GeForce what?
So what is GeForce 9800 GTX+?Actually, NVIDIA pre-announced it during the launch of Radeon HD 4800-series cards, no doubt to steal some of their thunder.
Available from next week, in quantity, according to NVIDIA, GeForce 9800 GTX is nothing more than a die-shrink that's accompanied by higher frequencies for the core and shaders, as compared to present GeForce 9800 GTX -a GPU that's based on the larger 65nm process.
The half-node process, 55nm, does a couple of things for NVIDIA. For one, it matches what AMD has been doing since last November - the month that Radeon HD 3870 was officially released. Secondly, it makes the GPU cheaper to produce, because NVIDIA's manufacturing partner, TSMC, can fit more on to a regular wafer.
There has been no further architectural tweaks to speak of, and it's sensible to presume that it will benchmark akin to the voluminous number of pre-overclocked GeForce 9800 GTXs currently out in the marketplace.
Let's take a look at the vital specs.
| Graphics cards | NVIDIA GeForce 9800 GTX+ 512 | NVIDIA GeForce 9800 GTX 512 | NVIDIA GeForce 8800 GTS 512 | NVIDIA GeForce GTX 280 | NVIDIA GeForce GTX 260 | ATI Radeon HD 4870 512MiB | ATI Radeon HD 4850 512MiB | ATI Radeon HD 3870 512 |
|---|---|---|---|---|---|---|---|---|
| PCIe | PCIe 2.0 | |||||||
| GPU clock | 738MHz | 675MHz | 650MHz | 602MHz | 576MHz | 750MHz | 625MHz | 775MHz |
| Shader clock | 1,836MHz | 1,688MHz | 1,625MHz | 1,296MHz | 1,242MHz | 750MHz | 625MHz | 775MHz |
| Memory clock (effective) | 2,200MHz | 2,200MHz | 1,940MHz | 2,214MHz | 1,998MHz | 3,600MHz | 2,000MHz | 2,250MHz |
| Memory interface, and size, | 256-bit, 512MiB, GDDR3 | 256-bit, 512MiB, GDDR3 | 256-bit, 512MiB, GDDR3 | 512-bit, 1,024MiB, GDDR3 | 448-bit, 896MiB, GDDR3 | 256-bit, 512MiB, GDDR5 | 256-bit, 512MiB, GDDR3 | 256-bit, 512MiB, GDDR4 |
| Memory bandwidth | 70.4GiB/sec | 70.4GiB/sec | 62.1GiB/sec | 141.7GiB/sec | 111.9GiB/sec | 115GiB/sec | 64GiB/sec | 72.8GiB/sec |
| Manufacturing process | TSMC, 55nm | TSMC, 65nm | TSMC, 65nm | TSMC, 65nm | TSMC, 65nm | TSMC, 55nm | TSMC, 55nm | TSMC, 55nm |
| Transistor count | 754M | 754M | 754M | 1,408M | 1,408M | 965M | 965M | 666M |
| Die size | 230mm²? | 330mm² | 330mm² | 576mm² | 576mm² | 260mm² | 260mm² | 192mm² |
| Double-precision support | No | No | No | Yes | Yes | Yes | Yes | Yes |
| DirectX Shader Model | DX10, 4.0 | DX10.1, 4.1 | ||||||
| 320 FP32 scalar ALUs, MADD dual-issue (unified) | 128 FP32 scalar ALUs, MADD dual-issue + MUL (unified) | 128 FP32 scalar ALUs, MADD dual-issue + MUL (unified) | 128 FP32 scalar ALUs, MADD dual-issue + MUL (unified) | 240 FP32 scalar ALUs, MADD dual-issue + MUL (unified) | 192 FP32 scalar ALUs, MADD dual-issue + MUL (unified) | 800 FP32 scalar ALUs, MADD dual-issue (unified) | 800 FP32 scalar ALUs, MADD dual-issue (unified) | 320 FP32 scalar ALUs, MADD dual-issue (unified) |
| Peak GFLOPS | 470/705* | 432/648* | 416/624* | 622/933* | 477/715* | 1,200 | 1,000 | 496 |
