Shootout: 780 Lightning vs 290

[Silverforce11]

As far as I see in in a lot of reviews and user results, GK110 scales very well with OC but it needs a good vram OC to achieve that. Whereas Hawaii scales very well with core OC (if reference fan is allowed to go jetmode to prevent throttling!) but doesn't need vram OC as much.

Both are pretty dang good TBH and if a 780 Lightning can be had for similar price to an aftermarket R290, I would certainly recommend the 780. Currently this situation seem to only exist in USA and at etailers.

Elsewhere, there is a massive price separation for 780 Lightning vs R290 (780 Lightning and other high-end models go anywhere from $650 to $750 down here, R290 go for ~$490!). Therefore, I would expect a highly OC Lightning to smash the R290, that should be the "norm".

 

[omeds]

Honestly, I'd grab reference 780's over reference 290's too.

 

[Mondozei]

Honestly, I'd grab reference 780's over reference 290's too.

Reference cards *shivers*

 

[Imouto]

It's 100% true, easily verifiable via video. Would you like that? Would you like to me post a video?

Tomb Raider is a perfect showcase for the GPU since it has basically no cpu overhead.

I don't want a video. I said unrealistic, made up and misleading. Your card is modded to lock the frequencies and it isn't behaving like a retail one. There are several tons of reviews out there with a maximum perf gain of ~25% with way better testing methodology and reputation than you.

So yeah, your scenario is still unrealistic, made up and misleading.

 

[BallaTheFeared]

I don't want a video. I said unrealistic, made up and misleading. Your card is modded to lock the frequencies and it isn't behaving like a retail one. There are several tons of reviews out there with a maximum perf gain of ~25% with way better testing methodology and reputation than you.

So yeah, your scenario is still unrealistic, made up and misleading.

Well I wasn't expecting anything less from you tbh.

My rep is flawless, it doesn't get better only duration of infallibility can improve at this point. ()

Who cares what they're getting, or what retail is, this is an end user thread? Real people with real hardware who are capable of backing up what they say.

I see you are still confused, this is showing 40%+ gains over cards tested at boost clocks of 900 or less. 780s come with all different boost clocks out of the box but always have the same mem freq.

What this is showing is core scaling from 900mhz to 1300mhz on the 780 with bandwidth is perfectly fine. It isn't showing reviewed 780s are capable of 40% gains (unless they were tested @ or below 900mhz and OC the same). :\

 

[el etro]

EDIT: 1306/906=43% Overclock on observed core clock.

40% more performance is a thing that i really doubt. But is perfect(linear)/almost perfect overclock scaling, a thing that GK 110 proven that it can do....



-------------------------------------------------------------------------------------------------------------------



Do all GTX 780 cards with stock bios go as far as your GTX 780 goes? Is this overclock stable for 24/7 or is a benchmark-run-only clock?

Do all GTX 780 go to this clock with the modded BIOS? Only modded BIOS+ modded afterburner + 100% Fan can make every GTX 780 card past the 1250Mhz mark?

My GTX 670 give to me some awesome results like 60 FPS in Farcry3(Maxed out 1080p 8xMSAA with Forceware331.65), but it never give to me the 42FPS average in Crysis 3(Maxed out no AA 1080p) that Guru3d claims.

 

[f1sherman]

Balla compared his 780 OC with 780 on Hardware.fr clocks. Nothing wrong with it IMHO.
Though that does look like one hefty OC.

 

[blackened23]

The "uber" mode used by hardware.fr for the 780 is perhaps the silliest thing I've read. Looking on the surface, you'd think they're overclocking the card for their GK110 so called "uber mode". All they do is up the sliders to maximum temperature limit and maximum power limit. That's it. They don't overclock the card and they don't touch the clockspeeds at all. Whatever....

Their "uber" scores are stock clockspeed. Just to be clear. Since some folks here throw those uber numbers around as if they're overclocked numbers. They are stock numbers.

 

[el etro]

The "uber" mode used by hardware.fr for the 780 is perhaps the silliest thing I've read. Looking on the surface, you'd think they're overclocking the card for their GK110 so called "uber mode". All they do is up the sliders to maximum temperature limit and maximum power limit. That's it. They don't overclock the card and they don't touch the clockspeeds at all. Whatever....

Their "uber" scores are stock clockspeed. Just to be clear. Since some folks here throw those uber numbers around as if they're overclocked numbers. They are stock numbers.

This "mode" is for to give an improvised challenger for AMD cards like 6990, R9 290(s)... it is added only for information purposes, like Ryan adds the data of uber mode of R9 290x to the charts he make.

 

[Imouto]

Hardware.fr clocks are not stabilized or locked clocks, they're averages that some people seems to don't understand. They even mention the max boost their cards are reaching.

