Shootout: 780 Lightning vs 290

[Termie]

This is a great comparison, and some huge overclocks.

Just to emphasize, though - you got the OCs on a pitiful 290 (the reference) and the very best 780 (the Lightning). Most 780s will not OC like that.

 

[BallaTheFeared]

There are a couple things to note:

First the 290 is on water, not reference air.

Secondly the core goes high but the vram does not on his old Lightning. F2F has this card now, it is one of the first Lightnings to roll out of MSI and due to the shortage it has the Elpida memory.



I didn't have my card long enough but when I compared his Bioshock score to mine one thing become quite clear to me...

I got a huge uplift off my initial vram OC (My card came out of the box at 1163MHz with stock vram).

Here is our back and forth when I first obtained the card...

F2F: @ 1320/1650 Avg - 116.41

Balla: @ Stock 1163 core boost Avg - 106.04

F2F: I was getting Avg - 111 fps with 1150

Balla: 114.44 avg @ stock core /w +550 memory offset

Balla: 117.80 Avg @ 1215/1777 (Did I win yet?)



Why am I posting this?

Well I found it strange that my card at 1215/1777 was faster than his Lightning at 1320/1650... Now weather that is because of Haswell vs Ivy @ 1080p with these cards or something else I can't say as my time was cut short.

However I do plan to explore it a bit later both with F2F and with my own card. If the memory offset makes that much of a difference at 1080p, I imagine it would make more of a difference at 1600p but I have neither a card to test or a 1600p monitor, but face just got a 1440 so I will start pestering him instead of you Elfear 😉

 

[Jacky60]

That 780 pretty much craps all over the 290, excellent OC and I suspect its not unusual. Nvidia are delivering the performance as usual when under a bit of competitive pressure. Good review Elfear thx.

 

[Imouto]

This is a great comparison, and some huge overclocks.

Just to emphasize, though - you got the OCs on a pitiful 290 (the reference) and the very best 780 (the Lightning). Most 780s will not OC like that.

There are a couple things to note:

First the 290 is on water, not reference air.

20 phase vs 6 phase.

Seems legit.

 

[BallaTheFeared]

20 phase vs 6 phase.

Seems legit.

Marketing matters? Or was his card on LN2?

Classified is 12+2, HoF is 8+2..

Oh and I took my reference 470s to 960 core, those had 5 phases in total with a 215w TDP at 607 MHz.


You can stack as many phases on a board you want, a core will only go so high.

 

[Zanovar]

Good stuff elfear and very interesting reading,*thumnb up*,hmm a great clocking (780/class/light or 290 aftermaerket(luckish of the draw) the longer this goes on the less interested i become*shrugs*.i will probably end up buying one of these cards just about the same time they are announcing their 20nm cards😛.:S

 

[Imouto]

Marketing matters? Or was his card on LN2?

Classified is 12+2, HoF is 8+2..

Oh and I took my reference 470s to 960 core, those had 5 phases in total with a 215w TDP at 607 MHz.


You can stack as many phases on a board you want, a core will only go so high.

That's just one of the likely huge differences between both cards Reference 290/X are known for being weaksauce for OC. VRMs for once, top binned chips secondly and the list is likely to go on forever. I'd just like to point you the graph where the 780 Lightning goes all the way to 640 watts with the 290 nowhere close to that. If you're trying to compare max OCed numbers this isn't a fair metric by miles.

 

[Smartazz]

A part of me wants to sell my 290x for a 780 Lightning. I'm wondering if I could get an even trade thanks to the miners.

 

[blackened23]

You could probably sell for a high price to a miner on ebay to even perhaps cover a 780ti. Or you could wait a month for more 290X aftermarket cards, assuming the prices stabilize. If you do opt to sell the reference 290X, i'd do it soon. Miners are still buying, not sure if that pace will maintain.

 

[BallaTheFeared]

That's just one of the likely huge differences between both cards Reference 290/X are known for being weaksauce for OC. VRMs for once, top binned chips secondly and the list is likely to go on forever. I'd just like to point you the graph where the 780 Lightning goes all the way to 640 watts with the 290 nowhere close to that. If you're trying to compare max OCed numbers this isn't a fair metric by miles.

