Shootout: 780 Lightning vs 290

[Elfear]

The 290 was definitely lacking in the power delivery department. When I started upping the volts it started acting very erratic. Not "you've hit the limit of the chip" erratic where you get random artifacts, but really weird stuff like the DVI signal going all wonky with crawling text. I also couldn't get both the memory and core stable at max oc. I had to choose one or the other. There's generally a bit of a trade-off but not the big discrepancy I saw with the 290.

With the 780 I was within ~25Mhz of the max capability of the memory even with the core at 1440Mhz. With the 290 I could loop Heaven at 1000/1500 all day but I had to drop the memory to 1450 with the core at 1150 and 1400 at 1250 or higher on the core. In some benchmarks I also got worse performance at max oc than I did at 1150/1450. Personally I think that was due to the PCB being unable to provide clean power to both the memory and core at the voltage I was using.

The intent of the shootout was to show a very high-end 780 against a reference 290. Most 780's will max out much lower which is verified by the average oc on HWbot (1167/2097 on air). Interestingly the average oc for the 290 is 1163/1901 on air (1250/1480 on water).

So while I gave the nod to the Lightning, the 290 was certainly no slouch. In more than one benchmark the 290@1150/1450 was hanging right there with the 780@1340/1650. At the same clocks the 290 is generally faster but due to the massive oc headroom of the Lightning, they came out pretty equal.

 

[Silverforce11]

The intent of the shootout was to show a very high-end 780 against a reference 290. Most 780's will max out much lower which is verified by the average oc on HWbot (1167/2097 on air). Interestingly the average oc for the 290 is 1163/1901 on air (1250/1480 on water).

So while I gave the nod to the Lightning, the 290 was certainly no slouch. In more than one benchmark the 290@1150/1450 was hanging right there with the 780@1340/1650. At the same clocks the 290 is generally faster but due to the massive oc headroom of the Lightning, they came out pretty equal.

There's no doubt the Lightning is a great product, thats why its premium $ is justified by many enthusiasts who love to push their OC. There's very few other cards on the market that can really sustain the kind of wattage we've seen on your Lightning card.

And here: http://hwbot.org/hardware/videocard/radeon_r9_290x/
Average clocks on Liquid Nitrogen? 1319 Mhz. Reference board is crap for big OC, its clear.

 

[BallaTheFeared]

lol must be only for AMD boards then! You've identified the logic on the R290 PCB that creates this function? I mean powertune would simply throttle the card, which the OP wasn't reporting so evidently it wasn't drawing more power than powertune allows/AMD sanctioned.

Anyways that is clearly not what is happening here, since the card took 525w in one game and was unable to draw that much in others. So obviously connecting simple dots power delivery isn't the reason the system consumption went down when moving from 1080p to 1600p in some titles, or that the R290 on water only drew 435w in M2033 when it is capable of drawing 500+ as was already shown in the same review. The reason it didn't draw anymore wasn't because of phase power. Oh and system draw drops going from no DOF to DOF on, guess which component probably was working harder causing higher system consumption?

 

[Imouto]

Sorry your explanation makes no sense at all.

No, it's you being offside all the time.

Card uses the same power at higher clocks and scores higher with more voltage.

Check the Bioshock chart again.

- Same FPS figures.
- More power for the 1250 Mhz run.
- Again capped at 500W.

Check the Crysis Warhead chart.

- 1.5 FPS difference.
- 525W wall for the 1250 Mhz run.

Check the Crysis 3 chart.

- Both runs at ~480W.
- 2 FPS (4%) difference.

Check the Heaven chart again.

- Both runs at 480W.
- Different load with low CPU power demand and high GPU demand.

Yeah ok, now about that bridge you were selling...

My point all along is that the reference 290 is way different than the 780 Lightning, please stop moving goalposts or claiming that I'm the one moving them. Looks like you finally noticed something and you refuse to acknowledge it.

Also it's highest consumption was 525, and it often was sub 500.. You want to talk about uarch bottlenecks and how that works?

The power consumption is for the whole system, of course there will be differences. The 1250 Mhz run is done under water, way more efficient.

 

[Silverforce11]

lol must be only for AMD boards then!

Which is interesting you bring that up, because in the past, NV boards would blow up their mosfets (gtx570, gtx690!) causing fires when pushed whilst no such thing was reported for AMD boards.

If you want to know if its powertune, you have to ask Elfear to do a clockspeed over time chart during his runs. I wouldn't be shocked to see clocks don't stick to his OC level when its pushed above what the board can handle.

