AMD Launches New Dual-GPU FirePro: S10000 Aimed At Datacenters, Virtualization
AMD has announced a new FirePro graphics card at the International Conference for High Performance Computing, Networking, Storage and Analysis conference today (hereafter abbreviated as SC12). It's aimed at server workloads and data center deployments, but it's also the most powerful GPU Sunnyvale has ever deployed, bar none.
Rumors of a dual-core Radeon 7990 have floated around since before the HD 7000 series actually debuted, but this is the first time we've seen such a card in the wild. Say hello to the S10000.
How does this card fit into AMD's product roadmap? That's a bit complicated. Nvidia has explicitly prepped the K20/K20X for servers and workstations and spent years establishing itself as a player in the HPC space. AMD, in contrast, doesn't have much of a presence in those markets.The S10000 is something of a jack-of-all trades and it's designed to tackle multiple markets at once.
On paper, the S10000 is a serious beast. We can't give you comparative specifics quite yet, but the single and double-precision rates listed above are higher than anything debuting today from Intel or Nvidia, as is the card's memory bandwidth. The flip side to these figures, however, is the eye-popping power draw: At 375W, the S10000 needs a pair of eight-pin PSU connectors and draws far more electricity at peak than anything else launching today.
None of this, unfortunately, makes the S10000 a shoo-in for the HPC/supercomputer market. AMD has ceded this space without a fight to Nvidia and Intel, and the company hasn't released any information to suggest its launching a new GPU initiative. Instead, the S10000 is supposedly aimed at the virtualization market. In theory, its dual-GPUs on a single-card are a good way to improve GPU virtualization density inside a single server; a single S10000 draws 375W compared to 550W for a pair of W9000's, while offering nearly identical performance.
The S10000 is rather clearly going after Nvidia's VGX K2, which is based on the GTX 690, but has a TDP of just 225W. Unlike the NV cards, the S10000 is equipped with video outputs, and can drive 4x Display Ports and a single DVI. It's not a bad move from AMD, but we rather wish the company had something like Nvidia's VGX K1. That card drops four mid-range Kepler GPUs on a single PCB and would seem to be a better ambassador for the benefits of GPU virtualization to a company looking to dip its toe in.
Then again, maybe we're wrong, and virtualization customers are banging down AMD's door to get their hands on faster cards. We discussed AMD's virtualization pitch when it debuted the S9000 a few months back, so we'll link you there if you have more questions. As for whether or not AMD will launch a W-version or consumer flavor of the card, that may depend on how well the HD 8000 family is coming along. The next iteration of AMD's GPU architecture, dubbed Sea Islands, is rumored to offer better power efficiency and overall performance despite also being produced on 28nm technology. If so, it may simply make more sense to wait.
Rumors of a dual-core Radeon 7990 have floated around since before the HD 7000 series actually debuted, but this is the first time we've seen such a card in the wild. Say hello to the S10000.
How does this card fit into AMD's product roadmap? That's a bit complicated. Nvidia has explicitly prepped the K20/K20X for servers and workstations and spent years establishing itself as a player in the HPC space. AMD, in contrast, doesn't have much of a presence in those markets.The S10000 is something of a jack-of-all trades and it's designed to tackle multiple markets at once.
On paper, the S10000 is a serious beast. We can't give you comparative specifics quite yet, but the single and double-precision rates listed above are higher than anything debuting today from Intel or Nvidia, as is the card's memory bandwidth. The flip side to these figures, however, is the eye-popping power draw: At 375W, the S10000 needs a pair of eight-pin PSU connectors and draws far more electricity at peak than anything else launching today.
None of this, unfortunately, makes the S10000 a shoo-in for the HPC/supercomputer market. AMD has ceded this space without a fight to Nvidia and Intel, and the company hasn't released any information to suggest its launching a new GPU initiative. Instead, the S10000 is supposedly aimed at the virtualization market. In theory, its dual-GPUs on a single-card are a good way to improve GPU virtualization density inside a single server; a single S10000 draws 375W compared to 550W for a pair of W9000's, while offering nearly identical performance.
The S10000 is rather clearly going after Nvidia's VGX K2, which is based on the GTX 690, but has a TDP of just 225W. Unlike the NV cards, the S10000 is equipped with video outputs, and can drive 4x Display Ports and a single DVI. It's not a bad move from AMD, but we rather wish the company had something like Nvidia's VGX K1. That card drops four mid-range Kepler GPUs on a single PCB and would seem to be a better ambassador for the benefits of GPU virtualization to a company looking to dip its toe in.
Then again, maybe we're wrong, and virtualization customers are banging down AMD's door to get their hands on faster cards. We discussed AMD's virtualization pitch when it debuted the S9000 a few months back, so we'll link you there if you have more questions. As for whether or not AMD will launch a W-version or consumer flavor of the card, that may depend on how well the HD 8000 family is coming along. The next iteration of AMD's GPU architecture, dubbed Sea Islands, is rumored to offer better power efficiency and overall performance despite also being produced on 28nm technology. If so, it may simply make more sense to wait.
