A common question I am guaranteed to get asked almost on a daily basis: ‘Why should I pay so much more for NVIDIA’s Enterprise series of cards over the consumer offering?’
So, let’s put you in the situation that a lot of people find themselves in; you are in charge of a project and you want to purchase the correct platform for your users (GPU based server or workstation), so you have two choices: you can either go for the consumer cards or the professional series of cards. But other than the massively different price points, what are the primary differences between these products?
If you would like to see more articles by myself or the Novatech team please go to: https://www.novatech.co.uk/deeplearning/articles/
Let’s start off with the price. Currently you can purchase a 1080TI Consumer GeForce card for around £800 inc VAT (highest performing card in the series). Now, let’s look at the same category card in the professional series; V100 is currently the flagship product for NVIDIA and will set you back around £10,000 inc VAT. As you can see there is a big difference, (£9000 based on today’s prices.) For such a big difference, you will be expecting a lot in return!
NVIDIA working closely with Independent Software Vendors (ISV’s) such as SAP, Autodesk, and many others to ensure these applications are fully certified on NVIDIA professional cards. Now in contrast, GeForce (Consumer cards) do not have the certifications meaning that users can suffer from with a lack of performance. If we just look at Tesla, then we can see optimisation and certification applied on the cards in the following category areas:
- Computational Finance
- Climate Weather and Ocean Modelling
- Data Science and Analytics
- Defence and Intelligence
- Deep Learning and Machine Learning
- Manufacturing: CAD CAE
- Media and Entertainment
- Medical Imaging
- Oil and Gas
- Research: Higher Education and Super Computing
- Safety and Security
Now if we look at one of the above category sections like Media and Entertainment, we can see hundreds of certified applications. We have to understand and remember the certification process NVIDIA delivers is designed to increase the user experience with the product in the desired area of use. Now this process obviously costs NVIDIA a lot to implement and maintain. The benefit to the user is clear, rock solid platforms with little or no instability, far greater performance, working in the most optimised and efficient way.
ISV’s will normally always support customers who run their applications on certified NVIDIA products. If you were using GeForce products, I imagine your support levels would be between minimal to non-existent.
For the Deep Learning crowd that I am normally speaking to, think about Frameworks and what NVIDIA has done to improve the performance of GPU’s using these. Frameworks with the optimised versions by NVIDIA, in some cases, work at X2 speed for non-optimised Frameworks.
For most users, Graphic card reliability is of paramount importance; in today’s modern Data Centres, it is of critical importance to eliminate any down time. Downtime is basically the worst KPI for a professional user that you can come across. For system support or system administrators, it is obviously a very serious concern as well because typically these individuals are targeted against down time!
So how is NVIDIA managing downtime on its professional series of product? (Note: The below are examples only and do not apply to all products).
- Thicker PCB boards that can handle heat dissipation are generally tougher.
- Thicker gold plating on PCIe connectors which offer better electrical integrity and corrosion resistance.
- Components on professional cards such as fans are very carefully selected to ensure they provide high reliability and long life.
- ECC (Error Correcting Code) memory on high-end Quadro cards to ensure data integrity by correcting for occasional single-bit errors.
- In most cases significantly larger memory capacity, this will result in far greater performance for applications using AI, textures, vertex buffer objects, CAD, image manipulation.
- Designed and built by NVIDIA to a single specification for 24x7 reliability and stability.
- Unified driver with deterministic schedules for better partner planning.
- Certified by leading OEMs.
- Application specific testing by NVIDIA.
