AMD FirePro R5000 Introduction



Company: AMD
Author: James Prior
Editor: Charles Oliver
Date: February 24th, 2013

AMD FirePro R5000 Introduction

Feb. 25th 2013, AMD unveils the new FirePro R5000 card, designed for remote graphics. This card is based on the Pitcairn ASIC from the Southern Islands family of GPUs, currently seen in the AMD Radeon HD 7800 series (excluding the XT) and the FirePro W5000, W7000 and S7000 cards. As you can see, AMDs differentiates their target market for their product with a prefix: W for Workstation, S for Server, and now R for Remote replacing RG for Remote Graphics. The AMD FirePro R5000 isn't just the W5000 with a new name, it has a new capability : PC over IP (PCOIP) that enables graphics over internet protocol Ethernet networks, first seen in the FirePro RG220.

AMD FirePro R5000

The GPU itself features three outputs, two mini-DP 1.2 display outputs and an RJ-45 jack with Gigabit Ethernet support. The card features TeraDici's TERA2240 chip which encodes and encrypts the display output of the GPU at high resolutions. How high? Dual 2560x1600 resolution at 60fps, or four 1920x1200, also at 60fps. The video stream encryption can be 256-bit AES, making it suitable for broadcast over IP networks without the need for a dedicated one to one VPN tunnel. AMD claims no competition from NVIDIA, certainly they have nothing similar off the shelf, and their comparison product is a $2K+ USD KVM extender over fiber network device. AMD was coy about pricing in our phone briefing for some reason - they'll give out all the technical details but not the price, not exactly a vote of confidence in their embargoed participants.

The FirePro R5000 replaces the existing RG220/220A products currently in the market, as it offers a much higher level of performance - the RG220 uses a tweaked 'Luigi' RV710 chip - RV711 - and has a TDP of 35W; the new R5000 is 150W and offers orders of magnitude more performance. With the RG220 product you need two cards to support quad displays, this is now reduced to a single R5000, which also feature upgrades like support for DirectX 11.1 / OpenGL 4.2 and OpenCL / C++ AMP support, and continues AMD's FirePro ISV certifications. Anyway, the FirePro W5000 retails for around $450, the FirePro RG220 for around $500, so it seems likely that the new card will jump in around the same price point to offer a large bang for the buck in this market. If AMD opts for a new higher price point, closer to the $1000 mark, it's still a remarkable value compared to Matrox's solution but reduces the value proposition vs. the traditional use cases for FirePro W5000 performance cards.

Competitive Landscape

So why does this product exist, and what is it used for? A big trend for computing is the cloud, under the banner of 'move the data, not the compute'. There are many organizations, both for profits business and academia, with end users that have heavy computing needs, think CAD/CAE, media/entertainment processing, that rely on both CPU horsepower and GPU horsepower. The current practice is to put high end high cost workstations under desks, assigned to individual users. This gives three problems to the organization: data security/protection, 'silo'ed resource utilization, and lack of mobility to move with workforce/projects.

Power User Target Market Segments

Looking at the first issue, many of these users will have local datasets. If this data is needed by multiple people there either needs to be a time consuming check in/out procedure or a maintenance window when backup and integration into the global dataset for the project or company is performed. The window for data corruption to occur is a worry to many businesses, as is the window for stupid user tricks to expose the dataset to the outside world. This can be through seemingly innocuous web surfing and getting exploited via zero-day attack or bad decision making on searches/link clicks, web filters only go so far when even website like MSNBC can be unwitting hosts to malware. The rest of the users will have a client/server connection to the server, querying the data needed to processed or updated. This has the advantage of simplifying backups, integration and updates, but adds processing load to both client and server, adds latency to client software, and increases bandwidth requirements.

PC over IP

The current obvious solution to the problem is virtualization: move the workstation into the server room and all the inherent benefits therein - high power server clusters with fat bandwidth connections, gigabit Ethernet is nothing compared to modern Fiber Channel networks, Infiniband, and 10GbE connectivity. However, what virtualization doesn't do well yet is graphics virtualization - getting GPU power to your virtual desktop is tough, and we solved one problem and made another. Why not just put the workstation as a server in the server room then, and remotely access it?

Benefits of FirePro PCoIP

What the remote graphics solution permits is for the user to be on a thin or software client connection to a centralized server and all the processing to be done on that remote rack mount workstation or server, streamed back to the thin client or software session on your compatible computing device. You see the pixels, it sends back USB to the FirePro which makes it a kind of IP KVM with great performance, it's got good enough quality to be used for professional graphics work and refresh speeds making it seamless. Now you've got your business app running in the datacenter, protected and supported, and you can indulge your non-business related tasks on the local client without exposing the data or server to it. Win-win.