He's still pumping over 1.39v and 1.34v into the cards, lol... On water, yeah.

Must be the phase power, not allowing it to use 1.5v for 24/7 OC, makes sense in the context of a thread like this.

290/x are clocking where a similar (slightly tweaked) uarch clocked on the same node... :hmm:

You're comparing a faster card, which is putting more load on the total system and on air to the draw of water cooled slower cards...

It's just yet another 780 OC > R290/x OC review, not sure how much further I'd read into it than that. OC is lottery, the only guarantee is stock.

 

[Smartazz]

You could probably sell for a high price to a miner on ebay to even perhaps cover a 780ti. Or you could wait a month for more 290X aftermarket cards, assuming the prices stabilize. If you do opt to sell the reference 290X, i'd do it soon. Miners are still buying, not sure if that pace will maintain.

I don't think there's enough demand on the 290x to upgrade to the 780 Titanium. If I could, then that's def the best option. I got a water block for the 290x, but I'm tempted to sell the block and the card and pocket the difference to a 780 Lightning.

 

[Imouto]

You're comparing a faster card, which is putting more load on the total system and on air to the draw of water cooled slower cards...

No, it's not. In Crysis 3 charts the max OCed 290 isn't even a fps lower than the max OCed 780 Lightning and it's pulling 155 (+31%) more watts. If we go by his own testing methodology the 290 wasn't watercooled in that test.

It's just yet another 780 OC > R290/x OC review, not sure how much further I'd read into it than that. OC is lottery, the only guarantee is stock.

No, this is just how data is manipulated to fit agendas.

 

[blackened23]

On that note, the GK110 with 1.3 volts clocks insanely high. The reference stock BIOS does not allow that high of a voltage, but essentially, 1300mhz+ becomes trivial with 1.3V on the GK110. The point here being, voltage matters more than power phases. If a scant reference GK110 can hit 1300mhz in a trivial fashion with the 1.3V afterburner unlock, i'd expect similar results with Hawaii.

I just don't see how near 1.4V being pumped into Hawaii somehow limits it. Power phases merely provide clean power to the chip and that matters more for LN2 overclocking. And by the way, that's why the lightning has so many power phases. That card is designed from the ground up for LN2 extreme overclocking and the lightning line of SKUs always has been. As far as air or water overclocking, higher voltage generally allows a chip to meet the upper range of its potential. With the GK110, even reference based Titans are hitting 1300mhz with ease using the voltage unlock through afterburner as i've mentioned..

 

[blackened23]

No, this is just how data is manipulated to fit agendas.

So we're back to the "just you wait" mentality to see Hawaii's overclocking? "Just you wait" for the 290X lightning? Uhh. Okay. I don't see how data is being manipulated here. The burden of proof is on you if you want to tell us how great Hawaii can overclock with more power phases and voltage. The GK110 is proven in terms of overclock scaling with additional voltage. Again....the burden of proof is yours, not ours, if you want to show us a 1300mhz Hawaii chip with more power phases/voltages/what have you.

So far that has proven to not be the case. But if you have more data proving otherwise, please share. Oh and BTW. The 7970 lightning did not overclock all that much better than other aftermarket 7970s. But, I guess we'll just have to wait again.

All that said, the 290X performs really well at 1200MHz. VERY well. But it doesn't have the overclocking headroom of GK110 IMO. If the lightning changes that i'll revise my opinion, but this is based on countless web reviews hitting overclocks at 300-400MHz above stock fairly easily. (reference GK110 780 boost stock clock = 900mhz.) With custom boards like the classified, 500-600MHz above stock has been achieved through voltage scaling. Whereas 200-250MHz seems to be the extreme upper limit so far with Hawaii, even with near 1.4V being pumped through the chip.

 

[BallaTheFeared]

No, it's not. In Crysis 3 charts the max OCed 290 isn't even a fps lower than the max OCed 780 Lightning and it's pulling 155 (+31%) more watts. If we go by his own testing methodology the 290 wasn't watercooled in that test.

Yeah he's putting 1.4v through the reference cooler :awe:

No, this is just how data is manipulated to fit agendas.

^

Let's spin it like the other side.

R290 - $530
Water Block - $120
Pump/Res - $140
Fittings/Tubing - $100


R290 was benched with a $900 setup cost for the GPU alone.