 

[Elfear]

There's no doubt the Lightning is a great product, thats why its premium $ is justified by many enthusiasts who love to push their OC. There's very few other cards on the market that can really sustain the kind of wattage we've seen on your Lightning card.

And that's what makes me think the 290 would have done better with a better PCB. The Lightning never skipped a beat at 640W, just rock solid. The 290 was very hard to pin down a max oc and became unstable once higher voltage was pumped through it.

The reference 7970's I've used with the same number of power phases didn't exhibit the same erratic behavior. Makes me think Hawaii just needs better voltage regulation and 6 phases isn't cutting it (for max overclocks anyway, the card acted just fine up to ~1200Mhz).

 

[BallaTheFeared]

Elfear, were your 1275/1400 runs done on reference air or water?

 

[Elfear]

Elfear, were your 1275/1400 runs done on reference air or water?

Water. I can't remember if I ever went for max clocks on air but i don't think so.

 

[Silverforce11]

And that's what makes me think the 290 would have done better with a better PCB. The Lightning never skipped a beat at 640W, just rock solid. The 290 was very hard to pin down a max oc and became unstable once higher voltage was pumped through it.

The reference 7970's I used with the same number of power phases didn't exhibit the same erratic behavior. Makes me think Hawaii just needs better voltage regulation and 6 phases isn't cutting it (for max overclocks anyway, the card acted just fine up to ~1200Mhz).

7970 haven't been shown to the consume the kind of power we're seeing here, these big dies (Hawaii and GK110) really suck the juice once you crank it.

I had made this observation a long time ago, that it looks like the reference board was barely enough on mosfets from AMD, they took the R280X/7970 design and threw in Hawaii on it, not a good move on their part as the HSF is not capable and neither is the power delivery to handle Hawaii properly (not allowing big OCs).

 

[blackened23]

The 7970 lightning did not overclock any better than other aftermarket cards. The additional power phases help for extreme LN2 overclocking, but for air? Not really.
The asus DC II 290X achieved overclocks of 1125-1150 at most review websites. 1200 was an outlier - most websites could not achieve stability past 1150.

I guess we're in the never ending cycle of "just you wait". Just wait for the card being released in Feb. That one will REALLY overclock. I guess. So we're waiting on the lightning 290X. Is that what i'm reading here? Just wait for the lightning 290X. Okay. Gotcha. That one will REALLY overclock. The average overclocks of the DC II were just an anomaly.

So this affects a prospective purchaser..how? The 290 and 290X are priced ridiculously at the moment in the states with the 780 being a far better value. So should the purchaser "just you wait!" for a lightning 290X? All for the hopes of better overclocking? The GK110 is tried and proven to overclock well, so far Hawaii is hitting a wall regardless of the cooling used. But I guess we have to wait for X and Y card being released 2 months from now.

 

[Silverforce11]

Actually here's a good test with a watercooled die but bare VRMs.

https://www.youtube.com/watch?v=7-Xc8F_ioac

A lot of clock speed throttling once its pushed and VRMs got hot (operating less efficiently), its not even able to hold a 1150mhz OC.

Whereas 7970 boards, even the infamous XFX DD with its bare VRM and 105C operating temps didn't throttle as it was still enough to supply the die. This really shows reference board has barely enough and doesn't like being pushed.

@Blackened, I only saw Asus D2 R290X review OC which up the vcore at HC (1195mhz) & Guru3d (1175mhz). Other sites left default volts. Any other reviewers up the voltage?

 

[BallaTheFeared]

Most 780's will max out much lower which is verified by the average oc on HWbot (1167/2097 on air).

1167 is most likely the gpuz base or boost clock, not the actual clock.

2097 is 402 GB/s, you sure that's right? You're at 316 GB/s with your Lightning.

 

[Elfear]

The 7970 lightning did not overclock any better than other aftermarket cards. The additional power phases help for extreme LN2 overclocking, but for air? Not really.
The asus DC II 290X achieved overclocks of 1125-1150 at most review websites. 1200 was an outlier - most websites could not achieve stability past 1150.

From what i've read here, apparently, 1150 is easily doable on the reference cooler. So I guess either review websites are lying, or something is amiss. Apparently the reference 290s sold to users here just overclock that much better. Whatever.

