Updated List of Video Card GPU OVERALL Performance VP Ratings - TITAN update!

[Will Robinson]

Well Bo,I don't really agree with your video card politics but I have to give you mad props for a very good thread.
Obviously took a lot of work and is commendable.
Well done.:thumbsup:

 

[Zanovar]

Well Bo,I don't really agree with your video card politics but I have to give you mad props for a very good thread.
Obviously took a lot of work and is commendable.
Well done.:thumbsup:

hear,hear good stuff Bofox.:thumbsup:

 

[BoFox]

Thank you both!

Anybody wanna guess what GTX Titan will be rated at?

 

[Will Robinson]

Would that be for Frames per $Dollar
or Smooothness/watt?:whiste:

 

[railven]

It actually runs pretty cool for its temperament, and can be overclocked like MAD!!! Truly the world's fastest!! $1500 is about as high as it gets now. Will we be seeing an overclocked "Mars III" dual-Titan card for $2000 on Newegg?

The day I buy a $2,000 vidjah cahd is the day I won the lottery, or lost my mind, home, and responsibilities. Haha.

 

[BoFox]

Ha!

Oh, come on, guess!! Just a rough number.. what do you think GTX Titan would rate at?!?

I haven't gotten around to the lesser mobile GT(X)-600M series based on GK107, but I went ahead and added:

Geforce GTX 675MX 4GB (DX11.1) -- 110 VP
Geforce GTX 670MX 3GB (DX11.1) -- 94 VP

Whew!!! Let me call it a day!

 

[badb0y]

For Titan I expect 365-375 range.

 

[BoFox]

UPDATE:

Finally added the following:
Geforce GTX 660M 2GB (DX11.1) -- 66 VP

Geforce GT 650M 2GB (GDDR5) (DX11.1) -- 58 VP

Geforce GT 645M 2GB (GDDR5) (DX11.1) -- 56 VP

Geforce GT 640M 2GB (GDDR5) (DX11.1) -- 51 VP

Tons of fine-tuning was done with the mobile GPU ratings, also!
TITAN will be next!!! I'm guessing around 350 VP!

 

[hyrule4927]

UPDATE:

Finally added the following:
Geforce GTX 660M 2GB (DX11.1) -- 66 VP

Geforce GT 650M 2GB (GDDR5) (DX11.1) -- 58 VP

Geforce GT 645M 2GB (GDDR5) (DX11.1) -- 56 VP

Geforce GT 640M 2GB (GDDR5) (DX11.1) -- 51 VP

Tons of fine-tuning was done with the mobile GPU ratings, also!
TITAN will be next!!! I'm guessing around 350 VP!

660M is underperforming the 560M? Certainly makes me feel better about the old 460M in my laptop.

 

[BoFox]

Well, there are two versions of 560M - the 2GB version is the much slower one, having only 40GB/s bandwidth and 2/3 the ROPs. I should include this version, so that Nvidia doesn't "get away with it", hehe! (Thanks for the reminder!)

UPDATE-

I did a quick algorithm approximation, since there are not many reviews on this particular configuration that was sneaked in after Notebookcheck.net did all the reviews on the 1.5GB version.
http://technopat.net/forum/blogs/burak-alkan/36-gtx-560m-128-bit-2gb-slower-than-192-bit-1-5gb.html

First, what I did was apply 3% performance penalty for every 10% reduction in bandwidth.
Then I applied another 1% penalty for every 10% reduction in pixel fillrate.

Since the 1.5GB version had 50% more bandwidth, that meant 15% for seriously low bandwidth of 40GB/s, even relatively to its specs.
50% more ROPs meant another 5% for a considerable reduction in pixel fillrate, with only 16 ROPs rather than 24. 1% per 10% increase in pixel fill rate is usually modest especially if there is that much of a reduction, but since the card is a bit weak, and most people won't be using AA with it, it's not that modest really..
So that's a total of 20%.
67 VP divided by 1.20 (20%) equals 55.8 VP, but it should be rounded down to 55VP because 40GB/s is really low, perhaps it should be a bit more than 16% penalty rather than 15%.

That's the generic algorithm according to the "GPU Theory of Relativity" currently under development. Usually, on average, there is about 2% performance for 10% of bandwidth, but in this case with abnormally low bandwidth ceiling, 3% is a tiny bit modest.

Added: Geforce GTX 560M 2GB (DX11) -- 55 VP

 

[railven]

You ready to add in Titan? No one cares about mobile parts right now. Haha, get back to work! Someone crack the whip!

