R9 290 or GTX780 ??

[BallaTheFeared]

I just was looking at the contrast from the benchmark to the actual game, pretty clear difference between the two.

/wave hand

Nothing to see here.

 

[Elfear]

Rather than all this back and forth about benchmarking at random clocks, why don't we find a realistic average overclock for both cards and bench.

HwBot puts the average overclock for each card on air at:

780 - 1170/2068 (4,150 submissions)
290 - 1157/1865 (298 submissions)

The memory overclocks are obviously wrong, but the core clocks look pretty close to the results below (if you assume unlocked voltage on the 290s).

Based on the results of various reviews sites we get:

780 - 1163/1680 (results compiled from this review thread)
290 - 1093/1422 (results compiled from this review thread)

The 780 results from the review sites above reflect overclocking with extra voltage (+38mV) while the 290 results are on stock voltage. Afterburner can be used to add voltage control on the stock 290 bios using this guide so you can overclock a lot higher without voiding the warranty.

So based on the results above, what should we set the average overclock at for each card to get some real world benches?

 

[BallaTheFeared]

Pretty sure those HWBot numbers are pure bunk.

It's either base clock or something else, also 2068 mem while pure win is simply unrealistic for anything but cherry Samsung 780 Tis. Not to mention my GHz boosts to 1163 out of the box.

Also 1865 mem for 290 is just a joke, who are they kidding?

I think 1300 is perfectly acceptable, some 780s will hit it on stock voltage, some crappy ones like mine will require more voltage than stock but it seems within the realm of reasonable clocks as far as GK110 goes. It's not like I have some boss card, 1300 for a Classified or Lightning for example isn't even worth posting about. In fact if that's all they hit you'd be more likely to see that owner post a complaint than anything else.

 

[blackened23]

1170 average for HWBOT? The guys using LN2? That sounds right to you? 1865 on VRAM on the 290. This sounds right to you? Come on. That can't be based on reality.

As far as real world benches. TPU, Guru3d, and most review websites are really conservative in their overclocks. They're not considered stable unless they're......24/7 game stable. Oddly enough, synthetics tend to do better in overclocked suicide runs without crashing, whereas crysis 3 will kill an overclock.
So why am I saying this? An overclock isn't considered stable unless it's stable. And that's what these websites do. They're also conservative in their overclocks, TPU achieved 15-20% faster than Titan performance with mild overclocks barely over 1200. Or sometimes less than 1200. Linus hit on this very topic in his latest WAN show from last Friday. Overclocks aren't considered stable. Unless they're really stable. I can do a full Valley run at like 60-100MHz higher than what I consider a "stable" overclock. But I can't play Crysis 3 or Metro: LL for 45 minutes at that clockspeed. Again, not stable unless it's REALLY stable. In everything. So I can't take anyone that does nothing but 2 minute Valley suicide runs (and this is basically, all of HWBOT) seriously in terms of "real world".

So back to HWBOT. LN2 benchmarkers are getting 1170MHz as an overclock on a GTX 780? Uh. Meanwhile, we have overclockers at actual tech review sites getting higher than that on air with aftermarket 780s. So what gives here. Apparently air 780s are getting higher than LN2 780s. Does this even remotely make sense? Guys that MOD their 780s with special VRMs and go all out with LN2? 1170MHz? But then at the same time, they can get 1865VRAM on their 290s.

So we have TPU, Guru3d, hardwarecanucks, among others that have achieved mild overclocks with 15-20% better than Titan performance. You ask for real world benches. There's your "real world" benches. We can go back and forth on this all day long, but the bottom line is practically every PC review website on the net has put the GTX 780 aftermarket cards through the paces in terms of overclocking. Those results have consistently indicated significantly faster than Titan performance. Hexus, Guru3d, hardwarecanucks, HardOCP, TPU, I could go on here. But I don't need to. There's your real world benches, we don't need to discuss it. It's been tested and done, all of these websites have demonstrated overclocks on 780s which result in 10-20% faster than Titan performance. HWBOT certainly isn't real world benching. Those guys are LN2 nuts that use every possible cheat they can, so LN2 overclockers are getting 1170mhz on GTX 780s? Uh. Okay. 1865 VRAM on the 290? Uhm. Yeah.

