[H] MSI Radeon R9 290X Gaming 4G Video Card Review

[toyota]

I only looked at the Metro results so far and there is NO WAY a 780 ti gets that much better performance than the 290x. in fact on techpowerup they got 55 fps for the same MSI 780 ti on the same settings where hardocp is getting 69 fps. I know that is real gameplay compared to the benchmark but it makes no sense for the 780 ti to do 14 fps better than the bench while the 290x is getting only 6 fps more than the bench.

 

[Attic]

The problem is that, aside from the 1 BF4 graph (which may very well be a mix up of numbers on the part of the author), the OC still outperforms the stock results on every game they tested. In order for the card to be throttling, it would have to be throttling at both stock and overclocked, yet the throttle when OC would still not be as severe as the throttle when not overclocked? It just doesn't make sense to blame the review on throttling. Reverse the numbers on the single apples-to-apples BF4 graph and everything suddenly checks out. The OC'd results beating stock across the board, the performance scaling relative to the overclock, and the power consumption of stock vs. overclocked....

Yeah, no throttle going on here, other than the 780 TI throttling the 290x.

I thought the idea was that additional voltage, +50mv, was applied to the OC settings of the 290x in [H] review while no additional voltage was applied when not overclocked. Not sure how that would affect performance numbers of a higher core value but intermittent throttling vs lower core with no throttling, but I think if the bug in the 14.2's was affecting [H] results that what we're seeing from [H] lines up with that bug.

There was a pretty heavy overclock put on the 780ti so we may just be seeing OC 780ti show it's wings, but I think there's more to the performance numbers of the 290x OC in this review which look low independent of the the 780ti.



Look at the performance numbers here for the 290x XFX when oc'd to 1130/5800 vs what identical clocks on the 290x from [H] recent review show.
http://www.hardocp.com/article/2014/02/24/xfx_r9_290x_double_dissipation_overclocking_review/5

XFX 290x (1130/5800)
Metro Last Light -64.6
Far Cry 3 - 48.7
Crysis 3 -52.0

MSI 290x (1130/5400)
Metro Last Light -58.3
Far Cry 3 -44.6
Crysis 3 -42.2

Other games show similar performance drops in the MSI 290x review. Perhaps it's memory anomaly, but the XFX 290x at 1130/5800, decent OC shows a more believable representation of performance of the 290x when OC'd.

 

[KaRLiToS]

It's weird. Because all my R9 290x overclock up tp 1275mhz/1650mhz if I add enough powertune and voltage.

 

[tviceman]

I only looked at the Metro results so far and there is NO WAY a 780 ti gets that much better performance than the 290x. in fact on techpowerup they got 55 fps for the same MSI 780 ti on the same settings where hardocp is getting 69 fps. I know that is real gameplay compared to the benchmark but it makes no sense for the 780 ti to do 14 fps better than the bench while the 290x is getting only 6 fps more than the bench.

I just benched my 780 @ 1189 / 6900 to compare with the techpower msi gaming 780 ti results, and I got 53.15 fps, only 4% slower than the MSI gaming 780 TI stock (which boosts quite high without an overclock) and a whopping 30% faster than the reference 780 scores they have listed.

My point is, I think there's been some notable driver improvements since that review came out, which is dated JANUARY 7th (4 months ago).

 

[VulgarDisplay]

It's weird. Because all my R9 290x overclock up tp 1275mhz/1650mhz if I add enough powertune and voltage.

On what drivers? This bug only occurs on 14.x drivers. The other thing is that you may be throttling and not notice if you aren't actively monitoring.

Also since your cards are water cooled it makes sense that their lower temp means less power use. They can clock higher before hitting the hard wattage wall a lack of power limit increase has.

 

[Silverforce11]

@Attic
From that result, the R290X is clearly bandwidth starved... :/ (it's not)

[H] also had the worse ASUS R290X, horribly loud fans to keep it from overheating when other review sites had a completely different result. I don't blame them, ASUS did have a bad batch. But here they are slack of late. Benching this card that has much worse performance that prior cards and not bothering to ask why? If there's such a huge variation between runs, look into it [H], otherwise its just slack. Are we to think that the benches vary by that much is normal?

 

[toyota]

I just benched my 780 @ 1189 / 6900 to compare with the techpower msi gaming 780 ti results, and I got 53.15 fps, only 4% slower than the MSI gaming 780 TI stock (which boosts quite high without an overclock) and a whopping 30% faster than the reference 780 scores they have listed.

