7870 XT vs EVGA 660 SC Signature 3GB

I recommend AMD to anyone looking to build an HTPC, NVidia to anyone looking to get into F@H.... For gaming, I recommend people to buy based on the games they own and which manufacturer those games favor.

AMD cards tend to struggle in WoW because it is a TWIMTBP game....same as every other game Blizzard produces.

 

nvidia the way its meant to be played

 

You guys have to remember its also luck of the draw.

You could get a 660ti that has a rubbish core and the same with the HD7870. So overclocking will go out the window and for a lot of people that is a big problem.

Also, AMD have come leaps and bounds with their drivers since they launched the never settle ones back in Novemeber. BF3 for the most part shot up by about 70% better frame rates on a lot of cards.

I will admit, I went from an overclocked GTX480 to my HD7970 and I was not that blown away by the performance. But since these new drivers have emerged and that fact that I have been lucky with my card being such a good overclocker it out performs most of Nvidia's offerings (bar the TITAN and the heavily overclocked 680's) but also cost about £50 cheaper as well.

 

well I done did it I am trading 2 nvidia cards for a powercolor vortex 7970? or something to that effect for my backup rig. going to try it in my main rig after testing in my back up rig to see what kind of cpu bottleneck is present. its a confirmed working card as I seen people with 7970/50 recently getting dud cards here

 

We're not saying that AMD cards are bad. We're simply saying that they don't provide the best overall experience in all situations....which is true of NVidia as well. There are cases where each card fails to provide the best possible experience....and games where each card excels.... It's just a matter of picking the card that will do so in the majority of conditions under which it'll be used.

I loved my HD7870....even with it struggling in WoW.....but the lack of support in F@H was a complete deal breaker for me and made my GTX660 a much better choice.

 

syk what do you use that 640 for fah and physx?

 

Nvidia for F@H? Have you seen what the GTX680 scores on Core_17? It scores like 8k something.

My 7970 gets 46K PPD at a modest overclock of 1125MHz.

Since the 680 released, I've been searching for clues on which would be faster for folding. It's a main part of my decision when picking a graphics card.

As soon as I knew the 680 had crippled compute, the decision was clear. That was almost a year ago btw.

Look at them both now, the 7970 scores a bit under 40k at STOCK. Nothing even comes close.

So no, no Nvidia for F@H please. Maybe Titan, yes, but until they solve the OpenCL performance, or feel like doing so, because, you know, they believe CUDA is the future, and do not pick up on trends well.


@all this thread: this isn't a war between whether AMD is better or Nvidia is better. OP is asking about two cards: 7870XT and GTX660. NO CF, NO SLI, etc...

I pick the 7870XT for multiple reasons, one is OC-ability (7950 / 7870 OC like crazy), and similar / worse / better performance (haven't looked at benches much TBH) than the GTX660 which shifts the decision to other factors.

The second, and I will repeat it again and again, the crippled memory bus of the GTX660 / 660Ti, which while it doesn't *seem* to hurt any performance now, there's no guarantee it won't later, and that "later" might be very soon.

I'm chill with AMD and Nvidia drivers. Both have sporadic issues, both work fine. High frametimes is more of an issue that came and will pass, and anyways it doesn't occur everywhere, and it's a work in progress. It shouldn't be an overblown issue, just like the GTX6xx VSync / stutter issues weren't overblown (simply because "Nvidia does not make any mistakes").

In that case, I'd look at the hardware, and the idea of paying the same or more for weaker hardware does not sit well with me, honestly.

EDIT: @sykozis: a 7850 @ 950MHz scores around 16300PPD with a member in here. Almost twice a GTX680. Just so you know.

 

How can you make a post about the topic being 7870xt vs 660 and then talk about folding numbers from a 7970 and 680 and a 7850?

 

Simply replying to Sykozis's suggestion. Do I need to make a new topic to reply to something he said?

Off-topic discussions happen; I'm simply stating this topic should not drift so far as to not answer the OP's question.

No need for smart replies.

 

not really, you are simply not taking your own advice about the topic being a 7870xt vs gtx660. off topic is fine.

 

As I said, you don't need to make a big deal about it. I mentioned my opinion regarding the topic at hand. In this case, I took my own advice.

 

I meant the XT version of the 7870. It's quite a bit faster. +if you OC it then there's no comparison. (It already trades blows with 660Ti).

This is what the OP should go for and what I promote. I'm no fanboy, although some things I say may seem as me taking sides. That is at all not what I mean to do. Most importantly I want the OP to get the most for his money's worth.

Look up some reviews and google a bit, and you'll see.

 

God I don't think I can read anymore threads like this the **** that gets spat out in these forums... lmao OP most people in these forums are bias one way or another. Read multiple reviews on both cards ( seems some websites are also bias for either red or green aswell) and decide what you want to get. Either way you will most likely be happy.

 

My GTX560ti was pulling 25k ppd on core_15 which is the most common for nvidia. My GTX660 pulls 20k on core_15. My HD7870 never managed to fold at all. I've never actually seen an nvidia card run core_17.... AMD cards also see a performance loss when running SMP+GPU that nvidia doesn't.

 

That used to be true on older cores. It was disappointing, as Pande Labs didn't seem to have quite gotten the OpenCL core right. The CUDA-based core was still the better performer.

Now with Core_17 (Beta), things are the opposite. The Radeons are using 1-2% CPU, and the GTX680 (Kepler) are using a full CPU core.