As far as I can tell the GTX 780 is more or less proven in terms of overclocking. All of these websites, despite having conservative overclocks, are getting excellent performance scaling. Now, the 290X scales well too. But I haven't seen a magnitude of a 20% performance increase in any of these overclocks. I haven't seen any website get past 1200MHz either. Most of them hit a wall in the lower 1100s, in fact.

Now, aftermarket 290X cards may change that. The DC II 290X didn't. Once again, hit a wall in the 1100-1200MHz area per reviewers. But other aftermarket 290X cards? We'll see. Perhaps the lightning 290X will do better. Or a Toxic 290X. Or the Tri-X? (that is reference PCB isn't it?) It is entirely plausible that the current overclock results are due to lack of power delivery. I don't know. I'd say the best PCB thus far is the DC II, but is hitting a wall in the 1100-1200 area like other cards are.

 

[Elfear]

Pretty sure those HWBot numbers are pure bunk.

It's either base clock or something else, also 2068 mem while pure win is simply unrealistic for anything but cherry Samsung 780 Tis. Not to mention my GHz boosts to 1163 out of the box.

Also 1865 mem for 290 is just a joke, who are they kidding?

Umm, I stated that in my post

"The memory overclocks are obviously wrong,

I think 1300 is perfectly acceptable, some 780s will hit it on stock voltage, some crappy ones like mine will require more voltage than stock but it seems within the realm of reasonable clocks as far as GK110 goes. It's not like I have some boss card, 1300 for a Classified or Lightning for example isn't even worth posting about. In fact if that's all they hit you'd be more likely to see that owner post a complaint than anything else.

I guess it comes down to how you want to compare cards. As others have mentioned, hitting over 1300Mhz consistently takes a modded bios. Review sites hit 1163Mhz on average with added voltage so I don't think your claim of 1300Mhz being easy is true.

1170 average for HWBOT? The guys using LN2? That sounds right to you? 1865 on VRAM on the 290. This sounds right to you? Come on. That can't be based on reality.

Did you read my post? Let me quote the parts you didn't catch:

The memory overclocks are obviously wrong...

HwBot puts the average overclock for each card on air

So we have TPU, Guru3d, hardwarecanucks, among others that have achieved mild overclocks with 15-20% better than Titan performance.

Define "mild overclock". The reviews I posted up hit 1163Mhz as their max overclock. In Anandtech's 780 review they saw average boost clocks of 990Mhz and the 780 was 10% slower than a Titan. So with perfectly linear overclock scaling (which is unrealistic on average) the 780 would have to hit 1287Mhz to be 20% faster than the Titan. Realistically it would be more like 1350Mhz with 80% scaling. That doesn't sound like a "mild overclock" attainable by most 780's to me. Feel free to post average overclocking results though if you have better sources.

Look you guys can compare any way you want to (it is an open forum after all ) but posting random benchmarks at random clocks seems like a poor way to come to a valid conclusion.

 

[BallaTheFeared]

Like I said, with b1 out 1163 is a joke. That's the out of box boost for GHz.

I'd say 1212 min for new 780s with no mods, 1300+ with skynet; 1.3v is just a AB softmod.

That's not even including top bins like HoF/Classy/Lightning.

 

[YBS1]

Balla, your FPS graphs look worrisome... there is no way that is playable at ANY framerate. Too many dips and spikes. Same goes for YBS1. You should get that fixed before comparing performance, it is a total mess.

You're seeing an oddity inherent in the benchmark, much like I observed with early build versions of the Tomb Raider benchmark, it would report minimums much lower than anything remotely seen during the benchmark itself, this was resolved after a few patches. This is also something widely know with the Heaven benchmark, it will always report a minimum that is less than half of the lowest observable performance point during the bench. Now why this shows on some systems and not others using Last Light, I don't have a clue. What I can assure you of is at no point during the benchmark was my observable framerate fluctuating wildly back and forth in a way that the black line indicates. Running FRAPS alongside the benchmark shows a minimum of ~47fps without any obvious problems in the frametimes. I had this same type of graph plot with the original game too, using both AMD and nVidia hardware.