My point is, I think there's been some notable driver improvements since that review came out, which is dated JANUARY 7th (4 months ago).

you must be using a magical driver then because even at a lower 2560x1440 I cant even get but 45 fps with my 780 clocked at nearly the same as yours. I am on 334.89 drivers which are newer than hardocp is using.

 

[tviceman]

you must be using a magical driver then because even at a lower 2560x1440 I cant even get but 45 fps with my 780 clocked at nearly the same as yours. I am on 334.89 drivers which are newer than hardocp is using.

I'm on 335.23. It's MAGICAL, just like you said.

Would you like me to run it again with afterburner and all that other good stuff on the screen?

EDIT: HAHA I have my 4770k running a little slower than yours, 4.2ghz. I've had HT disabled lately, though I doubt HT makes much of a difference either way.

 

[toyota]

I'm on 335.23. It's MAGICAL, just like you said.

Would you like me to run it again with afterburner and all that other good stuff on the screen?

ok but hardocp is not using that driver so my point still stands. plus 2560x1600 is over 10% more demanding than 2560x1440.

HT actually helps a little bit in the benchmark for this game.

 

[Attic]

@Attic
From that result, the R290X is clearly bandwidth starved... :/ (it's not)

[H] also had the worse ASUS R290X, horribly loud fans to keep it from overheating when other review sites had a completely different result. I don't blame them, ASUS did have a bad batch. But here they are slack of late. Benching this card that has much worse performance that prior cards and not bothering to ask why? If there's such a huge variation between runs, look into it [H], otherwise its just slack. Are we to think that the benches vary by that much is normal?

Yep, that was one of my first thoughts. I thought it was clear the results in their review were abnormal (though [H] explained the massive crush the 780ti put on the 290x by the OC on the 780ti) and deserving of some critical evaluation that I felt was lacking in their conclusion. Their 290x XFX OC'ing review appears to indicate something unique was going on with either the 290x XFX OC or the 290x MSI OC.

If the 780ti is pummling the 290x by 25% then the 780ti is a no brainer and the 290x is out of it's league, but the conclusion in the review kinda laid it out like the 290x is a good/great card at it's price.

 

[tviceman]

Is it entirely possible they are using different benchmarks of the same game each time they put out a new review?

 

[Schro]

I keep an eye out over here from time to time as well. Will get around to answering on the [H] side of the discussion as well shortly. I'm currently in the process of moving a couple of states away, so my rigs and data are packed at the moment.

I only looked at the Metro results so far and there is NO WAY a 780 ti gets that much better performance than the 290x. in fact on techpowerup they got 55 fps for the same MSI 780 ti on the same settings where hardocp is getting 69 fps. I know that is real gameplay compared to the benchmark but it makes no sense for the 780 ti to do 14 fps better than the bench while the 290x is getting only 6 fps more than the bench.

Techpowerup does not explicitly state which level they used for testing, so I do not think that you can make a fair comparison between our data (taken from the Red Square map, from the start up to the end of the shootout) and their data, unless, of course, they did the exact same runthrough. Performance variations within a single game are rather significant, so you can't average a frame rate from one level and then compare it to a frame rate on another level.

I thought the idea was that additional voltage, +50mv, was applied to the OC settings of the 290x in [H] review while no additional voltage was applied when not overclocked. Not sure how that would affect performance numbers of a higher core value but intermittent throttling vs lower core with no throttling, but I think if the bug in the 14.2's was affecting [H] results that what we're seeing from [H] lines up with that bug.

There was a pretty heavy overclock put on the 780ti so we may just be seeing OC 780ti show it's wings, but I think there's more to the performance numbers of the 290x OC in this review which look low independent of the the 780ti.



Look at the performance numbers here for the 290x XFX when oc'd to 1130/5800 vs what identical clocks on the 290x from [H] recent review show.
http://www.hardocp.com/article/2014/02/24/xfx_r9_290x_double_dissipation_overclocking_review/5

XFX 290x (1130/5800)
Metro Last Light -64.6
Far Cry 3 - 48.7
Crysis 3 -52.0

MSI 290x (1130/5400)
Metro Last Light -58.3
Far Cry 3 -44.6
Crysis 3 -42.2

Other games show similar performance drops in the MSI 290x review. Perhaps it's memory anomaly, but the XFX 290x at 1130/5800, decent OC shows a more believable representation of performance of the 290x when OC'd.