Check out this spreadsheet, see what I mean:
https://docs.google.com/spreadsheet/ccc?key=0Aspl6-RkaxtFdDZfZHI5SGpIYVFYbG1EUVpZTm5oOUE#gid=4

@40: GTX680 @1128MHz, ~6875PPD, CPU usage for FAHCore_17: 12.5%
7970: 2,10,11,12,16,23,32: CPU usage 0-2%, ~46k PPD @1125MHz.
7870: 17: SMP 10 on

Surely a great boost in folding power for Pande Labs as all those running 7950s, 7870s, 7950s, and 7970s can now chip in.

 

OpenCL runs on both CPU and GPU, which is why AMD cards use CPU time and suffer a performance loss when running SMP+GPU. With CUDA, the CPU is only responsible for the graphics driver, which is why there's minimal performance penalty.

BTW, I can't verify your claim of Kepler cards using a full CPU core.... So, how exactly did AMD get around OpenCL running on both CPU and GPU? If OpenCL is implemented properly, it's going to use CPU time.

According to F@H records, an HD7970 only turns out approx. 42K ppd on current projects and has to be quite heavily OC'd to exceed that... A GTX680 at stock clocks is putting out 42-44K ppd on current projects....

I found that statement in the F@H section rather funny since you just claimed it doesn't happy....

The GT640 was removed for testing Lucid MVP....but when it's installed it generally runs F@H 24/7 as it doesn't incur any performance penalties for the system at all. I can still game and benchmark without any performance loss....and still turn in WU's in F@H, albeit rather slowly.

 

What F@H records? Can you link me please? Or do you mean the spreadsheet?

If you're talking about the spreadsheet:


7970@1050MHz (13.51% OC over 7970, 5% OC over 7970GE): 46305PPD, 1% CPU usage on FX-8150 @4.5GHz.


7970@1090MHz (17.84% OC over 7970, 9% OC over 7970GE): 43608PPD, 0-1% CPU usage on i7 3930K @4.3GHz.
7970@1090MHz (17.84% OC over 7970, 9% OC over 7970GE): 44923PPD, 0-1% CPU usage on i7 3930K @4.3GHz.
7970@1100MHz (18.92% OC over 7970, 10% OC over 7970GE): 47266PPD, 0.6% CPU usage on E5-2680 @3.1GHz.


7970@1000MHz (stock 7970GE): 22197PPD, 0-2% CPU usage on i7 970 @3.6GHz. Weird result here.


7970@1125MHz (21.62% OC over 7970, 12.5% OC over 7970GE): 48,772PPD, 0.6% CPU usage on E5-2680 @3.1GHz.


7970@1125MHz (21.62% OC over 7970, 12.5% OC over 7970GE): 49,821PPD, 0.6% CPU usage on E5-2680 @3.1GHz.


7970@1050MHz (13.51% OC over 7970, 5% OC over 7970GE): 42,767PPD, CPU make and usage not listed.


29847PPD with i7 3770 @3.9GHz SMP 7. CPU usage for FAHCore_17, 12.5%.


30780PPD with i7 3770 @3.9GHz SMP 7. CPU usage for FAHCore_17, 12.5%. Misread the 6875 for PPD last time, sorry.

Almost matching scores for the last two results.


None of the 7970s are OCed heavily. The max clockspeed here, 1125MHz, is a clockspeed that can be reached by the vast majority of 7970s at STOCK voltage. Check the 7970 launch reviews, reviewers cranked the slider to 1125MHz on Overdrive and left it there before unofficial overclocking / non-reference cards were available.

The rigs with 680s were scoring ~30k PPD with SMP on.

Here are the results for my rig: i7 930 @3.61GHz + HD7970 Core_17:

First, 7970 @ 925/1375 @1175mV (stock):

Spoiler

Second, 7970 @1125/1375 @1175mV (stock voltage):

Spoiler

Heavily OCed? nope. 47k > 42k? yep.

Third, SMP7 + Core_17:

Showing HD7970 PPD:

Spoiler

Showing SMP7 PPD:

Spoiler

Just now, took a snip of Task Manager showing CPU usage for Core_17:

Spoiler

1%.


For comparison, here's a friend's system with a GTX670 at stock (isn't OCed), i5 3570:

Spoiler

22769PPD. CPU usage is 26.5% (~25%, 1 core).

It was fluctuating from around that number to 30.6% or so. Why? Read the following.

Did I say anything about the card using ZERO percent CPU??

I said it uses 1-2%, and I can verify this on my setup. So yes, OpenCL is using CPU time.

Obviously, having a 12% load from SMP and a 2% load from Core_17 would worsen performance on both, albeit slightly. On Core_16, if I had SMP8 + GPU running, the GPU usage would hover around 60%. Once I turned off SMP it rose back to 95-99%. I used to run SMP7 + GPU for similar results, the GPU usage would be maintained at 95-99%. Now even with SMP8, it is maintained at 90%+.

The following replies might be even funnier to you:

And that explains why I've been seeing >25% CPU usage from Core_17 on that GTX670 under Windows 8.

 

Default clock for HD7970 is only 925mhz. 1125mhz is more than 20% OC, which most consider a heavy OC.

First you claim GTX680 only gets 8k PPD on core_17 (a beta core)....then you claim a GTX670 gets 22k PPD....so, which is it?

You're basing all your information on a "beta" core....which means unoptimized and bugged. Also means more likely to return errors. Most of us that are folding, are running stable cores, not betas because you get no points if errors are returned. So, how about we stick to current projects and cores that are officially supported and stable? Beta performance means nothing. So, my post stands as factually correct as all current (read: official) projects are done on Core_15 and Core_16. To get Core_17, you have to use the -beta flag, according to the nice people running the Folding@Home project.