| Data sampling and filtering | 64ppc address and 64ppc bilinear INT8/32ppc FP16 filtering, max 16xAF | 64ppc address and 64ppc bilinear INT8/32ppc FP16 filtering, max 16xAF | 64ppc address and 64ppc bilinear INT8/32ppc FP16 filtering, max 16xAF | 80ppc address and 80ppc bilinear INT8/40ppc FP16 filtering, max 16xAF | 64ppc address and 64ppc bilinear INT8/32ppc FP16 filtering, max 16xAF | 40ppc address and 40ppc bilinear INT8/20ppc FP16 filtering, max 16xAF | 40ppc address and 40ppc bilinear INT8/ 20ppc FP16 filtering, max 16xAF | 16ppc address and 16ppc bilinear INT8/FP16 filtering, max 16xAF |
| Peak fillrate Gpixels/s | 11.8 | 10.8 | 10.4 | 19.264 | 16.128 | 12 | 10 | 12.4 |
| Peak Gtexel/s (bilinear) | 47.2 | 43.2 | 41.6 | 48.16 | 36.864 | 30 | 25 | 12.4 |
| Peak Gtexel/s (FP16, bilinear) | 23.6 | 21.6 | 20.8 | 24.09 | 18.432 | 15 | 12.5 | 12.4 |
| ROPs | 16 | 16 | 16 | 32 | 28 | 16 | 16 | 16 |
| Peak TDP (watts, claimed) | 155 | 155 | 140 | 236 | 182 | 160 | 110 | 105 |
| Power connectors (default clocked) | 6-pin + 6-pin | 6-pin + 6-pin | 6-pin | 8-pin + 6-pin | 6-pin + 6-pin | 6-pin + 6-pin | 6-pin | 6-pin |
| Multi-GPU | SLI - three-board | SLI - three-board | SLI - two-board | SLI - three-board | SLI - three-board | CrossFire - four-board | CrossFire - four-board | CrossFire - four-board |
| HybridPower | Yes | Yes | No | Yes | Yes | No | No | No |
| Outputs | 2 x dual-link DVI w/HDCP, native HDMI 5.1 (via S/PDIF) | 2 x dual-link DVI w/HDCP, HDMI 7.1 (native, on GPU) | 2 x dual-link DVI w/HDCP, HDMI 5.1 (native, on GPU) | |||||
| Hardware-assisted video-decoding engine | NVIDIA's PureVideo HD - full H.264 decode and partial VC-1 decode | AMD UVD - full H.264 and VC-1 decode | ||||||
| Reference cooler | dual-slot | dual-slot | dual-slot | dual-slot | dual-slot | dual-slot | single-slot | dual-slot |
| Retail price (default-clocked model) | £149** | £129*** | £139 | £319 | £219 | £175 | £117 | £89 |
* calculated on a three-FLOPS-per-clock-cycle basis.
** estimated pricing.
*** hard to find in-stock.
Analysis
The GeForce 9800 GTX+'s differences from the regular GTX have been emboldened.
Higher core and shader clocks mean that it has greater fillrate and GFLOPS throughput, but we've already seen a slew of pre-overclocked GTXs ship with even-higher frequencies.The real bonus for NVIDIA with respect to lowering costs comes from the smaller manufacturing process, leading to a die-size that's around one-third smaller. Make two and get the third free, if you will.
Interestingly, peak power-draw remains the same, at ~155W, necessitating dual 6-pin PCIe connectors, but the rest is pretty much standard GTX fare. We're nonplussed that the GDDR3 memory has been left at 2,200MHz, although we suspect partners to boost this further with pre-overclocked GTX+ variants.
Think of it as a faster 9800 GTX and you're pretty much there.
With retail models expected to be priced from £149, NVIDIA has taken old technology - GeForce 8800 GTS is similar enough, really - and made it cheaper to produce. The pricing is also slap-bang in the middle of Radeon HD 4850 and HD 4870, too.
Let's take a physical look.