Unlike consumer cards, professional cards from NVIDIA have to undergo a battery of hardware tests to ensure that they meet NVIDIAS very high standards. It’s also important to note that unlike popular belief, Professional cards from NVIDIA are not identical to the consumer versions. It is true that the base architecture is the same, but very early on in the production process they branch off and go on different journey as identified above and below. Also, all products have to undergo but are not limited to:
- Thermal tests
- Shake and shock
All of the above are designed to prove that the card will operate with sufficient margin even under even the most of demanding of environments. This is a key point to understand; Professional cards are designed to handle workloads in environments that the typical consumer cards could never withstand. Imagine a consumer card being used in a server room environment, running 100% of the time, typically under full load, and probably running at far higher temperatures than normal. In short, you are stressing the card to limits that are far greater than it was designed to do, so evidently there is a far higher percentage that the product is going to fail, or not return the performance curve you desire.
The manufacturing process is very important here as well. Consumer cards which are after-launch manufactured by a series of OEM’s (Original Equipment Manufacturers) and their testing and assembly process can offer varying degrees of quality. However, with the Professional series this is not the case. The manufacturing process is very tightly controlled by NVIDIA to ensure that only the highest quality product is associated with the name Quadro and Tesla.
Enterprise grade product with enterprise grade support:
Warranty and support with consumer cards is a world apart from what you will experience with the professional series of product. In short, when a consumer card fails, you will be dealing directly with the board partner (EVGA, MSI, ASUS), and as such you will be relying on them to help you. For the most part, the consumer board partners are very good on warranty and support, however the systems are not designed for immediate replacement of product (24-48 hours timeframe) and their customer support teams would also find it hard to support you, as they are designed for consumer card issues, not enterprise grade support issues.
Imagine you have purchased consumer cards to be used in a server room environment or to go directly into a Data Centre, but the key thing is that regardless of environment you are NOT using the cards for playing games. Now imagine you have a double bank holiday weekend and on the Thursday night you start a training run on some models you are working on, or equally you could be running any type of workload not just ML ones. Then when the card fails, because of the double bank holiday you get no customer support from you card manufacture. Then on the Tuesday after the bank holiday due to several different reasons like system admin not being available, you are still unable to process a support ticket for your problem until later in the week (Thursday). THEN you have to wait for 1-2 weeks for the board partner to replace your product or fix the issue.
Well the above is not so imaginary, this actually happened to a client I know and as a result the system the card was in was down for nearly three weeks before the problem was identified and fixed.
If you are a consumer and your gaming system was down for three weeks this would cause a serious problem for you! No CoD for three weeks (nightmare, as my kids would say) but imagine if this happened to a business, a business that depends of these systems being in constant usage. For the clients I deal with, this is simply not a scenario they would except.
It is also very important to note that the refresh rate or product lifecycle is far longer for Enterprise class products. This is because NVIDIA purposely throttles back the releases of new Enterprise products as they know and understand that Enterprise class customers don’t want to be doing constant refreshes of technology, whilst insuring that the product in the market is constantly updated with new software versions to maximise on performance.
- Global technical pre-sales and post-sales support
- Long product lifecycle
- Constant product updates
- Deep workflow experience across vertical industries (Education, Manufacturing, Healthcare and so on)
- Enterprise class support
- Guaranteed response times of max 1 working day
- Regular maintenance releases for all versions of CUDA
- Dedicated portal (knowledge, FAQ, troubleshooting, articles)
- Long term software support 3 years availability from CUDA 6.0
For a full list of additional services: www.nvidia.com/enterpriseservices/additionalservices
So when should you choose the more expensive professional card over the cheaper consumer card…?
Before I summarise, let me add this point: Consumer cards are basically sports cars, built for speed first. They run fast and hot and are NOT designed to run at full speed for long periods of time. Think F1 cars, they run very fast and do a fantastic job in the correct environment (Silverstone, Hockenheimring), but try racing them in a typical urban environment… you could experience some serious overheating, and will you really use the available performance of the car sat at a traffic light?
Professional grade cards however are not only built for speed, but for endurance and stability. They are designed with all three aspects in mind, as this is what’s most important to Enterprise environments. So, while I use the comparison of F1 cars and consumer graphic cards, I will say for the Enterprise cards you are the driver in the Dakar Rally. You need speed to win, and you need fantastic reliability to make it to the end of the race, and you also have to be intelligent to navigate the different terrains you will encounter!