Load isn't the same for a whole game session or benchmark run. On a very light load the frequency can reach 1019 Mhz and screw all your testing at locked 900 Mhz. You also can't quantify the variability of the game session or the benchmark run. Your card locked at 900 Mhz may do way better than a retail card at stock in certain portions of the game/benchmark and worse in others.

This isn't like getting an average speed of your trip going to see your parents these holidays.

Looks like you don't even understand how GPU Boost works.

Also, the Uber mode from hardware.fr is the max you can get from a stock card while removing the power and temp factors effectively telling you how much your mileage may vary.

 

[BallaTheFeared]

EDIT: 1306/906=43% Overclock on observed core clock.

1306/901 = 40% observed fps gain

40% more performance is a thing that i really doubt. But is perfect(linear)/almost perfect overclock scaling, a thing that GK 110 proven that it can do....

40% assumes at 900MHz baseline, 40% at 1GHz would be 1,400Mhz. It would entirely depend on what you wanted to use as a point of reference. Also it should be clear by now that cards like my GHz which boost out of the box close to Ti reference performance gain a decent amount of performance just from an initial memory OC even with "stock boost" so it shouldn't be used to take the performance of a 1163MHz 780 with stock vram then add 13% because the 1163 is running away from the bus and you need that bus for scaling to start with.

-------------------------------------------------------------------------------------------------------------------

Do all GTX 780 cards with stock bios go as far as your GTX 780 goes? Is this overclock stable for 24/7 or is a benchmark-run-only clock?

I can't say, I haven't used all 780s.

It's been stable every time I've used it but I've had my card for two days, over Christmas, it's not like I've gone extensive with it.

Do all GTX 780 go to this clock with the modded BIOS? Only modded BIOS+ modded afterburner + 100% Fan can make every GTX 780 card past the 1250Mhz mark?

Yes/No? Also my card was running 68C with 68% fan speed, there is another 1600RPM for the fans to use, and 68C is still pretty far from the point of throttle.

Of course this is on air, with balanced settings. Would I run it 1350/1800 1.3v 24/7? No, probably not for several reasons, yeah it's fast but it's also inefficient. I'm at 1080p and I really don't need it, it would be like driving around town in first gear with a Ferrari. By the same token I wouldn't run a 290 at 1.4v on water 24/7.

My GTX 670 give to me some awesome results like 60 FPS in Farcry3(Maxed out 1080p 8xMSAA with Forceware331.65), but it never give to me the 42FPS average in Crysis 3(Maxed out no AA 1080p) that Guru3d claims.

Maybe you were cpu limited?

 

[BallaTheFeared]

Hardware.fr clocks are not stabilized or locked clocks, they're averages that some people seems to don't understand. They even mention the max boost their cards are reaching.

Yes they're the average clock speed the card used over an entire run, average, meaning below and above possible but all things accounted for the average clock speed was what they posted.

Load isn't the same for a whole game session or benchmark run. On a very light load the frequency can reach 1019 Mhz and screw all your testing at locked 900 Mhz. You also can't quantify the variability of the game session or the benchmark run. Your card locked at 900 Mhz may do way better than a retail card at stock in certain portions of the game/benchmark and worse in others.

I can if it hit 1019 then 840 as often and for as much time, thus the reviewer was confident in saying "avg" of "x" clock. That's how avg works.

This isn't like getting an average speed of your trip going to see your parents these holidays.

Ok.

Looks like you don't even understand how GPU Boost works.

You're probably right, I still haven't figured out what average means. It would seem though hw.fr is using heat to control the boost, via getting the card to throttle down below max boost it might hit coming out of the gate, then they evaluate the performance based on a heated up reference 780 which is probably limited by heat/fan speed in the actual testing to obtain the sub 900MHz avg boost speed.

Also, the Uber mode from hardware.fr is the max you can get from a stock card while removing the power and temp factors effectively telling you how much your mileage may vary.

Oddly below the 1019MHz though, aye?


If it makes you feel better I can use 993MHz fixed to figure out the SCALING of the core with overclocking. When looking at core SCALING it doesn't matter what fixed point I use, so long as that point is fixed. The numbers should line up regardless of if I start at 800 or 1100.

Wouldn't make much of a difference if the objective was to see performance gains with overclocking relative to the hw.fr numbers, since sub 900 is slower than the R290 and 993 is faster.

 

[f1sherman]

My undervolted 290 is at 87C mining now.
Stock clocks, 70% fan and near opened window.

2 weeks ago, same settings, same ambient temp, it was sitting at 83C

How's that for wear & tear??
Not that it surprises me much.

 

[f1sherman]

Funny thing is... fan RPM went UP.

70% used to be 3856 RPM, and now it's 3927-4000 RPM

 

[blastingcap]

AMD had a driver update that speeds up the fan on some models

 

[f1sherman]

Nah, that's not it. THAT driver is 13.11 V9.2; November 8.