 

[YBS1]

[*]Or to "still waiting for the AMD guys to obliterate us", even though that was never the claim.

Again with the goalpost moving, now I am accused of owning a golden sample R9 290X because it can run at 1200 core. Another attempt to justify claiming my results are invalid and don't count anyway?

Actually, that was the claim someone else made earlier in the thread. You do seem to have a golden sample 290X, no need to be offended by that, that's fantastic. I'm assuming you're on air and from what I've seen thus far 1200mhz 290's of either variety are quite rare. It's great that you came in here to bench it with us. I have no issues with the 290's, they are terrific performers and were a (relative) performance bargain at the original price, albeit with a crippled reference cooler that brought some downsides along with it.

 

[Elfear]

Like I said, with b1 out 1163 is a joke. That's the out of box boost for GHz.

I'd say 1212 min for new 780s with no mods, 1300+ with skynet; 1.3v is just a AB softmod.

That's not even including top bins like HoF/Classy/Lightning.

Balla I respect your opinion and all but do you have any hard data to back up your claims of average overclocks? Getting a B1 stepping isn't guaranteed and even if you do you're still limited to +38mV unless you start flashing bioses.

 

[YBS1]

120hz, lightboost, gsync.

All gimmicks almost no one uses. Only games that can hit 120hz without CPU bottlenecks are old engine based games.

Really....

Frames, Time (ms), Min, Max, Avg
20444, 120000, 117, 201, 170.367

Frames, Time (ms), Min, Max, Avg
19989, 120000, 92, 201, 166.575

Frames, Time (ms), Min, Max, Avg
20492, 120000, 95, 201, 170.767

Frames, Time (ms), Min, Max, Avg
9819, 59437, 122, 201, 165.200

Want to take a wild stab at what multiplayer game those numbers are from using the Ultra preset? I'll give you a hint, I don't need Mantle.

 

[Elfear]

120hz, lightboost, gsync.

All gimmicks almost no one uses. Only games that can hit 120hz without CPU bottlenecks are old engine based games.

1440p at 60 FPS is pretty much the desired resolution now for single monitor setups. Thanks to the cheap panels in the market.

I've used a 30" 1600p LCD for years and after trying a 24" 120hz Lighboost panel I game on the latter almost exclusively. If you haven't seen Lightboost in action, you need to check it out. Amazing motion clarity.

 

[Mondozei]

I complained about having to turn my monitor on and off, and having to disable ULPS for crossfire to function properly... Other than that not so much.

It's not like I complained about the stutter, because I was aware of it prior to the purchase assuming it would be fixed and we'd see the memory rewrite everyone was talking about.

It's really telling though, when my 7950 CF was whoopin some 670/680 b-u-t-t not one person from "Team Nvidia" moaned like a moron about bias.

You do not understand. People are telling you you're biased to the card you have at the moment. Because you need to justify the purchase to yourself, like a moron, as you would have said.

 

[Silverforce11]

Like I said, with b1 out 1163 is a joke. That's the out of box boost for GHz.

I'd say 1212 min for new 780s with no mods, 1300+ with skynet; 1.3v is just a AB softmod.

That's not even including top bins like HoF/Classy/Lightning.

I got 1.2ghz undervolted to 1.15vcore on my 7950, doesn't mean most cards would get it..

The average for 7950s is only ~1.15ghz! For whatever golden sample that hits 1.25ghz+, there were duds that couldn't even pass 1.1ghz with volts. Hence, average.

I have yet to see multiple reviews of 780s by professional sites that consistently get above 1.25ghz without going fancy modded bios, outside of [H] and their binned HOF. A lot of reviews put it around the 1.15ghz mark.

While indeed 780s are amazing overclockers, so were 7950s, but lets be realistic with averages.

And finally, you cannot assume most gamers who come to forums asking for advice is willing to go maximum balls to the wall on their 780s with heaps of voltage for it to draw 500W for their gaming enjoyment. Highly unrealistic scenario outside of benchmark e-peen runs. So to say to them, 780s are automatically faster since it can run at 1.3ghz+ is meaningless because who outside of a few is willing to put up with that kind of power consumption spike for a few extra fps, not to mention using modded bioses?

