NVIDIA GeForce GTX 770 Graphics Card Review

Add Your Comments

It’s release day again for NVIDIA, and this time we have the second card in the 700 series. We had the GTX 780 a week ago, and now it’s the GTX 770’s turn for the spotlight. Based on the same GK104 core as the GTX 680, the GTX 770 could be just a rebrand, but maybe there’s more. Continue on to see what we can uncover.

Specifications & Features

The following specifications may look familiar to some, especially those that have owned a GTX 680. The GTX 770 is very similar to the GTX 680; it uses the same GK104 core with identical shader/TMU/ROP counts, the same bus width of 256 bits, and the same amount of VRAM. The only difference that can be seen between the GTX 770 and the GTX 680 using GPU-Z is the higher default clocks on both the core and VRAM. However, there is a little more to the GTX 770 than a GTX 680 with a new sticker and increased clocks.

GPU-Z v0.7.1 Specifications

GPU-Z v0.7.1 Specifications

The following are a few of the more interesting new features of this generation of NVIDIA GPUs. If you would like some more detailed info on the new features that comes with the 700 series GPUs, please take a look at our NVIDIA GeForce GTX 780 review published last week.

GPU Boost 2.0

GPU Boost now uses temperature rather than power draw to dictate the boost clock. GPU Boost 2.0 sets the core clock as high as it can while maintaining a set temperature target (default is 80 °C). This temperature target can be increased up to 94 °C by the end user to potentially let the GPU boost higher.

Adaptive Temperature Controller

While trying to stay at the set temperature target, GPU Boost 2.0 takes advantage of a new fan controller that helps reduce drastic fan speed fluctuations. When the fan fluctuations are smoothed out, it makes the GPU less audible to the end user when fan speed changes are taking place.

GeForce Experience

  • Drivers Updated Automatically – GeForce Experience replaces NVIDIA Update, and it notifies you of new drivers and can automatically download them.
  • Optimal Playable Settings – GeForce Experience can change graphical settings to the values it deems most optimal base on the PC’s hardware.
  • ShadowPlay – Taking advantage of the built-in H.264 video encoder on-board Kepler GPUs, ShadowPlay keeps a running recording of the last 20 minutes of gameplay. Since ShadowPlay uses a hardware encoder, it has less overhead than solutions such as FRAPS.

The GeForce GTX 770

These pictures should look rather familiar if you’ve been keeping up with recent GPU releases by NVIDIA, but I assure you that this is indeed a new GPU. The GTX 770 sports the same reference cooler as the GTX TITAN and GTX 780, which means it has the same industrial look to it.

NVIDIA GeForce GTX 770

NVIDIA GeForce GTX 770

NVIDIA GeForce GTX 770

NVIDIA GeForce GTX 770

NVIDIA GeForce GTX 770

NVIDIA GeForce GTX 770

The GTX 770 also features the LED backlit “GEFORCE GTX” logo that was introduced with TITAN and later found on the GTX 780. The next two images are a couple of angled shots of the fan end.

LED Backlit GeForce GTX Logo

LED Backlit GeForce GTX Logo

NVIDIA GeForce GTX 770

NVIDIA GeForce GTX 770

NVIDIA GeForce GTX 770

NVIDIA GeForce GTX 770

As usual, we have two SLI connectors, but I’m unsure if up to three or four cards will be supported in SLI. There wasn’t any mention of the number of GTX 770s supported in SLI in the whitepaper. So, it could be four like the GTX 680 or three like the GTX 780. The GTX 770 uses an 8-pin and a 6-pin power connector, up from the twi 6-pin used on the GTX 680. When talking about power connectors it’s as good a time as any to mention the TDP, which is 230 W. That’s 35 W higher than the GTX 680, so there’s another difference between the two. The display outputs haven’t changed and they include a Dual Link DVI-D, Dual Link DVI-I, HDMI, and DisplayPort.

SLI Connectors

SLI Connectors

PCIe Power Connectors

PCIe Power Connectors

Display Output Connectors

Display Output Connectors

Now it’s time to pop the cooler off and take this thing apart. After removing 22 screws and eating a banana for my cramped forearm, we have the reference cooler off of the GTX 770.

Reference Cooler Removed

Reference Cooler Removed

Taking a closer look at the PCB, we can tell that it’s completely different from the reference GTX 680 PCB. It looks closer to the TITAN or GTX 780 PCB than the GTX 680 PCB, but it’s different from those as well. I have higher 4K x 3K resolutions of the PCB pics if you’re interested in a specific section of the GPU.