780 Lightning is $500

http://www.newegg.com/Product/Produc...82E16814127754

😱

 

[Zanovar]

On that note, the GK110 with 1.3 volts clocks insanely high. The reference stock BIOS does not allow that high of a voltage, but essentially, 1300mhz+ becomes trivial with 1.3V on the GK110. The point here being, voltage matters more than power phases. If a scant reference GK110 can hit 1300mhz in a trivial fashion with the 1.3V afterburner unlock, i'd expect similar results with Hawaii.

I just don't see how near 1.4V being pumped into Hawaii somehow limits it. Power phases merely provide clean power to the chip and that matters more for LN2 overclocking. And by the way, that's why the lightning has so many power phases. That card is designed from the ground up for LN2 extreme overclocking and the lightning line of SKUs always has been. As far as air or water overclocking, higher voltage generally allows a chip to meet the upper range of its potential. With the GK110, even reference based Titans are hitting 1300mhz with ease using the voltage unlock through afterburner as i've mentioned..

NM

 

[blackened23]

NM as well!

 

[Imouto]

Some people should really revisit a glossary of electrical terms. Voltage is meaningless without the current. As shown here the Lightning is being fed way more and better.

 

[BallaTheFeared]

Some people should really revisit a glossary of electrical terms. Voltage is meaningless without the current. As shown here the Lightning is being fed way more and better.

Someone should stop cherry picking graphs.

If the 290 was capable of clocking higher it would be capable of drawing more power.

 

[Silverforce11]

No denying it, the Lightning is an awesome card, good to see MSI keeping up their rep for this brand. At this OC level its certainly worth the premium.

However, these big dies certainly suck a lot of juice when cranked up, and its a bit scary to see ONE GPU pull that much watts. Almost defies the laws of physics!

 

[Imouto]

Someone should stop cherry picking graphs.

If the 290 was capable of clocking higher it would be capable of drawing more power.

And clocks are likely to be hold back because of power delivery, both not constant voltage and not enough current likely due to powertune.

The 290 seems to be hitting a wall at ~500W power consumption. Just check the Heaven benchmark and you'll see both the 1250Mhz and the 1150Mhz runs with the same power consumption. Back to the Crysis 3 bench both 1250Mhz and 1150Mhz use the same power. And for TFSM's sake check the Bioshock chart, that 1250Mhz run is screaming for more power.

 

[BallaTheFeared]

Sorry your explanation makes no sense at all.

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

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


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

 

[Silverforce11]

Those VRMs are magical, 6 of them can have NO limits on power delivery?

 

[BallaTheFeared]

No, their limit would be when they popped.

You get cleaner delivery with more phases, and you can push more power before catastrophic failure. Only one of those can slightly help core OC.

Just look at cpu motherboards, 8+2 or 32 doesn't matter the chip can do what the chip can do. Or 4 if you go cheapo with AMD, what happens then? Chip isn't limited, vrms go ka-boom!

I'm not surprised I'm having this conversation though...

First the results were with a reference card doing 1285/1400 with 1.39v, then someone tried to pass off limiting phases of power causing a chip to take more voltage, clock higher, and perform better while using the same amount of power because in some 5th dimension that made sense when they typed it out.

 

[Silverforce11]

Absolutely not true. Lack of mosfets don't simply go boom, they have built in protection mechanisms to prevent overload, by cutting back on power delivered, causing instability or throttling.

My mITX Z68 MB was one of the first mITX board available with Z68 for OC, but it only had several phases and every single user (check out [H] forum on SFF rigs) on this board experience throttling of CPU OC above 4.3ghz for i5 and above 4.1ghz for i7s. It didnt even come with heatsinks so as soon as it gets too hot, it loses efficiency and throttles more. I had to install zalman copper mosfet sinks on them to stop throttling above 4.3ghz.

As soon as the next iteration came out with more mosfets and heatsinks on them, mITX owners could push the OC as high as ATX.

Edit: There is a direct relationship for mosfet operations, higher temps lower their efficiency causing more input required per output. Each mosfet has a maximum current, hotter mosfets in combination with not enough, would therefore cause throttling, or for gimped no-safety boards, pop!