Maybe in the past the differences between professional and consumer cards were much smaller, so the argument for professional cards was harder to justify. But when we look at the differences we see today, I feel the gap if now far wider than it’s ever been previously.
So what is the best choice for you?
In preparation for writing this I did some homework to see what information can be freely found to help users understand what’s best for them. I found a lot of outdated information, the information I found was not easy to locate, or did not justify the points clearly. But more importantly from the customers Novatech deals with, there is a massive knowledge gap here. People simply don’t know why they should choose one over the other. The other issue is there are too many pessimists on the forums blasting NVIDIA for the changes made to its EULA.
As a professional reseller of NVIDIA products, Novatech will always recommend users to adopt Tesla or Quadro over consumer cards. The reason for this is simple, and there are three reasons of note:
- The amount of downtime that can be associated through incorrect purchases has a massive effect on companies who rely on these platforms running 24x7, not to mention the increase in workflow output.
- Since the EULA changes made by NVIDIA it is simply not worth the risk anymore. however with the combined risks I have mentioned above it seems like an illogical course of action to take?
- Rightly or wrongly NVIDIA have a new direction when it comes to the segmentation of their products. It’s correct as NVIDIA knows it products better than anyone, and I feel this is NVIDIA trying to maintain their market leading position by insuring that customers get the correct product for the correct environment / workflow. They have to do this to insure the experience people are paying for is given to them, but more importantly NVIDIA want the customer satisfaction levels to be as high as possible and that means correct positioning of products. This is the direction NVIDIA has set, and as users we have to accept and understand this. I don’t believe like some forums suggest that this is NVIDIA maximising their current market position. NVIDIA are a company set up and managed with engineers at its core. This is them making sure users are getting what they pay for, besides I don’t think currently NVIDIA has a pressing desire to show more growth than its already getting to shareholders.
So, what to do?
I think a pinch of logic is needed here. Let’s say you are a big bank or any other big organisation; I think the argument between professional or consumer card will be very straight forward and uncomplicated one.
But what if you are in academia, or a company of reasonable size (SME) and budgets are tight, and ROI has to be proved at all times? Would the argument for consumer cards be stronger… it could be, yes, but how do you know for sure? In this instance you should do the following:
Speak to an authorised NVIDIA reseller like Novatech.co.uk and ask for professional advice! If they don’t mention the things I have discussed above, then they maybe are not positioned correctly to help you
Contact NVIDIA directly, explain your situation and take advise from them
Proof of Concept speak to NVIDIA they will recommend you to certain resellers who have the capacity to offer PoC units to you for testing.
Once you have done the above, only then will you have the correct information to base a judgement on. Don’t go into this on your own, rely on the partner networks NVIDIA offers, speak to NVIDIA directly, and use PoC services.
Academia is slightly different to most normal adopters due to the fact that they are normally very focused on budgets and ROI, but also in Academia they may be using the product in a singular fashion, so not utilising the full advantage offered by NVIDIA due to the nature of the experiments or usage of the card. However, I do not see this as a problem… again the key thing is to communicate with your reseller of choice, if you feel they are not able to advise you correctly you should be speaking to somebody else. In my opinion in the UK there are only a handful of real GPU specialist resellers who know their stuff. Whenever I have spoken to NVIDIA about a problem that a University is facing with GPU alignment I have always found them to be very understanding and nearly in all cases a solution is found.
If you would like to engage with Novatech on this matter, or any matter to do with NVIDIA technology, then please speak to one of my team who will be happy to assist you. (Paul Sansbury [email protected] 02392 322 594, Luke Tiernan [email protected] 02392 322 794 to see how we can assist and support your projects)
Novatech are specialists in GPU / Deep learning / Machine Learning platforms.