I guess we're in the never ending cycle of "just you wait". Just wait for the card being released in Feb. That one will REALLY overclock. I guess. So we're waiting on the lightning 290X. Is that what i'm reading here? Just wait for the lightning 290X. Okay. Gotcha. That one will REALLY overclock. The average overclocks of the DC II were just an anomaly.

On air the extra power phases probably didn't help as much but we're talking apples and oranges too since Hawaii and especially GK110 are bigger chips. I'd be curious to see whether the average reference 780 overclocks as well as the average Lightning or Classy (both under water to remove the limits of the cooler) with 1.35-1.4V. I'll bet they probably don't but that's just guessing.

We'll have to see if that proves true with Hawaii. Could be the core is reaching it's limits at 1250Mhz and no amount of extra power phases will help. But maybe a more robust PCB will allow consistent oc's in the 1300+ range.

 

[BallaTheFeared]

Reference cards aren't binned like these chips we're talking about, but generally given enough samples you'll see reference cards getting close to the top sku's... The difference is there is a ton more variance with reference unbinned chips vs select binned chips that go into cards like the classy, lightning and even the HoF.

1300+ was exceedingly rare with Tahiti, Hawaii is GCN 1.1 on the same node... I think people are hoping for more than is to be expected in such a scenario.

 

[blackened23]

I wonder if the Asus DC II with a full coverage block would fare better:

http://www.techpowerup.com/196026/n...r-the-asus-radeon-r9-290x-directcu-ii-oc.html

I don't know how many power phases a lightning 290X has, but the asus has 8. I still maintain that it doesn't make much difference outside of exotic cooling (eg. LN2).

 

[Elfear]

1167 is most likely the gpuz base or boost clock, not the actual clock.

I can't imagine it's the boost clocks though. I don't know of any 780's that boost to 1167Mhz out of the box. The other Nvidia cards look about right too for air overclocks (at least on the core): 770 is 1233/2534, 680 is 1220/2023, and 670 is 1136/2111.

2097 is 402 GB/s, you sure that's right? You're at 316 GB/s with your Lightning.

Ya the memory ocs look funky. Water looks right for the 290 but the air average must be wrong.

 

[BallaTheFeared]

I dunno what to make of their Nvidia numbers.

My 780 GHz at stock was...

Core Clock 1017MHz

Boost Clock 1071MHz

Actual Boost 1163MHz

 

[Imouto]

@Elfear

What was the Power Limit for the 1250 Mhz and the 1150 Mhz setups?

Also I can't remember a single review of a 7970 having the Power Limit unlocked like the GTX 780 Lightning does.

 

[blackened23]

I can't imagine it's the boost clocks though. I don't know of any 780's that boost to 1167Mhz out of the box. The other Nvidia cards look about right too for air overclocks (at least on the core): 770 is 1233/2534, 680 is 1220/2023, and 670 is 1136/2111.

Uhm. What. Clarify your meaning here? Every Kepler based GPU boosts higher than advertised out of the box. Kepler's boost is variable with the lower number advertised on the box as a guaranteed boost. The boost with Kepler GPUs is not static as well. If you buy two different GK110 cards, they will have different out of the box boost speeds. Both being higher than advertised. Aside from this, B1 stepping GK110 chips are boosting even higher than A1 chips, out of the box. Most 780ti's are boosting in the mid 1100s despite the advertised boost being much lower - i'd attribute this to the newer stepping being used on all 780ti's, and a few select 780s.

The reference GTX 780 has a guaranteed boost of 900mhz but like I said. Every Kepler GPU boosts higher than that in actual gaming. What's on the box is the lower end of the spectrum printed as a guarantee with the actual boost being far higher, generally speaking. Most GK110 cards from what i've seen regularly boost 90-100MHz higher than what is advertised and printed on box, this has been true since the GK104.

 

[Elfear]

I dunno what to make of their Nvidia numbers.

My 780 GHz at stock was...

Core Clock 1017MHz

Boost Clock 1071MHz

Actual Boost 1163MHz

*Shrugs* I wish we had another database of average overclocks to compare with HWBot. Some of the clocks do make you scratch your head.

I guess the point I was trying to make though is that most 780's outside of the Classified and maybe the HoF aren't going to hit the same clocks as the Lightning. As far as the potential of each company's architecture, the comparison wasn't very fair.

 

[Elfear]

Uhm. What. Every Kepler based GPU boosts higher than advertised out of the box. Kepler's boost is variable with the lower number advertised on the box as a guaranteed boost.