So that fix had already been included in my starting 13.11 v9.5 driver

 

[24601]

Nah, that's not it. THAT driver is 13.11 V9.2; November 8.

So that fix had already been included in my starting 13.11 v9.5 driver

Secret remote driver control from AMD?

Kinda like Intel Micro-code updates?

 

[f1sherman]

Im not overly worried about +100-150 RPM.
Though this is the first time I've seen fan going UP with time.

But if you combine that with 4C temp increase... in 2 weeks. That makes me one worried customer.
Though not too worried. I have my local warranty, and I registered with XFX.

BTW note that I never wished to test Mr Bauman and his built to operate on 95C.
No, I said lets step back a little. And keep it at 87-88C

 

[Leadbox]

Im not overly worried about +100-150 RPM.
Though this is the first time I've seen fan going UP with time.

But if you combine that with 4C temp increase... in 2 weeks. That makes me one worried customer.
Though not too worried. I have my local warranty, and I registered with XFX.

BTW note that I never wished to test Mr Bauman and his built to operate on 95C.
No, I said lets step back a little. And keep it at 87-88C

How certain are you that your ambient temp is constant?

 

[Imouto]

snip

You don't understand that the problem is removing the GPU Boost feature. You think of FPS as speed while it's a variable work done over time. That's why your Rome 2 bench looks like a rollercoaster and your max FPS figure in Tomb Raider is 40% higher with your card clocked 40% higher.

As observed by hardware.fr their card peaking 1006 Mhz boost frequencies or 16% higher. If you couple that with lower workloads it destroys your 900 Mhz average assumptions.

 

[f1sherman]

How certain are you that your ambient temp is constant?

&*$@
Hit wrong button - Wall of text gone.

Pretty sure. If anything the ambient temp was HIGHER(edited) back then. I clearly remember wearing less clothes.
Opening my window would have caused temp to shoot down 87->80-82, and now it's sitting at 87C with opened window and with higher fan RPM.

 

[blastingcap]

My reference 7970 brand new had full-load temps 7C lower than it does after several months of operation, and that's with routine maintenance trying to get rid of dust a few times a year. Dust to the fan and the case filters if any, can cause temps to rise. However 4C in 2 weeks is a LOT, too much to explain with just dust alone. Could be other factors like ambient temps changed, or if you misremembered temps, etc.

Edit: yeah, humidity can vary too and impact cooling efficiency

 

[f1sherman]

There's much more humidity in the air today, and that can't be good for temps.
Dust is my enemy #1. I'm using filters and doing all I can to keep it out of my case.

 

[wand3r3r]

Well I wasn't expecting anything less from you tbh.

My rep is flawless, it doesn't get better only duration of infallibility can improve at this point. ()

Lol. Posting histories are quite revealing including yours but let's get this thread out of the biased gutter it's in.

While the 780 core scaling may be decent, cherry picking and modding your card etc don't prove the point you think they do.

 

[BallaTheFeared]

You don't understand that the problem is removing the GPU Boost feature. You think of FPS as speed while it's a variable work done over time. That's why your Rome 2 bench looks like a rollercoaster and your max FPS figure in Tomb Raider is 40% higher with your card clocked 40% higher.

As observed by hardware.fr their card peaking 1006 Mhz boost frequencies or 16% higher. If you couple that with lower workloads it destroys your 900 Mhz average assumptions.

You seem to think Kepler bounces around from base to max boost during benchmark runs, and that is simply not true.

Also you seem to be projecting v-sync based variance in clocks based on a workload not at 99% usage then using that to concoct some inane scenario where a 780 which is already being throttled due to thermals is now reducing/increasing clock speed based on load which should already be at or near 99%.

Kepler will downclock with vsync on if the workload is low enough to warrant it, however it doesn't do it randomly based on nothing when vsync is off and usage is maxing out.

No Rome 2 is a rollercoaster because the load is dynamic, each area in the Rome 2 benchmark has different things going on which are harder or easier to render based on the dynamics of the image being shown. Which should be very obvious since cpu and gpu speeds are fixed.

Tomb Raider gained 40% because the core scaled with the added memory speed perfectly well, and because Tomb Raider the benchmark has very little cpu overhead, unlike Metro 2033 or Rome 2.

No they observed a max boost on their cold run of 1019MHz, when using uber in their hot box run their observed boost clock speed was 993.

Lol. Posting histories are quite revealing including yours but let's get this thread out of the biased gutter it's in.

While the 780 core scaling may be decent, cherry picking and modding your card etc don't prove the point you think they do.

If you have proof that I've lied about my results with video cards please, please share it here so everyone can see what I lair I am.

This isn't about being bias one way or another, you seem to have misunderstood the comment. Likewise I never saw you call my integrity into question when I posted the #1 fastest 7950 CF results in our forum benchmark threads.