AMD and Nvidia have both refreshed their FirePro and Quadro lineups with announcements at at SIGGRAPH 2014.
AMD announced the FirePro W7100, W5100, W4100, and W2100, all based on a range of AMD GPUs ? including Tonga. The lineup is 4K capable, they offer improved performance across the board and carry double the amount of VRAM compared to the past generation.
| AMD FirePro W Series lineup |
| |
AMD FirePro W7100 |
AMD FirePro W5100 |
AMD FirePro W4100 |
AMD FirePro W2100 |
| Stream Processors |
1792 |
768 |
512 |
320 |
| ROPs |
Unknown |
16 |
16 |
8 |
| Memory Clock |
5GHz GDDR5 |
6GHz GDDR5 |
5.5GHz GDDR5 |
Unknown |
| Memory Bus Width |
256-bit |
128-bit |
128-bit |
64-bit |
| VRAM |
8GB |
4GB |
4GB |
2GB |
| Double Precision |
Unknown |
1/16 |
1/16 |
1/16 |
| TDP |
Unknown |
Unknown |
50W |
26W |
| GPU |
Tonga |
Bonaire |
Cape Verde |
Oland |
| Architecture |
GCN 1.1 |
GCN 1.1 |
GCN 1.0 |
GCN 1.0 |
| Display Outputs |
4 |
4 |
4 |
2 |
The W7100 gets a brand new GPU named Tonga. Compared to the previous Tahiti part, the new GPU is a smaller, lower performance GCN 1.1 part that slots in under the Hawaii GPU used in the W9100 and W8100. AMD is not disclosing the GPU clockspeed or offering any floating point throughput performance numbers, although we know that the W7100 will feature 1792 stream processors. Coupled with that is 8GB of GDDR5 clocked at 5GHz sitting on a 256-bit memory bus.
With the W7100, AMD is doubling W7000?s memory from 4GB to 8GB. The Tonga GPU process 4 triangles/clock versus W7000?s 2 tris/clock. Overall compute/rendering performance should also greatly be increased due to the much larger number of stream processors (1792 vs. 1280).
Like the W7000, the W7100 is equipped with 4 full size DisplayPorts, allowing for a relatively large number of monitors to be used with the card. A new GCN 1.1 display controller can drive 3 4K@60Hz displays .
The W7100 is designed be a budget alternative to the W9100 and W8100, offering reduced performance but at a much lower cost.
The second card being introduced today is the W5100, which is based on AMD?s Bonaire GPU. W5100 replaces the outgoing W5000 and utilizes a slightly cut-down version of Bonaire with 768 SPs active. It is clocked at approximately 910MHz, which puts its compute performance at 1.4 TFLOPS for single precision. Feeding W5100 is 4GB of VRAM attached to a 128-bit memory bus and clocked at 6GHz.
With 4 full size DisplayPorts the W5100 has the same 4K display driving capabilities as the W7100. However with lower performance and half the VRAM, it?s targeted towards lower power usage scenarios.
The third of today?s new FirePro cards is the W4100, which is based on AMD?s older Cape Verde GPU. It has just 512 SPs active and its power consumption is 50W. AMD has paired the card with 2GB of VRAM.
The final FirePro card being introduced today is the FirePro W2100, which is AMD?s new entry-level FirePro card. It is based on AMD?s Oland GPU and has 320 SPs active attached to 2GB of memory on a 64-bit bus.
At a TDP of just 26W, the W2100 is AMD?s lowest power, lowest performance card.
Wrapping things up, AMD has not revealed final pricing for these cards at this time, and their availability is expected in Q4 of this year.
AMD also announced that they will be bringing their low-level Mantle API over to the FirePro family.
Seperately, Nvidia announced their own refresh of their Quadro lineup.