There are other factors to consider when comparing numbers from a review that I do compared to one that Brent does. While we do use the same level and similar routine, how I move through that level will not always be the same as how he moves through a level. Also, we do not always record FPS data for the same length of time - for example, his Crysis 3 runthrough is nearly 10 minutes (which I did for quite a while), but more recently I have been curtailing it a bit earlier in the level. If the average frame rate is different on the first half of the map compared to the second, you will see differences in the results between reviews. I would suggest that our average FPS numbers are not great to compare across reviews or reviewers even though they will be great direct comparisons within that particular review (or perhaps to another review done around the same time by the same reviewer). Our observation about best playable settings, in general, will be fairly consistent across our reviews (unless there's an edge case where we make a judgment call).

As for the overclocking side of things, the card did NOT throttle and OC performance at +25mV as well as +40mV was much lousier than it was at +50mV. With regards to the BF4 numbers, I suspect I put a couple numbers in the wrong cell. I was wrestling with that chart for about an hour before I got it right (adding a 4th card to that was NOT fun), so I suspect I put the wrong number in. The OC'ed performance did give a better average FPS and I'll go back through my data and notes once the movers deliver it to my new house (read: might be a bit).

The other thing to keep in mind is that we do our BF4 testing in full 64 player servers, so there is a significant amount of variability depending on the action within the server. Prevalence of explosives, number of deaths and other shenanigans can influence the average FPS number by a significant amount.

It's weird. Because all my R9 290x overclock up tp 1275mhz/1650mhz if I add enough powertune and voltage.

You must have quite a good card. I've tried at least a half dozen 290X's and can't seem to get one to stabilize beyond 1130MHz. I had the MSI one up to about 1180, but it wasn't stable in all of the games I threw at it, so it got knocked down to 1130 before it would no longer crash.

Is it entirely possible they are using different benchmarks of the same game each time they put out a new review?

We don't use benchmarks. We play through a particular level of the game (as specified in the best playable settings chart). As a general rule, we don't change levels between reviews unless there's a good reason. I've explained other sources for variance between [H] reviews above.

 

[tviceman]

ok but hardocp is not using that driver so my point still stands. plus 2560x1600 is over 10% more demanding than 2560x1440.

Yet my card still poops all over your card at the same resolution and clock speeds! MWAHAHAHHA. No longer will I just look down upon owners of other vendor-based cards, I shall look down upon all 780 owners who cannot defeat me in benchmarking battles. TO THE DEATH!

HT actually helps a little bit in the benchmark for this game.

Don't tempt me.

 

[tviceman]

We don't use benchmarks. We play through a particular level of the game (as specified in the best playable settings chart). As a general rule, we don't change levels between reviews unless there's a good reason. I've explained other sources for variance between [H] reviews above.

That was what I meant by "benchmarks." My stupid vocabulary! You are doing what I suspected - recreating new runs through each game you benchmark with each new video card review.

 

[blackened23]

David replied:

http://hardforum.com/showthread.php?p=1040741775#post1040741775

The card did NOT throttle and OC performance at +25mV as well as +40mV was much lousier than it was at +50mV.

Myth debunked. By the way, plenty of people got throttling on 14.1 do not get it with 14.2 with the newest afterburner. But in any case. AS I SAID. Both David and Brent monitor GPU clockspeeds while benching. If a card throttles, they know. They have stated this in many reviews.

These results really aren't out of line as some here would suggest. Go look at the Asus 290X DC II maximum overclock vs 780ti maximum overclock *reference* review. The 780ti won by a commanding lead in every benchmark IIRC. Because GK110 scales better with overclocking.

So this review has the MSI gamer 290X overclocked, and the 780ti gamer overclocked. So let's use some common sense here based on the fact that the reference 780ti max OC beat the 290X DC II max OC by a huge margin. 780ti overclocked == GK110 benefits more from overclocking than the 290X. These results aren't much different than the aforementioned Asus DC II 290X vs 780ti maximum overclock shootout. The reference 780ti won handily there, just like the MSI gamer 780ti is winning handily here. Not hard to figure out. At stock settings, the cards are close. At maximum overclocks? They're not close, GK110 takes the lead.

Now David did mention he's double checking the BF4 results. But the review results were not affected by throttling whatsoever, as the GPU clockspeeds were monitored the entire time with Afterburner's overlay.

 

[Silverforce11]

@Schro
If you guys bench things differently between runs and person doing the testing, it would lead to huge variations which we've seen of late in your reviews.

For example, person A will do a run through at a particle area of a level, that is either more or less punishing on GPUs compared to person B.

Why do you guys not standardize the run and play through? Picking a random section is going to skew the result and not representative. One only has to look at at certain areas of the game more often that leads to a sharp FPS drop, to affect the final average FPS. Example: Crysis Waterfall scene here on AT they used to love testing it.

