of rendering jobs (and using more power than I was previously) how do providers typically deal with this kind of situation? What should I expect to happen from my provider?
Normally this shouldn't be a problem, however best to ask the specific provider ahead of time just to make sure.
PUREVOLTAGE.COM Premium Colocation, Dedicated & VPS Hosting
Thanks guys for your feedback, I think dual GTX 1080ti's are my best option right now.

I have a question with regards to gpu power consumption and how it affects a dedicated server quote initially and moving forward . A quoted price takes into consideration the anticipated use / workload being run on the server because gpus can potentially use a lot of power if they are running hard frequently. If my needs change in the future, and for example I am running a lot of rendering jobs (and using more power than I was previously) how do providers typically deal with this kind of situation? What should I expect to happen from my provider?
For the provider it's a roll of the dice. After making the sale there's not much they can do about it. It's usually more relevant if someone wants to use GPUs for transcoding, because that will typically use no more than 75w even if the gpu has a tdp of 250w. For transcoding I would typically recommend one or mote GTX 1070s, however. The transcode performance is nearly identical, the gpu costs less money, and if for some reason the card is not used for transcoding, the maximum power use is more like 150w, vs 250w for a 1080ti, so it's less risky to quote based on the lower transcoding power consumption.

For anything other than transcoding, the provider needs to set aside enough capacity on their power circuits so that you can run the cards at full power. A cost difference between 24/7 use or not mostly comes into play if the provider is charged a metered usage fee for power. Even then the majority of the cost is for the equipment designed to keep the server online -- generator, ups, cooling -- and not the electricity itself. Those items have to be sized for peak load and not average load, so in most cases a big price difference only comes into play if the gpus are used for transcoding. Thereare some edge cases where it will cost a little less for occasional use vs 24/7 use, but not by a huge margin.

Where you might see this impact you as a customer, is if the provider really closely monitors things, they might quote you a higher price for future servers if you're using substantially more power than they budgeted in your initial quote. Again that would make more sense if you were getting a discounted quote on the assumption the use case was transcoding, lessso for other uses.
IOFLOOD.com -- We Love Servers
I agree with @Purevoltage and @funkywizard very good points

The blower is good at pushing the hot air out the back which is usually desirable, but I've had blowers fail before and usually prefer a 3 fan design versus a single blower. Extra case fans are also a good thing to have as @funkywizard said.

As far as the power usage goes, if the server is rendering 24/7 or close to it you might start to trip the circuit breakers and a few components might fail prematurely. If there's any components that aren't fully up to spec or have defects, they're more likely to fail under heavy load.
Blower is really the only options at least when it comes to 8 GPU's in a system that we have seen otherwise there could be a lot of issues.
We have seen some big issues in 1U systems as well that an AI client brought to us, since then we have built out multiple systems for them while the two 1U systems they had built can overheat and shutdown under large and long loads even with great cooling there is only so much air in 1U it can push. All of the 2 and 4U systems have never had any issues.

As for the circuit breakers this shouldn't be an issue unless circuits are overloaded which typically providers will max ports out at 80% usage so a bit of extra should trigger an alert.
PUREVOLTAGE.COM Premium Colocation, Dedicated & VPS Hosting
? Custom built Dedicated Servers to fit your needs. AMD/Intel servers, 10G,20G+ High Bandwidth
Enterprise Hardware, with 6 Global Locations - Seattle | Chicago | Dallas | Los Angeles | New York | Amsterdam
I agree with @Purevoltage and @funkywizard very good points

The blower is good at pushing the hot air out the back which is usually desirable, but I've had blowers fail before and usually prefer a 3 fan design versus a single blower. Extra case fans are also a good thing to have as @funkywizard said.

As far as the power usage goes, if the server is rendering 24/7 or close to it you might start to trip the circuit breakers and a few components might fail prematurely. If there's any components that aren't fully up to spec or have defects, they're more likely to fail under heavy load.
Triple fan is ok for a single gpu system with plenty of empty space around the gpu. For any other use case I would avoid it.

Once you pack two gpus right next to each other in adjacent pcie slots, you'll need all the help you can get keeping them cool in a gpu intensive workload such as mining or deep learning. The blower fan models do far better than triple fan models in that scenario. Even then, the gap is so small for the air to get into the fan intake on the obstructed gpu, you need high-pressure supplemental fans to keep the temperature down, even with the gpu fan at maximum.

This is why systems with 3 or more gpus are a niche item -- there's no way to keep them cool without a specifically engineered cooling solution. For mining, that can be "attach a pair of 15k rpm 40mm fans to the gpu exhaust grill". For a less MacGyvered solution, typically you'll see chassis filled with double-stacked high rpm 80mm fans, for example 8 gpu systems will have fans that draw a few hundred watts amd sound like a jet engine. And even then, blower fan model gpus with the gpu fan set to maximum is a must to keep them from overheating.

As to fan failures, it happens occasionally, but not as common as you'd assume. Out of 50 or so GPUs mining for about a year, 24/7 100% fan speed, we had 5 or so fans fail from triple-fan model cards, and one fan fail from a blower model. So it happens but not something to be super worried about. To be honest, we lost more fans due to security ID badges accidentally contacting the fan blades than any other reason.

As to failed parts, that's very unlikely (other than fans) at 24/7 operation unless the card is overheating. We try to keep ours under 70C if possible. The GPUs start to throttle above 80C, and are rated typically for operation at 90C. So there's plenty of headroom if you do it right.

One other point are water cooled gpus. The EVGA SC2 Hybrid 1080ti is an excellent card, getting you temperatures way lower than anything else, even if you pack them in like sardines. It's not a very "server friendly" solution however, trying to find a place to put all the tubes and radiators. It takes what would otherwise be a clean looking 4u server and making it a rather messy 5u or 6u solution. But it sure gets rid of any hassles with trying to cool the cards.