The Quadro K5200 is NVIDIA?s new second-tier Quadro card. Based on a cut down version of NVIDIA?s GK110 GPU, the K5200 is anupgrade to the K5000 thanks to the high performance and features found in GK110.
| NVIDIA Quadro lineup |
| |
Quadro K5200 |
Quadro K4200 |
Quadro K2200 |
Quadro K620 |
Quadro K420 |
| CUDA Cores |
2304 |
1344 |
640 |
384 |
192 |
| Core Clock |
650MHz |
780MHz |
1GHz |
1GHz |
780MHz |
| Memory Clock |
6GHz GDDR5 |
5.4GHz GDDR5 |
5GHz GDDR5 |
1.8GHz DDR3 |
1.8GHz DDR3 |
| Memory Bus Width |
256-bit |
256-bit |
128-bit |
128-bit |
128-bit |
| VRAM |
8GB |
4GB |
4GB |
2GB |
1GB |
| Double Precision |
Unknown |
1/24 |
1/32 |
1/32 |
1/24 |
| TDP |
150W |
105W |
68W |
45W |
41W |
| GPU |
GK110 |
GK104 |
GM107 |
GM107Unknown |
GK107Unknown |
| Architecture |
Kepler |
Kepler |
Maxwell |
Maxwell |
Kepler |
| Displays Supported |
4 |
4 |
4 |
4 |
4 |
The K5200 ships with 12 SMXes enabled and utilizes a 256-bit memory bus. NVIDIA has put the GPU clockspeed at 650MHz while the memory clock stands at 6GHz. Meanwhile the card has a total of 8GB VRAM.
NVIDIA has clamped K5200 at 150W, which is much lower than the full-fledged K6000 but is 28W more than the K5000.
Display connectivity includes a DL-DVI-I port, a DL-DVI-D port, and a pair of full size DisplayPorts, with the card able to drive up to 4 displays.
NVIDIA?s second new Quadro card is the K4200, which is based on the GK104 GPU comprised of 7 active SMXes (1344 CUDA cores) paired with 4GB of VRAM. Clockspeeds stand at 780MHz for the GPU and 5.4GHz for the VRAM.
Moving on, NVIDIA?s third Quadro refresh card is the K2200, which replaces the GK107 based K2000 and is the first Quadro product to utilize one of NVIDIA?s newest generation Maxwell GPUs, tapping NVIDIA?s GM107 GPU.
NVIDIA is aiming to produce an especially potent sub-75W card for K2200. Here NVIDIA uses a fully enabled GM107 GPU ? all 5 SMMs (620 CUDA cores) are enabled ? and it?s paired with 4GB of VRAM on a 128-bit bus. Meanwhile based on NVIDIA?s performance figures the GPU clockspeed should be just north of 1GHz while the memory clock stands at 5GHz.
The K2200 has 1.3 TFLOPS of single precision compute/shader performance and 4GB of VRAM. Display connectivity is identical to the new K4200 and the outgoing K2000. The K2200 can drive up to 4 displays by utilizing a mix of its DL-DVI port, two DisplayPorts, and a DisplayPort MST hub.
NVIDIA?s fourth Quadro card is the K620, again based on a Maxwell GPU
which features 3 SMMs (384 CUDA cores) and 2GB of DDR3 operating on a 128-bit memory bus. Like K2200 the GPU clockspeed appears to be a bit over 1GHz, and meanwhile the memory clockspeed stands at 1.8GHz.
NVIDIA?s last new Quadro card is the K420, based on a Kepler GPU . From a performance perspective this card won?t see much of a change ? the number of CUDA cores is constant at 192 ? but memory bandwidth has been doubled alongside the total VRAM pool, which is now 1GB.
NVIDIA says that the Quadro refresh cards will be shipping in September.
Nvidia also announces the entry-level GeForce GT 720. Based on the 28 nm GK208 silicon, the chip features 192 CUDA cores, and a 64-bit wide DDR3 memory interface, holding 1 GB of memory. It features core clock speeds of 797 MHz, and memory clock speeds of 1600 MHz DDR.
NVIDIA is capturing the sub-$50 market with this passively-cooled GPU.