 

[Mondozei]

Basically Ballas claim boils down to his assertion that hitting 1300 is normal for most 780 cards yet he cannot give a source for averages with a large enough pool to justify his claim. Surprising!
P.S breaking your warranty is not something most people are willing to do hence it cannot be counted as average.

And he prefers to avoid using 1440p in his benchmarks, the de facto benchmark standard today. The reason is simple: his card loses out on that. Fanboyism is hilarious when exposed.
And as I said previously, this is a fanboyism that changes on the card the person has, like most fanboys.

To OP: read Elfear's and RussianStallions's posts in this thread. Both are unbiased and objective. They have no agenda to self-rationalize their purchase of a card but give you good advice regardless. And I say this as an owner of an (older) Nvidia card.

 

[TreVader]

From what I can see, the 780 is an inferior card to the 290 and there are a few people who purchased 780s right before the 290s came out and are now doing there best to convince themselves that they didn't waste their money. ATM they are right, the 780 @ 500 is not a waste of money if the 290 is >520, but as soon as the 290 drops below 520 the 780 becomes a worse deal and if it gets anywhere near MSRP then pretty much the entire nvidia lineup looks horribly price-gouged.


AT MSRP The 780ti is a few percent faster than the 290x for an extra $150. The 780 is $100 more than the 290 yet is slower, and Titan is slower than the 290x for an extra $450. This is at stock clocks (which, believe it or not, most people stick with!).

The 780 appears to make up for it's pretty pathetic price/perf ratio with a big OC, but from what i've seen the results Balla gives are not even close to typical. Most people are getting about 1200 core out of their 780s, whereas most people are getting about 1150 core out of their 290s. That doesn't really make the 780 any better than the 290 oc vs oc... they're basically the same with the 780 coming from a bit behind to catch up.


I think the thing that makes Hawaii so amazing is it's offering 64 ROPs and a 512 bit memory bus. Should AMD decide to put 7Ghz vram and up core clocks, there is no way even the 780ti will keep up.

I really am disappointed with the way nvidia is gouging people whenever it feels it can get away with it. I haven't bought a Video card since my 6800 Ultra, but I really liked nvidia back then and at the time they weren't doing this crap. Now that AMD is really giving them a beating I think they might reconsider their strategy.

 

[Pandamonia]

Really....

Frames, Time (ms), Min, Max, Avg
20444, 120000, 117, 201, 170.367

Frames, Time (ms), Min, Max, Avg
19989, 120000, 92, 201, 166.575

Frames, Time (ms), Min, Max, Avg
20492, 120000, 95, 201, 170.767

Frames, Time (ms), Min, Max, Avg
9819, 59437, 122, 201, 165.200

Want to take a wild stab at what multiplayer game those numbers are from using the Ultra preset? I'll give you a hint, I don't need Mantle.

It's not battle field 4 on 64 player China rising with AA on max at ultra. I know that much

 

[TreVader]

From what I can see, the 780 is an inferior card to the 290 and there are a few people who purchased 780s right before the 290s came out and are now doing there best to convince themselves that they didn't waste their money. ATM they are right, the 780 @ 500 is not a waste of money if the 290 is >520, but as soon as the 290 drops below 520 the 780 becomes a worse deal and if it gets anywhere near MSRP then pretty much the entire nvidia lineup looks horribly price-gouged.


AT MSRP The 780ti is a few percent faster than the 290x for an extra $150. The 780 is $100 more than the 290 yet is slower, and Titan is slower than the 290x for an extra $450. This is at stock clocks (which, believe it or not, most people stick with!).

The 780 appears to make up for it's pretty pathetic price/perf ratio with a big OC, but from what i've seen the results Balla gives are not even close to typical. Most people are getting about 1200 core out of their 780s, whereas most people are getting about 1150 core out of their 290s. That doesn't really make the 780 any better than the 290 oc vs oc... they're basically the same with the 780 coming from a bit behind to catch up.