PCB Front

PCB Front

PCB Back

PCB Back

Looking at the GPU, we can see it’s the GK104 used in the 600 series. Now that we’re looking at the bare core, I can see the only difference I’ve come across between the GTX 770 and GTX 680 GPU cores and it’s the printed numbers: GK104-400-A2 for the GTX680 and GK104-425-A2 for the GTX 770. Unfortunately, I do not know if there are any underlying changes in the GTX 770’s core.

The Samsung K4G20325FD-FC28 GDDR5 RAM chips are one of the major differences between the GTX 770 and the GTX 680. These chips are rated for 7 GHz at 1.5 V. As far as I know, these are the fastest GDDR5 chips currently being used in consumer graphics cards. However, Samsung does list chips rated for 8 GHz in their product guide. Seeing those 8 GHz rated chips gives me hope of reaching 8 GHz on these chips as well.

Since we’re talking about VRAM, it’s worth noting that the GTX 770 will come in 4 GB flavors as well (like the GTX 680). So, that means this GPU will be allowed to have more VRAM than the GTX 780; kind of odd to me, but it is what it is.

GK104 GPU

GK104 GPU

Samsung

Samsung K4G20325FD-FC28

Now on to the reference cooler. The first shot is just an overall view of the bottom side of the cooler. Nothing too special here, there are thermal pads for the RAM chips and the VRM to transfer heat into the black top plate.

Reference Cooler Underside

Reference Cooler Underside

The thermal pads are the thickest I’ve ever seen. It seems like the cooler would have been designed to allow for thinner thermal pads so heat would be transferred quicker. However, RAM doesn’t produce a ton of heat, especially when only using ~1.5 V or so. Looking at the second picture, we can see one of the screws holding the fan in place was cross-threaded making it stick up a little crookedly. It shouldn’t cause any issues, but it’s definitely not something I like to see.

Thick Thermal Pads

Thick Thermal Pads

Cross-Threaded Screw

Cross-Threaded Screw

Here are all of the cooler components disassembled. We have the top plate with the fan mounted on it next to a mostly pointless fin stack, the actual heatsink itself, and the 3-part outer casing. The fin stack to the left of the fan may draw up a little heat from the VRM, but the heat would have to travel so far that I doubt it even does that.

Reference Cooler Disassembled

Reference Cooler Disassembled

Finally, after removing a total of 39 screws and eating another banana for preventative measures, the last piece of the cooler is uncovered. The fan is a Delta, model BFB0712HF. Unfortunately, this model is not in Delta’s product guides, and the closest I could find was BFB0712H. There is a thin piece of rubber that is stuck to the backside of the hub covering the label. This rubber seems to be for reducing possible vibration noise between the fan and top plate connection.

Reference Cooler Fan

Reference Cooler Fan

Test Setup & Methodology

Test Setup
CPU Intel i7 3770K @ 4.0 GHz
Motherboard ASUS Maximus V Gene
RAM 4×2 GB Corsair Dominator GT @ DDR3-1866 9-9-9-24
Graphics Cards NVIDIA GTX TITAN*
EVGA GTX 780 SC ACX
NVIDIA GTX 770
EVGA GTX 680
EVGA GTX 670 SC
MSI HD 7970 TwinFrozrIII OC Boost Edition*
HIS HD 7950 IceQ X2 Boost Clock*
Hard Drives 50 GB OCZ Vertex 2
2 TB Hitachi GST Deskstar 7K3000
Power Supply SeaSonic SS-1000XP
Operating System Windows 7 Pro x64 SP1
Graphics Drivers NVIDIA 314.09 (GTX TITAN)
NVIDIA 320.18 (GTX 780 and GTX 770)
NVIDIA 314.22 (GTX 680 and GTX 670)
AMD 13.4 (HD 7970)
AMD 12.1 (HD 7950)
Equipment
Tenma Sound Level Meter
Fluke 52 II Dual Input Thermometer
Kill-a-Watt Meter

 * These GPUs were previously tested by other reviewers.