The only conclusion from comparing your own reviews:
XFX R290X is magical.
MSI R290X is crap.
More recent AMD drivers lose massive performance.
Your testing methodology is skewing the outcome.

Which is it?

 

[Schro]

That was what I meant by "benchmarks." My stupid vocabulary! You are doing what I suspected - recreating new runs through each game you benchmark with each new video card review.

I make sure to differentiate that as most other review sites tend to run true "benchmarks".

Quite honestly, I do not go back and gut check the min/max/average FPS that I get in a particular game to our other reviews. I do tend to gut check against best playable settings where I feel like something is amiss, but it is quiet possible that I could have a differing opinion on a best playable setting compared to Brent or Grady.

 

[toyota]

Yet my card still poops all over your card at the same resolution and clock speeds! MWAHAHAHHA. No longer will I just look down upon owners of other vendor-based cards, I shall look down upon all 780 owners who cannot defeat me in benchmarking battles. TO THE DEATH!



Don't tempt me.

oops I had my card at stock. at 1124/6200 I got 50.25 fps which is line with what you got. I only do 1176/6500 as my max if its really important for a bench but I dont like to use it since it needs a bump in voltage.

 

[blackened23]

oops I had my cards at stock as. at 1124/6200 I got 50.5 fps which is line with what you got. I only do 1176/6500 as my max if its really important for a bench.

Toyota, keep in mind that these numbers are considered "maximum overclock" numbers. You are correct that the 780ti and 290X are relatively close at stock settings. However at maximum OC's, the 780ti pulls away by a very sizable margin.

I mentioned this before, but look at HardOCPs reference 780ti MAX OC versus Asus DC II 290X max OC review. The 780ti reference MAX OC handily beat the asus card in that review despite being a reference design. The same thing is happening in this review; what happens at stock clocks doesn't happen at maximum overclocks - the 780ti pulls way ahead, essentially. H is testing both cards at near their maximum overclocks. Therefore that would heavily favor the GK110.

For what it's worth, Linus (of linus tech tips) has tested maximum overclocks on all of these cards as well numerous times, and every test he did at maximum overclocks (on his youtube) has the GK110 winning. Yet at stock, the 290X can win many of those same tests. So the recurring them here is what happens at stock benchmark shootouts doesn't happen at maximum overclocks.

 

[Silverforce11]

Toyota, keep in mind that these numbers are considered "maximum overclock" numbers. You are correct that the 780ti and 290X are relatively close at stock settings. However at maximum OC's, the 780ti pulls away by a very sizable margin.

I mentioned this before, but look at HardOCPs reference 780ti MAX OC versus Asus DC II 290X max OC review. The 780ti reference MAX OC handily beat the asus card in that review despite being a reference design. The same thing is happening in this review; what happens at stock clocks doesn't happen at maximum overclocks - the 780ti pulls way ahead, essentially. H is testing both cards at near their maximum overclocks. Therefore that would heavily favor the GK110.

Obviously you forgot that the [H] ASUS DCII required very high fan speed to prevent it from hitting 94C.. something other sites do not experience, far from it. They had terrible scaling in performance for OC DCII when other sites disagree.

http://www.guru3d.com/articles_pages/asus_radeon_r9_290x_directcuii_oc_review,28.html

So either [H]'s DCII sample was bad (and ASUS/Gigabyte sent out bad samples per their own statements), or Guru and other sites are somehow magical samples.

 

[Attic]

snip.

Thanks for dropping in and explaining. That makes a lot of sense.

I didn't notice the reviews were done by different author's, had Brent pegged for all the recent stuff. That along with the different in game run thrus explains the sizable differences between the xfx OC review.

 

[blackened23]

Obviously you forgot that the [H] ASUS DCII required very high fan speed to prevent it from hitting 94C.. something other sites do not experience, far from it. They had terrible scaling in performance for OC DCII when other sites disagree.

http://www.guru3d.com/articles_pages/asus_radeon_r9_290x_directcuii_oc_review,28.html

So either [H]'s DCII sample was bad (and ASUS/Gigabyte sent out bad samples per their own statements), or Guru and other sites are somehow magical samples.

Hah. I guess linus had some bad samples as well then. I could find some more reviews with these so called "bad samples". How many reviews should I find with bad samples. There's legitreviews which shows 780ti winning across the board at 4k resolution with the newest WHQL's.