The best part is that you don't need to do this on a one to one basis, you can actually virtualize the workstation and use ESX or XenServer's directIO passthrough feature to provide the VM with exclusive, native access to the GPU. This is beneficial for density, the economies of scale show value for this approach. Replace dozens of 1S or 2S workstations with 2S or 4S servers running multiple VMs with DirectIO provided GPUs and you've got the instant benefits of virtualization available, such as thin provisioned storage, memory over commitment, increased CPU utilization, distributed resource scheduling, high availability, shared OS base images and so on. The end user can be connected via Virtual Desktop Infrastructure and/or thin client, or from a separate end-user application for use either on a user provided device or company assigned asset.

Remote Graphics

The reason for the TeraDici chip and Ethernet port is that to provide the best connectivity available, the remote access has to be smooth as well as secure. Latency needs to be minimized, hence a direct connection to the card rather than just an interpretation layer is available. As noted previously this makes a KVM over IP style solution available, sending back USB 2.0 over the connection to the card which also supports wake on USB and wake on magic packet functionality; energy saving is important to all levels of business now and the ability to remotely wake a datacenter hosted workstation or VM is important.

The last matter is mobility, as companies look for talent (or cheap workforces) around the globe, the ability to connect from anywhere is becoming more important. With essentially a streaming workstation environment a company can host a datacenter where they want and have a dispersed workforce, either because they offer professional services that leverage home office's horsepower to Git-er-dun, or use time zones to increase productivity - moving shifts to areas where it's close to normal business hours that white collar workers with the skills they need want to work, permitting the dead 16 hours a day the infrastructure isn't used to become partially or wholly used. Efficiency ++.

The two mini-DP ports are used for local connectivity, for mirroring the remote display. This permits troubleshooting and training operations to be simplified and speedy, no need for additional latency introducing products in the mix. DisplayPort 1.2 permits display chaining, and Dell just introduced a line of 2013 monitors supporting this feature, so it should be possible without much effort to host 4 displays from the card to allow this.

TeraDici TERA2240 Chip

The AMD FirePro R5000 offers several opportunities for businesses needing GPU performance and remote connectivity with great performance, as well as for organizations looking to optimize their spend on infrastructure and support. It also offers hosting providers a simple way to offer SaaS, with a familiar interface for the customer - a whole OS environment with dedicated GPU performance and ISV certification. This can be leveraged by businesses looking for short term projects completed by contractors to be supported without going to the expense of leasing or buying new equipment, or reducing overhead by moving to hosted high performance infrastructure.

The concept as a whole is sensitive to large scale disruption events like Hurricanes, Sandy showed us how payment processing for businesses all over the East Coast of the US was disrupted by damage to so called cloud provided services, as well as Netflix. As always there is risk with business, you have to decide where you want your pain point and how much you want to pay to mitigate it. The old school form of thinking is if you own the equipment, you can yell at the guy working on it till its fixed; that's good management right thar, I tell you what. The new school says I don't care who, I care about service and the contract says you'll provide this or pay me for loss of business, and if you can't there's another guy down the street who'll take my money and get me back in business quickly; that's free market competition and I'll be alright.

The licensing complexities from TeraDici owning PCoIP mean that AMD aren't using their own video encoding abilities and compute performance for encryption, instead using the TeraDici chip on the card. Perhaps if the partnership is a success there will be a Tahiti based card and a new TeraDici chip featuring 10GbE connectivity and 120Hz/Stereo 3D support. An APU FirePro would be a good idea as well, especially if heterogeneous compute business applications take off and APU's enter the datacenter; AMD helping to bring TeraDici into the HSA foundation would be a very smart move. Which brings us neatly to why you, Rage3D reader, gaming enthusiast, performance enthusiast, should care about the AMD FirePro R5000? Well, if I were a company looking to move into being the technology provider for cloud gaming, surround computing, and media entertainment companies, I'd probably want a strong, enterprise based relationship with the guys who hold the patents and the expertise for doing an integral part of that paradigm right, before I needed it. Wouldn't you?

At full height and length but single slot it's a smart size for rack mount workstation and servers, this is a smart product from AMD offering the right level of performance for the targeted customers and has all the right certifications, and it fits into existing infrastructure simply - no proprietary or high cost devices required. AMD should definitely work with one of their partners like HP, ASUS, MSI or PC Partner (Sapphire/Zotac) to bring an embedded APU thin client to market, complementing the FirePro R5000.