Methodology

  • GPU performance testing is done according to our video card testing procedures.
  • Peak system power consumption was measured using a Kill-A-Watt meter during both Unigine Heaven and 3DMark11′s Combined Test.
  • For cooler performance testing, I increased the temp target in Precision X to the max (94 °C) to let the GPU get at hot as it wanted up to the 95 °C maximum safe temp for operation. I used Unigine Heaven to load the GPU at each increment from 26-100% and let Precision X keep track of the max GPU temp reached during the test.
  • For noise testing, I first turned off all external sources of noise possible (TV, ceiling fans, A/C, etc.) and only fed 4 V to the Gentle Typhoon AP-15 fan on my Thermalright AXP-100. Then, I measured sound level at 10 cm from the intake side of the card and varied the fan speed manually using increments from 26-100%.
  • All testing was performed before disassembling the graphics card.

Performance Results

Synthetic Tests

When looking at our results, I’ll be comparing the GTX 770 to the GTX 680 since they both use the same GK104 GPU and also looking at AMD’s top offering, the HD 7970. The GTX 770 has +40/+1000 MHz  (core/VRAM) advantage over the GTX 680, so I definitely expect the GTX 770 to perform a little better in all tests. In the synthetic tests, the higher clocks of the GTX 770 do allow it to edge out the GTX 680 by about 4% on average, as expected. The GTX 770 is matched or beat out by the HD 7970 is all the synthetic tests, and it turns out the HD 7970 is 4.5% better than the the GTX 770 on average. So, the GTX 770 falls squarely between the GTX 680 and HD 7970.

gtx770_syn_graph

Game Tests

The game tests are more of the same with the GTX 770 beating the GTX 680. Again, when both GPUs are using the same core and memory sub-system, the one clocked higher is expected to be on top. However, the HD 7970 is still outperforming the GK104 GPUs in 5 of the 6 tests with the most pronounced difference being in Battlefield 3 where the HD 7970 is 25% better. In these game tests, the average performance increase of the GTX 770 over the GTX 680 is ~6%, and the average performance increase of the HD 7970 over the GTX 770 is ~9%.  So, again, the GTX 770 falls between the GTX 680 and HD 7970.

gtx770_game_up

Raw Data

Here’s a table of the raw performance data since it wouldn’t fit nicely on the charts above.

gtx770_raw_up

Overclocking

GTX 770 -vs- GTX 680 BIOS

The following slides show the BIOS differences between the GTX 770 and GTX 680. I pulled both BIOSes from my GPUs; the GTX 770 BIOS is on the left and the GTX 680 BIOS on the right.

The clocks, fan speed range, and power target are set in the common tab…nothing too interesting here. The boost tables are essentially the same for the two cards, but the GTX 770 just starts lower. The boost and clock states tabs show the core and VRAM clock settings for each of the four profiles (P00, P01, P05, and P08). The voltage tab shows the max voltage for the cards and the voltage set for each of four profiles (P00, P01, P05, and P08)…now this is the tab I find interesting. The max voltage for the GTX 680 is set at 1.150 V, but an additional 0.0125 V can be adjusted via software. Now, the GTX 770 has a “Max Voltage #2” of 1.200 V, then another 0.0125 V can be used bringing the actual max up to 1.212 V. This voltage difference is a notable change in the two cards for sure, and should allow for some higher core clocks for the GTX 770 than the GTX 680.

This slideshow requires JavaScript.

GPU Boost 2.0

The following graph shows the stock boost clock of this GTX 770 sample while running Heaven. Remember the vague definition of the rated boost clock: The average core clock that the average GPU will use during typical workloads. Even though the rated boost clock is 1085 MHz, the GTX 770 actually boosts to 1123.5 MHz according to the GPU-Z log. I believe every NVIDIA GPU that I’ve seen with boost, has always boosted higher than its rated boost clock.

stock_boost_graph

Stock Air Results

Okay, let’s see what the additional voltage can help us reach. First up is Firestrike and this GTX 770 sample was able to reach 1293 MHz on the core  for a 15% increase over the stock boost clock, 1123.5 MHz. I mentioned earlier that I had hopes to reach 8 GHz on the VRAM, and the 7 GHz rated chips were able to reach and exceed the 8 GHz by quite a bit, coming in at 8316 MHz. That’s a ~19% OC on the already fast 7 GHz chips. These clocks weren’t on the edge of Firestrike stability, seven looped runs completed successfully.

1293/8300 MHz OC

1293/8316 MHz OC

3dm_oc_boost

Next, the more strenuous and consistent load of Unigine Heaven. Both core and VRAM clocks did have to be reduced to run Heaven, but not by that much. The best I could do with Heaven was 1267 MHz on the core and 8208 MHz on the VRAM, still a 13% and 17% OC respectively. Again, not bad at all, and 5 consecutive Heaven runs were successful after finding the highest clocks where it would pass at least one run.