The reference GTX 780 has a guaranteed boost of 900mhz but like I said. Every Kepler GPU boosts higher than that in actual gaming. What's on the box is the lower end of the spectrum printed as a guarantee with the actual boost being far higher.

I get that. The Lightning is advertised at 1033Mhz but mine boosted to 1124Mhz. 1167Mhz seemed high if HWBot was reading the regular boost clock.

 

[ICDP]

Excellent work Elfear, thanks for sharing and taking the time.

The current US prices of AMD R9 cards make them impossible to recommend over Nvidia equivalents. Strange that this price gouging has not hit the rest of the world. R9 290 and 290X are still at MSRP here in the UK.

I have tested 2X MSI Gaming GTX780 in SLI. One gets ~1220 and the other gets ~1202 core overclock with stock BIOS. Both drop to ~1180 or so after a few minutes of gaming. I have also got a 290X and it will get 1170 core clock with +100mV. With a modded 780GTX voltage BIOS max clocks raised only 20-30MHz but had the benefit of zero throttling. At these more realistic over-clocks the GTX780 doesn't come close to matching R9 290X and trails by ~15% on average.

The vast majority of GTX780s are going to get nowhere near 1340 let alone 1400+. The silicone lottery applies and when making a purchase anyone who "expects" 1300+ as a guarantee on their GTX780 should be prepared for disappointment.

 

[Jaydip]

Check at [H], almost all HOFs reach ~ 1300.

 

[ICDP]

Check at [H], almost all HOFs reach ~ 1300.

Three points

1. You said it yourself that "almost" but not all HOFs reach 1300 and they are specifically binned GTX780s with high quality PCB and other OC goodies.
2. The vast majority of GTX780s sold are reference, or the non super duper editions. HOFs, Lightning's and Classy's do not make up the majority of GTX780 sales.
3. Reference or lower end custom cooled GTX 780s (WF3, MSI Gaming etc) are in the vast majority of cases not going to get anywhere near 1300 core clock. The average OC for stock BIOS GTX780s is lower than 1200. In order for me to sustain 1200+ core clock on both My MSI GTX780s I need to flash with a modded voltage BIOS and that is instant warranty void. Stock BOIS OC both throttle to ~1180 after a short period.

Of course the silicone lottery applies to R9 290/X cards as well but using a 1400+ OC GTX780 Lightning to claim the 780 is better than R9 290 = fail. Even Elfear concluded the only reason GTX780 looked good was the inflated price gouging on R9 290 cards in the US. Here in the UK the cheapest GTX780 is £80 more expensive than an R9 290.

Current price gouging in the US puts the highend GTX780s or even GTX780Ti in the same price range as R9 290 290X. There is no contest in the US, the GTX 780 is a superior purchase. The rest of the world it's the R9 290 that gets the nod as the better card in cost/perf.

 

[Jaydip]

Three points

1. You said it yourself that "almost" but not all HOFs reach 1300 and they are specifically binned GTX780s with high quality PCB and other OC goodies.
2. The vast majority of GTX780s sold are reference, or the non super duper editions. HOFs, Lightning's and Classy's do not make up the majority of GTX780 sales.
3. Reference or lower end custom cooled GTX 780s (WF3, MSI Gaming etc) are in the vast majority of cases not going to get anywhere near 1300 core clock. The average OC for stock BIOS GTX780s is lower than 1200. In order for me to sustain 1200+ core clock on both My MSI GTX780s I need to flash with a modded voltage BIOS and that is instant warranty void. Stock BOIS OC both throttle to ~1180 after a short period.

Of course the silicone lottery applies to R9 290/X cards as well but using a 1400+ OC GTX780 Lightning to claim the 780 is better than R9 290 = fail. Even Elfear concluded the only reason GTX780 looked good was the inflated price gouging on R9 290 cards in the US. Here in the UK the cheapest GTX780 is £80 more expensive than an R9 290.

Current price gouging in the US puts the highend GTX780s or even GTX780Ti in the same price range as R9 290 290X. There is no contest in the US, the GTX 780 is a superior purchase. The rest of the world it's the R9 290 that gets the nod as the better card in cost/perf.

I can easily clock my 780 to ~1300 as well.I do not run at those speeds as I play mostly swtor nowadays.Heck I have seen Titan owners with TI bios cross the 1250 mark, it was a max oc shootout and 780 just plain came out ahead.As for recommendation I don't care, people choose according to their needs, their is no one size fits all.