facebook rss twitter

Review: Nvidia GeForce GTX 770

by Tarinder Sandhu on 30 May 2013, 14:00


Quick Link: HEXUS.net/qabwrr

Add to My Vault: x

A lot of old, something new

Nvidia has tempted the moneyed enthusiast by launching two ultra-high-end cards this year. We first saw the Titan in February, followed by the GTX 780 just last week. These GPUs share a 7.1-billion-transistor die and are derived from the architecture powering the latest workstation Tesla cards. They're big, beefy and at over £500 too rich for most folk.

The firm's previous single-GPU graphics champ, GTX 680, has been around for well over a year now, which is practically ancient in the fast-paced GPU space. Thinking about it from a marketing and PR perspective, Nvidia needs 7-series cards in addition to the GTX 780, but how to do this when the next-generation architecture, codenamed Maxwell, is slated for 2014? Well, GTX 780 copies Titan's blueprint, but it's an expensive-to-produce GPU that's hard to scale down to lower price points.

Nvidia is solving this conundrum by, today, introducing GeForce GTX 770 - a GPU that perfectly copies the architecture found on the extant GTX 680. But surely the long-awaited GTX 770 isn't just a rebadged GTX 680, is it? That would be wrong on many levels, so let's roll out the Table of Silicon Purgatory and see what this new premium gaming GPU is all about.

GPU GeForce GTX TITAN (6,144MB) GeForce GTX 780 (3,072MB) GeForce GTX 770 (2,048MB) GeForce GTX 680 (2,048MB) Radeon HD 7970 GHz (3,072MB)
Launch date February 2013 May 2013 May 2013 March 2012 June 2012
DX API 11.1 11.1 11.1 11.1 11.1
Process 28nm 28nm 28nm 28nm 28nm
Transistors 7.1bn 7.1bn 3.54bn 3.54bn 4.3bn
Approx Die Size 551mm² 551mm² 294mm² 294mm² 352mm²
Processors 2,688 2,304 1,536 1,536 2,048
GPU Boost v2.0 v2.0 v2.0 v1.0 AMD-specific
Texture Units 224 192 128 128 128
ROP Units 48 48 32 32 32
GPU Clock/Boost (MHz) 837 (876)

863 (902)

1,046 (1,085) 1,006 (1,058) 1,000 (1,050)
Shader Clock/Boost (MHz) 836 (876)

863 (902)

1,046 (1,085) 1,006 (1,058) 1,000 (1,050)
GFLOPS 4,494 3,977 3,213 3,090 4,096
Memory Clock (MHz) 6,008 6,008 7,010 6,008 6,000
Memory Bus (bits) 384 384 256 256 384
Max bandwidth (GB/s) 288.4 288.4 224.3 192.2 288
Power Connectors 8+6 8+6 8+6 6+6 8+6
TDP (watts) 250 250 230 195 250
GFLOPS per watt 17.98 15.85 13.97 15.84 16.38
Current price $999 $649 $399 $449 $449

Analysis - GPU Boost v2.0 and higher frequencies

Pay close attention to the GTX 770 and GTX 680 columns. Both GPUs use the same Kepler architecture, encompassing 1,536 shaders, 32 ROPs and a 256-bit memory bus. The basic blueprint, then, is the same, but, as always, the devil is in the details.

The year or so since the GTX 680's launch has provided Nvidia and its manufacturing partner, TSMC, enough time to fine-tune the engineering process enough to tease out higher frequencies from the GK104 silicon; achieving loftier clocks during maturation is common for cutting-edge technology. Nvidia now feels comfortable in releasing the reference GTX 770 GPU with a core speed of 1,046MHz, rising to an average of 1,085MHz under load, and this represents a small gain over GTX 680.

One of the reasons why we believe the GTX 770 will run faster than Nvidia's own conservative 1,046/1,085MHz clock is down to implementing GPU Boost v2.0 for the new card. The more-granular approach to frequency and voltage is sure to pay dividends when the card is pushed.

As a result, we expect add-in board partners to run their respective GTX 770 OC editions much higher. But this process maturation is also likely to be used by the very latest GTX 680 GPUs on the market, too, meaning GTX 770 is simply taking the bulk of GK104 GPUs previously destined for the now-doomed GTX 680, and repurposing them as 'all-new' GTX 770s.

An insubstantial jump on the core isn't nearly enough to allow the chip company to get away with calling this new GPU the GTX 770, we feel, so Nvidia has upgraded the GDDR5 memory, too. We're impressed that the GTX 770 ships with 7Gbps modules as standard, up from the already-fast 6Gbps on the GTX 680/780/Titan, and this leads to a 17 per cent hike in available bandwidth.

Faster memory certainly burns through more power, so the increase in TDP over GTX 680 - 230W vs. 195W - doesn't come as a huge surprise. Like the GTX 680 variant, the bulk of the cards are to be outfitted with a 2GB framebuffer, though partners can choose to release 4GB versions, albeit with a tasty price hike.

Keen pricing

Speaking of price, reference-like GTX 770s are to be listed at $399 (£329) from today, putting them firmly in the crosshairs of the Radeon HD 7970 GHz and standard card. This undercuts the GTX 680 by around £40 yet, clearly, the new card offers more performance. It's therefore safe to assume GTX 680s will disappear very quickly once production of GTX 770s is ramped up.

And there you have it. The GTX 770 can quite legitimately be thought of as a slightly overclocked GTX 680 GPU - most probably using the exact core - that is fed by the fastest GDDR5 we've ever seen. Think of it as an ultra-clocked GTX 680 and you're pretty much on the money.

Knowing that this new card is little more than a GTX 680 OC with GPU Boost 2.0 baked in, is it then possible to flash existing GTX 680s BIOSes to the new, higher speed of the GTX 770 - and upgrade for free, if you will? It's doubtless possible, especially if the GTX 680's GDDR5 memory is capable, but we reckon it's semi-pointless; you risk damaging the card and any frequency hike can be easily software-loaded by using a third-party tool such as EVGA Precision. It's not as if loading a new BIOS paves the way for activating dormant cores or a wider memory interface, is it?

GTX 770s main purpose is two-fold: it's designed to deflect attention away from AMD's Radeon HD 7970/GHz cards by matching them on price and performance, and to reinvigorate Nvidia's premium GPU category just in time for Intel's Haswell CPU launch.