1267/8208 MHz OC

1267/8208 MHz OC

heaven_oc_boost

Cooling & Noise

Cooling Performance

The reference cooler performs pretty well when running at 50% fan speed or higher. Any slower, and the temperatures start to jump up quickly.

Recall that I had mentioned I set the temperature target to its max of 94 °C, then look at the max temp in the chart below… Yeah, the GPU couldn’t keep the card at or below 94 °C when the fan was set manually to its minimum of 26%. As designed, the GTX 770 did try to compensate by throttling the clocks towards the end of the benchmark run, but that just wasn’t enough. The max safe temperature for the card to run at is 95 °C; after that is when clocks will start to throttle.

cooler_graph

Sound Level

The dBA numbers graphed below come from the raw measurement at 10 cm and two estimated dBA levels calculated for different distances (1 m and 2 ft). Why estimate instead of measure at further distances? It’s because the meter I’m using gets more accurate as the dB increase, so I wanted to measure really close to the source to get the most accurate measurements. The following equation is what was used for estimation of sound level at different distances.

L2 = L1 – 20 * log10(r2/r1)

  • L1 = Sound level at reference distance
  • L2 = Sound level at desired distance
  • r1 = Reference distance
  • r2 = Desired distance

Sound level at 1 m is easy to calculate when measured at 10 cm since log10(1/0.1) = log10(10) = 1, so all that needs to be done is subtract 20 dBA from the measured numbers. That’s why I chose the 10 cm measuring distance :)

The reference cooler is especially quiet when operating under 50% fan speed, but as seen in the previous section, the performance will suffer quite a bit. Personally, I wouldn’t like to run the fan faster than 50-60%, that’s just my noise tolerance. I like keeping the noise around 30-35 dBA when the PC is at a reasonable distance away.

noise_graph

System Power Consumption

The power consumption when running the GTX 770 is 2 W less than the GTX 680 in 3DMark11, and 11 W higher in Heaven. So, pretty much the same power consumption from those two cards. Even though the GTX 770 has beefier PCIe power connectors and a higher TDP, I expected the power consumption to be about the same as the GTX 680 since the cards’ specs are basically identical except for higher clocks.

power_graph

Performance per Watt

Personally, I don’t care a lot about performance per Watt, but I know there are those that do. The GTX 770 does pretty good here and traded blows with TITAN, but the GTX 780 is where the performance per Watt is at. The HD 7970 does alright here as well, being mediocre in Heaven and 2nd in 3DMark11.

p-w_graph

Performance per Dollar

The MSRP on the GTX 770 is $399, which is surprising considering the GTX 680 is around $450. That’s right, the GTX 770 performs better than the GTX 680, but costs less. The pricing bumps the GTX 770 up to GTX 670 levels of performance/$. In the synthetic tests, the GTX 770, GTX 670, and HD 7970 are all about equal in performance/$, but in the game tests the HD 7970 beats the other two. The HD 7950 still remains at the top of performance/$. Some people like to also consider game bundles when talking about price, and AMD has better game bundles that include 3 AAA titles (Crysis 3, Bioshock Infinite, and Tomb Raider).

gtx770_p-d_up

Conclusion

Performance-wise the GTX 770 falls between the GTX 680 and AMD’s HD 7970; ~5% better than the GTX 680, but the HD 7970 being ~6% faster.

Cooling performance of the reference cooler is good when operating at 50% or higher, and the the noise starts getting high around the 60% mark. So, I would probably keep the cooler under 60% at all times. Noise tolerance varies from person to person, so my preference may not be the best for everyone.

Overclocking the GTX 770 was painless and results were very good, in my opinion. I definitely did not expect 1267-1293 MHz on the core and a huge 8208-8316 MHz on the VRAM. The additional 0.037 V core voltage over the GTX 680 certainly helped the GTX 770’s core clock run 52-78 MHz higher than I was able to get on the GTX 680 (1215 MHz). The Samsung GDDR5 chips on this card are beastly both at stock and overclocking, they are definitely the fastest VRAM I’ve seen on consumer cards.

The MSRP of $399 is what makes this card, if it was any higher, then it would be overpriced without a doubt. This pricing makes the GTX 770 a better performing and cheaper card than NVIDIA’s own GTX 680. At $399 the performance per dollar of the GTX 770 is just a little lower than the GTX 670.