I think the thing that makes Hawaii so amazing is it's offering 64 ROPs and a 512 bit memory bus. Should AMD decide to put 7Ghz vram and up core clocks, there is no way even the 780ti will keep up.

I really am disappointed with the way nvidia is gouging people whenever it feels it can get away with it. I haven't bought a Video card since my 6800 Ultra, but I really liked nvidia back then and at the time they weren't doing this crap. Now that AMD is really giving them a beating I think they might reconsider their strategy.

 

[el etro]

Point remains. 290X stock == Titan stock performance.

Not 290x locked at 1000Mhz...

 

[BertolomeoDiaz]

from what i see on the valley scores chart, guskline doesn´t seem to do much high overclocking, even though he has a custom water loop on his 670sli
https://docs.google.com/spreadsheet...bjQ2UWZQUDZ0dVE&single=true&gid=0&output=html
at oc out of the box with no voltage tweak, i think good custom r9 290 might be stronger

then there is the question of monitor resolution
if you use your full hd non 120Hz monitor i strongly recommend trying downsampling if you never tried it before (especially using such powerful cards/multi gpu setups)
it can be done very easily on nvidia cards (i hear it´s more of an issue on amd cards)
try it out with your gtx 670sli rig, it´s amazing for picture quality

downsampling on 1440p is probably more something for dual titans (6Gb)
but 1440p looks way better anyway, ds not so much recommended as with 1080p

then there is another question: if you ask for both team red and green you obviously don´t prefer either by much

if i had had 4 nvidias as my last cards (and i only know your current rigs) and i´d keep a 670sli setup for doing downsampling on my smaller monitor,
i think i´d go amd for the big monitor (also good suit for having double the vram over 670s), just for trying the other brand for a change (unless i needed an exclusive feature, or it were public consent that either of the choices were crappy, but it sounds like performance is pretty much eye to eye)

it´s exactly what i´m doing
2560x1080 monitor: sli 770s doing downsampling
6000x1080 triple monitor setup: upcoming r9 290 (i really hope it works without display port monitors or accepts my dp-->dvi active adapter)

just that i didn´t change from a load of nvidias, had more amd cards in the past
but that´s when i went 770sli for a change

then again, if the amds are blown up in price due to the mining, maybe 780 is the way to go,
at the same price, for competitive with each other cards, i´d always go nvidia, they somehow have that extra premium sparkle about them (but that´s purely a matter of taste)
it´s like a corvette or a porsche, the corvette will perform as admirably, but if given the choice, price aside, i´d always want the porsche (then again, i´m german, lol)

 

[BallaTheFeared]

Balla I respect your opinion and all but do you have any hard data to back up your claims of average overclocks? Getting a B1 stepping isn't guaranteed and even if you do you're still limited to +38mV unless you start flashing bioses.

Nope, only my two 780s that both boosted to your "max" with 1.16v, both would do over 1200 with 38mv added.

Jaydip hit 1300 on stock voltage.

tviceman hit 1200, but had to drop it back to 1176 after prolonged gaming because he didn't want to use the 38mv.


I don't think we should rule out bios mods, all skynet does is disable boost and unlock the power limit, 1.3v is still achieved by soft modding msi afterburner the same one would do for a R290.

I still like to think of Anandtech as an enthusiast website and not a virtual marketing hub. However in the event we're going to presume otherwise I still think roughly 1200 is good for B1 stepping 780s, which should be replacing all card at some point.

Here are the current bios skynet has for B1 chips:

 

[blackened23]

I still like to think of Anandtech as an enthusiast website and not a virtual marketing hub.

I don't KNOW. AMD pays this website heavily and has the AMD center on the front page and all. Speaking of which. I found the 290 reference review conclusions pretty funny despite the money that AMD throws non-stop at this website. Ryan Smith has cojones - not afraid to speak his mind despite the cash that AMD funnels into anandtech. Kudos for that.