 

[BallaTheFeared]

You ready to add in Titan? No one cares about mobile parts right now. Haha, get back to work! Someone crack the whip!

:thumbsup:

 

[f1sherman]

Further evidence is needed, but if CF behavior shown bellow is common, and not a fluke, CF scores need to be multiplied with 2/3, then further penalized for skipping every 3rd frame,
(and recommended only to your gf, moms bf and similar.)

If PCPER findings get confirmed I vote CF for the biggest tech joke ever :ninja:


We aren't ready to show our full sets of results yet (soon!) but the problems lie in that AMD's CrossFire technology shows severe performance degradations when viewed under the Frame Rating microscope that do not show up nearly as dramatically under FRAPS. As such, I decided that it was simply irresponsible of me to present data to readers that I would then immediately refute on the final pages of this review - it would be a waste of time for the reader and people that skip only to the performance graphs wouldn't know our theory on why the results displayed were invalid.

Many other sites will use FRAPS, will use CrossFire, and there is nothing wrong with that at all. They are simply presenting data that they believe to be true based on the tools at their disposal. More data is always better.

Here are these results and our discussion. I decided to use the most popular game out today, Battlefield 3 and please keep in mind this is NOT the worst case scenario for AMD CrossFire in any way.


http://www.pcper.com/reviews/Graphi...ance-Review-and-Frame-Rating-Update/Frame-Rat


Notice the thin silver band between the maroon and purple? That is a complete frame according to FRAPS and most reviews. Not to us - we think that frame rendered is almost useless.

 

[hyrule4927]

Further evidence is needed, but if CF behavior shown bellow is common, and not a fluke, CF scores need to be multiplied with 2/3, then further penalized for skipping every 3rd frame,
(and recommended only to your gf, moms bf and similar.)

So, when exactly did you become the exclusive authority on how Voodoopower ratings are assigned? Not saying that this issue shouldn't be taken into consideration, but I'm not sure why you are demanding that BoFox use your suggested calculation method . . .

Edit: To clarify, I'm not trying to jump on you here, I just thought your post was phrased in an oddly imperative fashion, considering the ratings are assigned at BoFox's discretion.

 

[f1sherman]

I am talking common sense approach. Not demanding anything duh...

We wait and see how this turns out.

This reminds me of "what do you want"

^_^

 

[wand3r3r]

So, when exactly did you become the exclusive authority on how Voodoopower ratings are assigned? Not saying that this issue shouldn't be taken into consideration, but I'm not sure why you are demanding that BoFox use your suggested calculation method . . .

Edit: To clarify, I'm not trying to jump on you here, I just thought your post was phrased in an oddly imperative fashion, considering the ratings are assigned at BoFox's discretion.

lol :thumbsup:

 

[BoFox]

It seems that 1920x1080 results for GTX Titan should not be considered, since some of the 1080p benches are already bottlenecked by the CPU. For example,

http://www.guru3d.com/articles_pages/geforce_gtx_titan_review,13.html

The VP ratings only cover the performance of the GPU, not the CPU. So I have to ditch 1920x1200 and below, in order to be able to fairly assess the GPU. After all, 90+% of the owners of Titan will most likely be playing on 1440p or higher.

@f1sherman, here something else I discovered about GTX 690 - that it STILL has microstuttering (EXTREME in some cases)..

 

[f1sherman]

you didn't really believe that that MS is non-issue with 690, did you

 

[BoFox]

At first, it seemed like GTX 690 pretty much tackled microstuttering issues with vastly improved "frame metering" algorithm, when comparing it against HD 6990 and GTX 590:

http://techreport.com/review/22890/nvidia-geforce-gtx-690-graphics-card/3
It even did much better than GTX 680 in SLI:

It was so impressive at first... I actually believed Nvidia that GTX 690 tackled the issue at large.

 

[BoFox]

Then now we're seeing this about GTX 690:
From: http://www.hardwarecanucks.com/foru...orce-gtx-titan-6gb-performance-review-15.html

For GTX 690, microstuttering is rather severe - there, Titan would feel just as smooth as GTX 690. The actual perceived smoothness would probably be only a few % in favor of GTX 690, like 65 "effective" fps vs Titan's ~60fps, after accounting for the microstuttering.

Rather mild microstuttering for GTX 690, but not bad enough to really hurt its distance ahead of Titan.