Not sure why you go to such extreme efforts to discredit a review website when the same theme occurs across multiple reviews. GK110 is better at overclock scaling, but both cards are close at stock settings. It's that simple really. No need for discrediting and/or far fetched theories. That does not mean the 290X is a bad GPU, so i'm not sure why some are taking such offense to this. 290X is a great GPU now that it is selling at MSRP levels - it's a much better value than the 780ti despite any performance difference. The mining tax of prior months is basically gone and done with now. That means you can buy a quiet aftermarket 290X GPU for sometimes 100$ less than the 780ti. What's not to like? 290X doesn't have to always be the fastest, because it's a much better value at a much better price point. It's a great card for those who favor AMD. But if you want the fastest as many do, you pay the premium for the 780ti. It's that simple.

Anyway, it is what it is. I'll let you dissect the information, and if you want to discredit websites on this, that's on you.

 

[toyota]

Toyota, keep in mind that these numbers are considered "maximum overclock" numbers. You are correct that the 780ti and 290X are relatively close at stock settings. However at maximum OC's, the 780ti pulls away by a very sizable margin.

I mentioned this before, but look at HardOCPs reference 780ti MAX OC versus Asus DC II 290X max OC review. The 780ti reference MAX OC handily beat the asus card in that review despite being a reference design. The same thing is happening in this review; what happens at stock clocks doesn't happen at maximum overclocks - the 780ti pulls way ahead, essentially. H is testing both cards at near their maximum overclocks. Therefore that would heavily favor the GK110.

no I am looking at the stock cards and compared the 780ti to the exact model on techpowerup. the only thing I was looking at was that the 780ti seems to be doing way better in the actual game than the 290x based on their respective performance in the built in benchmark. maybe just the part they are testing favors the Nvidia card. plus the built in benchmark is little more cpu intensive than the actual game so in real gameplay the 780ti can flex a little more.

 

[KaRLiToS]

There are other factors to consider when comparing numbers from a review that I do compared to one that Brent does. While we do use the same level and similar routine, how I move through that level will not always be the same as how he moves through a level. Also, we do not always record FPS data for the same length of time - for example, his Crysis 3 runthrough is nearly 10 minutes (which I did for quite a while), but more recently I have been curtailing it a bit earlier in the level. If the average frame rate is different on the first half of the map compared to the second, you will see differences in the results between reviews. I would suggest that our average FPS numbers are not great to compare across reviews or reviewers even though they will be great direct comparisons within that particular review (or perhaps to another review done around the same time by the same reviewer). Our observation about best playable settings, in general, will be fairly consistent across our reviews (unless there's an edge case where we make a judgment call).

So his Crysis 3 runthrough is almost 10 minutes? How can you replicate an identical scene each time you do the tests?

It is innacurate and will skew the results. Hence, there is so few differences between the 290x and the GTx 780ti that every FPS counts


As for the overclocking side of things, the card did NOT throttle and OC performance at +25mV as well as +40mV was much lousier than it was at +50mV. With regards to the BF4 numbers, I suspect I put a couple numbers in the wrong cell. I was wrestling with that chart for about an hour before I got it right (adding a 4th card to that was NOT fun), so I suspect I put the wrong number in. The OC'ed performance did give a better average FPS and I'll go back through my data and notes once the movers deliver it to my new house (read: might be a bit).

You put wrong numbers in the wrong cell and it's been like that since January? A couple of mistaken numbers can be inoffensive but I don't think you realize the amount of collateral damages it can do on the forums with the Nv Fanboys and AMD fanboys.

It can also mislead the choice of people.


The other thing to keep in mind is that we do our BF4 testing in full 64 player servers, so there is a significant amount of variability depending on the action within the server. Prevalence of explosives, number of deaths and other shenanigans can influence the average FPS number by a significant amount.

Exactly



You must have quite a good card. I've tried at least a half dozen 290X's and can't seem to get one to stabilize beyond 1130MHz. I had the MSI one up to about 1180, but it wasn't stable in all of the games I threw at it, so it got knocked down to 1130 before it would no longer crash.

No I don't have cherry picked cards. I don't even have hard time stabilizing my cards at +1200/1500 with + 200mv . I don't know how come you can't past 1130mhz.

I can reach 1300mhz on my cards with pt1 BIOS.



We don't use benchmarks. We play through a particular level of the game (as specified in the best playable settings chart). As a general rule, we don't change levels between reviews unless there's a good reason. I've explained other sources for variance between [H] reviews above.

.....

 

[tviceman]

oops I had my card at stock. at 1124/6200 I got 50.25 fps which is line with what you got. I only do 1176/6500 as my max if its really important for a bench but I dont like to use it since it needs a bump in voltage.

So my drivers aren't magical after all?