Based on performance, it might be a tad high in price, but there are other metrics to base worth off of than strictly performance (drivers, software, NVIDIA specific features, etc.). A small price drop to $375 would really boost the performance per dollar of the GTX 770 while making it cheaper than the HD 7970.

Overall, I think NVIDIA has a good GPU in the GTX 770 that fits in a performance gap while not being grossly overpriced. If someone is looking for a high-end GPU without completely breaking the break and wants to make use of some of NVIDIA’s features, then the GTX 770 would be the go to card.

Overclockers_clear_approved

Click the Approved stamp for an explanation of what it means.

– Matt T. Green (MattNo5ss)

Leave a Reply

Your email address will not be published. Required fields are marked *

Discussion
  1. bluezero5
    actually might be able.. both gk104... can they SLI?

    That's what I was thinking and at the very least you may be able to flash one card to match the other card logically.
    Disappointed at the performance at the price point across what is out there in the market. At stock the 770 does hang with the 7970 with synthetic benchmarks, it falls behind with the games, no surprise there.
    We all know the 7970 OC's like a beast and I would love to see the 770 at OC vs 7970 OC'd to say 1200/1600 running through the same set of benchmarks, I think it will help people with their buying decision as both cards sit at the same price points.
    Then the big kicker is the 2GB VRAM the 770 has, they should have put 3 GB on, come on Nvidia
    Leegit
    What is Tom talking about?
    "There’s also the issue of memory capacity to sort out. We didn’t have any trouble running out of steam on our 2 GB cards at 2560x1440. However, as you step up to 5760x1080 and higher, 3 and 4 GB models are more appropriate. Expect to pay somewhere around $50 more per GeForce GTX 770 4 GB."
    Source: http://www.tomshardware.com/reviews/geforce-gtx-770-gk104-review,3519-31.html

    Looks like he expects the cost of the 4GB GTX 770 to be $450 :shrug:
    bluezero5
    We need to be the first site to test this possibility !!! Hokie/Matto5ss !! do something !! :)

    I won't be able to try a flash until this afternoon. I have already saved my GTX 680 and GTX 770 BIOSes, and I've edited a copy of the GTX 770 BIOS to match the GTX 680 as best as I could. I figured this would come up :D
    Boost tables cannot be edited, so that the only major difference in my GTX680 BIOS and my edited GTX770 BIOS.
    You made a typo in the last few paragraphs. Should be "breaking the Bank" not "breaking the break" of is breaking the break a saying? I should also say great review!
    MattNo5ss
    Looks like he was right. Not sure what Leegit's post was pointing out.

    Didn't Nvidia say no 4GB versions? Or was that just for the 780?!?
    Leegit
    Didn't Nvidia say no 4GB versions? Or was that just for the 780?!?

    bluezero5
    Seeing new toys is always exciting.
    Soooo many things to try !
    My bet is that.. it can SLI. ( gotta bet on something exciting! )

    That's correct no GTX 780s with more than 3GB, GTX 770s will have 4GB versions though. This makes me wonder if the GTX 770 will do 4-way SLI like the GTX 680 as well, leaving the GTX 780 to support only 3-way.
    Brolloks
    Disappointed at the performance at the price point across what is out there in the market. At stock the 770 does hang with the 7970 with synthetic benchmarks, it falls behind with the games, no surprise there.
    We all know the 7970 OC's like a beast and I would love to see the 770 at OC vs 7970 OC'd to say 1200/1600 running through the same set of benchmarks, I think it will help people with their buying decision as both cards sit at the same price points.
    Then the big kicker is the 2GB VRAM the 770 has, they should have put 3 GB on, come on Nvidia

    Yeah, it sets between the GTX 680 and HD 7970 for performance, but at the price of the HD 7970. I think EarthDog ran his HD7970 OC settings at 1175/1705MHz. Once I confirm, I could run those vs my GTX 770 OC.
    The GTX 770 has a 256 (4x64) bit bus width. Since you have 4 64bit controllers, that means each controller can handle 2x 32bit memory chips. So, 8 chips total for 8x256MB = 2048MB or 8x512MB = 4096MB. I don't think I've ever seen 384MB chips, which you would need to get 3GB of VRAM.
    By default, the maximum voltage is maxed in the GTX 770 BIOS at 1.2125V. You can see that in the GTX770 BIOS screenie I posted above. The voltage is grayed out and cannot be changed from 1.2125V