Whoa! That's EXTREME microstuttering. We could just as well cut the fps counter in half for GTX 690 right there, almost! With the "quick" frame times at only 5ms, they are nearly non-existent, so you could just as well be looking at the "long" frame times. Overall, that would make it feel more sluggish than both Titan and 7970GE.

Finally, almost zero microstuttering for GTX 690. 7970GE has severe stuttering/jittering problems here.

While the 7970GE continues to stutter, the 690 has constant microstuttering there, alternating between ~10ms and ~22ms for each frame. Titan would actually be smoother than 690, with its red line right below the middle of 690's fat blue "stream". (BTW, Apoppin just said that with the latest beta drivers, the 7970 no longer has any stuttering problems in Hitman - and that it's actually putting out even higher fps.)

MILD microstuttering for 690, but...

Gosh, change the resolution and 690 gives EXTREME microstuttering. The 5ms frames are nearly non-existent in relation to the FAR (5 times) slower 25ms frames. I'd say that Titan is smoother, while 7970GE gives roughly equivalent experience right there.

More 5760x1080 results here: http://www.hardwarecanucks.com/foru...orce-gtx-titan-6gb-performance-review-16.html

With extreme microstuttering, FPS is almost half a lie!

PcPer shows a bit worse microstuttering with GTX 690 for Crysis 3:

I'm amazed at this chart, showing 50th percentile, 75th, 95th, and 99th!! That's another way to show it really well!

(But GTX 690 does a bit better at 2560x1440). http://www.pcper.com/reviews/Graphi...ance-Review-and-Frame-Rating-Update/Crysis-3-

Great job, PCPerspective and HardwareCanucks with using other means of measuring frame time "disparity" that Techreport basically pioneered with 99th percentile times and "time spent beyond X ms"!

 

[BallaTheFeared]

I think you missed his point, besides 690 stutter is no worse than 7970 single stutter...

It's the fact that CFX is counting frames that aren't rendered, inflating their "scaling"/fps scores.

 

[BoFox]

Yeah I know (and AMD's driver team is hard at work on this, resolving it for some games already), but Tomshardware is saying almost the same thing about GTX 690:

Now, here is one example of where GeForce GTX 690’s performance doesn’t appear to be accurately represented due to the point in the pipeline where Fraps takes its measurement. Average frame rates look good. The frame rate over time seems pretty normal. And the consecutive frame latency number appears remarkably low. However, playing through the game—a critical component of the evaluation, in our minds—uncovers a lot of stuttering. We’re not sure if the 2 GB of GDDR5 available to each GPU isn’t enough, or if Nvidia’s metering technology (which it hasn’t spent much time talking about yet, but is already public domain) isn’t handling Far Cry 3 well. Whatever the case, we assure you that this game isn’t fun to play on a 690 at 2560x1600 and up.

http://www.tomshardware.com/reviews/geforce-gtx-titan-performance-review,3442-5.html

BTW, I also like Tom's graph:

- Another way to present it in a chart, labeling it as "consecutive frame latency"! Way to go - review sites jumping onto the bandwagon that TR pretty much built! Microstuttering was an issue for long enough.. now AMD and NV know that the media is more well-informed!

 

[BallaTheFeared]

Bro we're not talking about stutter right now.

We're talking about additional unrendered frames counting towards FPS ticks.

 

[BallaTheFeared]

Maybe this will help Bobo...

Fraps (Frames per Second)

Here is a typical frame rate over time graph as generated by FRAPS. Looks good right? CrossFire and SLI are competitive with the advantage to the HD 7970s.

Actual Output to screen

This is the same graph with data gathered from our method that omits RUNT frames that only represent pixels under a certain threshold (to be discussed later). Removing the tiny slivers gives us a "perceived frame rate" that differs quite a bit - CrossFire doesn't look faster than a single card.

 

[BoFox]

(^^ That's only with BF3) Yeah, that's why FRAPS's measurement of frame times isn't always an accurate indicator of frame latency measurements.

OTOH, Toms also said this about 7970GE regarding the frame time measurement by FRAPs:

Based on real-world gameplay, we’re not convinced that the Radeon card’s result is palpable, even if its lag is several times higher than what Nvidia’s cards convey.

I'm awaiting PCPer's further testing, along with additional testing by other sites. Hopefully AnandTech will look into this also. I wonder how much AMD would have fixed the drivers by then.. Oh well! :|

EDIT - What if PCPer also discovered the same thing about GTX 690 in Max Payne 3 and Dirt: Showdown (see post #45), where the "fast" 5ms alternating frames are actually more like 0ms skipped frames?!?