Anyway, numerous websites have tested aftermarket 780s with excellent scaling. Most of these websites achieved 15-20% faster than Titan performance without BIOS mods. Without anything. Except +38mV. These were all aftermarket cards. Aftermarket cards do OC better than the reference, unless you BIOS mod the reference. The aftermarket cards do not need BIOS mods, generally speaking. The same could be true for other 290X aftermarket cards (aftermarket overclocking WAY better than reference) and I would expect that to be the case. We'll just have to wait and see. I will note the Asus DC II 290X didn't seem to overclock any better than reference, though, for some reason. On the other hand, it doesn't throttle while overclocking either (in performance mode).

These 780 aftermarket overclocks weren't exceptionally high either. Many weren't past 1200. Still waiting on the 290X review at hardwarecanucks, guru3d, techpowerup, hardocp, etc that shows a 290X to be 15-20% faster than stock Titan. For some strange reason HardOCP isn't even overclocking 290 or 290X cards. What's up with that.

Asus 290X review:

http://www.hardocp.com/article/2013/12/23/asus_r9_290x_directcu_ii_oc_video_card_review#.UsGdNvRDuaU

Notice what's missing. NO OVERCLOCKING SECTION. Every video card HardOCP has ever reviewed has had an overclocking section. I guess they're just making exceptions for the AMD 290X now. Odd. Whats up with that.

Anyway, of course the 290X could do better with aftermarket varities just like the 780 does. Completely plausible and we'll see. This goes back to the main point. The GTX 780 aftermarket cards are proven to overclock well and scale well. The 290X? It overclocks semi decently but I haven't seen anything mind blowing yet. But not many aftermarket 290X cards have been tested. Soo.......we'll see. We'll have more data in the coming weeks when more aftermarket 290X cards hit.

 

[PPB]

AMD pays to have the MKT centered in one place in the frontpage, I wonder who takes care for the rest of the site (including this forum)? :think:

 

[blastingcap]

AMD pays to have the MKT centered in one place in the frontpage, I wonder who takes care for the rest of the site (including this forum)? :think:

AMD has a stealth marketing campaign giving free stuff to "respected" "influencers" who spam forums, Facebook, and Twitter full of pro-AMD stuff, so actually, AMD pays for both. That's not including employees and AMD affiliates (e.g., subcontractors, AIB partners) who do not often identify themselves. I wish I were joking.

 

[rtsurfer]

I don't KNOW. AMD pays this website heavily and has the AMD center on the front page and all. Speaking of which. I found the 290 reference review conclusions pretty funny despite the money that AMD throws non-stop at this website. Ryan Smith has cojones - not afraid to speak his mind despite the cash that AMD funnels into anandtech. Kudos for that.

This is pretty awesome.

You went from questioning the credibility of the forum members to the credibility of the site reviewers themselves.

Considering these people have been reviewing stuff for years now anf all of a sudden they are AMD biased as they are paid yo hace a little sub-portal for AMD on their front page and mayne because AMD sends them HW first then other reviewers to review.

You said AMD is paying this site heavily.

How much heavily..?

Do you know.??

Get me a figure.

This is just ridiculous how people like you will go so far to prove their point, even trying to discredit respected Tech Journalists like Anand with years of experience under that belt. This is just disrespectful.

Do yourself a favour if you have doubts regearding the credibility of the reviews of this site then get off its forums.
GO SOMEWHERE ELSE..

 

[BallaTheFeared]

You went from questioning the credibility of the forum members to the credibility of the site reviewers themselves.

The only thing that happened was an R290x at 1200MHz failed to beat a 1300MHz 780.

So between the shilling and member call outs of bias there is a discussion going on about what is a fair clock speed for comparison between Hawaii which doesn't clock as well as Kepler.

I bios modded my 7950s, I got bios mods directly from MSI to flash my cards with. I happen to believe modding is fair practice, others wish to disagree. All this is fine, we can have conversations like this without the pretty awful level of crosstalk that is taking place in this thread which has sadly become the norm for this forum.

Some people don't overclock, some overclock only with default voltage, some use max voltage allowed and yet still others wish to get the most from their hardware and do so by various means including changing coolers, modding, soft mods, bios flashes and everything else under the sun.

I'm sure we can come to an agreement on where we should baseline the products, even with all the fud posts. All that is required is that they be ignored.