Upgrade from 780 SLI to 980Ti SLI

[ddogg]

Hey guys, I currently have two Superclocked 780s in SLI and was wondering if it was worth making the upgrade to 2 980Tis or wait another year for Pascal. I am gaming at 4k and besides a few good PC titles like the Witcher 3 and GTA V, I need to drop settings to get constant playable fps(30+).
Not sure if I should wait for Pascal or get the 980s as a stop-gap solution. I know the 980Ti SLI will comfortably play 4k at close to 60fps. My system is also completely custom water-cooled so I really don't want to break my loop.

My other system specs are i5-2500k @4.5, 16gb Ram, 2x500GB Samsung 850 EVO Raid-0

Thanks

 

[happy medium]

I think 2 980ti's will be better than a stop gap solution, especially overclocked.

I also think your 2500k will not fully push 2 high end Pascal gpu's. One Pascal wont be much of a upgrade over 2 980ti's.

I think you should grab 2 gtx980ti's now, and do a full system overhaul with 2 Pascal gpu's later.

gtx980ti's are on sale for 599$ with a free game.
http://www.newegg.com/Product/Produ...4500378&cm_re=gtx980ti-_-14-500-378-_-Product

You still should get good money for your 780's also.

Does your motherboard have pcie-3.0 16 x 16?

 

[Silverforce11]

Do NOT pay such ridiculous $$ for 2x 980Ti when Pascal mid-range will drop in early 2016 that will destroy it in performance for less money and less power use.

Think about the 2 x 580 vs the 2x 680 comparison, because we're dealing with a full node jump and a new architecture.

There's also the lack of wisdom in buying an NV architecture that is about to be made obsolete.

The reason your dual big-Kepler runs like donkey in Witcher 3 is because NV has moved their focus on optimizations to Maxwell. Take a look at other NV sponsored titles and compare how well Kepler runs it (hint: not very well compared to neutral titles).

When Pascal drop you can expect the same optimization focus shift to it instead of Maxwell.

You are basically in the position of early 2014, you have the choice of upgrading to 2 x 780Ti or wait ~half a year for a new-gen to arrive (but even better with Pascal, new-gen on new-node). If you had splurged $1500 or more for 2x watercooled 780Ti, you would feel really stupid once Maxwell lands.

So, hold onto your 2x 780 for ~6 months, that's my guess when we'll see mid-range Pascal mop the floor with current top-end stuff. Then you can upgrade to Pascal knowing it will be receiving all the optimization focus for the next ~2 years (or whenever Volta is due).

 

[alcoholbob]

Transistor cost has not gone down at 16FF though...a 16FF card might be more cost effective because better architecture or HBM, but it won't be because its on 16FF manufacturing node...

 

[Silverforce11]

Transistor cost has not gone down at 16FF though...a 16FF card might be more cost effective because better architecture or HBM, but it won't be because its on 16FF manufacturing node...

680 small mid-range chip = faster than big chip 580 at the same price.

New node + new uarch does that.

Mid-range Pascal should be faster than 980Ti.

 

[alcoholbob]

Pricing remains to be seen. Essentially what you are claiming is Nvidia and AMD will be forced to reduce their profit margins when 16nm FF hits by possibly 50%. We'll see if competition alone will do that. So far it hasn't been shown as both companies are pricing big die at $650 on the very cheap and mature 28nm.

I believe they will end up meeting in the middle. Nvidia and AMD will take a cut in gross margins and increase their prices as well.

 

[Mondozei]

I would say it's worth it. Next year, Pascal will likely be GP104, so if you're on GK110 now and upgrading to GM200 this year, would mean that you could go to GP100 at HBM2 in two years time. You get the benefits of the "full fat" architecture every time you upgrade.

I would add something, OP. You probably need to do a CPU upgrade.

 

[ddogg]

I think 2 980ti's will be better than a stop gap solution, especially overclocked.

I also think your 2500k will not fully push 2 high end Pascal gpu's. One Pascal wont be much of a upgrade over 2 980ti's.

I think you should grab 2 gtx980ti's now, and do a full system overhaul with 2 Pascal gpu's later.

gtx980ti's are on sale for 599$ with a free game.
http://www.newegg.com/Product/Produ...4500378&cm_re=gtx980ti-_-14-500-378-_-Product

You still should get good money for your 780's also.

Does your motherboard have pcie-3.0 16 x 16?

I think it has PCI-e 2.0, I would need to look it up online. x16 on 1 and x8 on the other.

I would say it's worth it. Next year, Pascal will likely be GP104, so if you're on GK110 now and upgrading to GM200 this year, would mean that you could go to GP100 at HBM2 in two years time. You get the benefits of the "full fat" architecture every time you upgrade.

I would add something, OP. You probably need to do a CPU upgrade.

Definitely. I am looking to upgrade my Mobo and CPU, possibly a i7-6820k (if one does come out) next year. I don't do much beyond gaming on this machine so haven't seen the need to upgrade but the 2500k is definitely showing it's age now.

 

[ddogg]

Do NOT pay such ridiculous $$ for 2x 980Ti when Pascal mid-range will drop in early 2016 that will destroy it in performance for less money and less power use.

Think about the 2 x 580 vs the 2x 680 comparison, because we're dealing with a full node jump and a new architecture.

There's also the lack of wisdom in buying an NV architecture that is about to be made obsolete.

The reason your dual big-Kepler runs like donkey in Witcher 3 is because NV has moved their focus on optimizations to Maxwell. Take a look at other NV sponsored titles and compare how well Kepler runs it (hint: not very well compared to neutral titles).

When Pascal drop you can expect the same optimization focus shift to it instead of Maxwell.

You are basically in the position of early 2014, you have the choice of upgrading to 2 x 780Ti or wait ~half a year for a new-gen to arrive (but even better with Pascal, new-gen on new-node). If you had splurged $1500 or more for 2x watercooled 780Ti, you would feel really stupid once Maxwell lands.

So, hold onto your 2x 780 for ~6 months, that's my guess when we'll see mid-range Pascal mop the floor with current top-end stuff. Then you can upgrade to Pascal knowing it will be receiving all the optimization focus for the next ~2 years (or whenever Volta is due).

This is why I am unsure whether to make the jump or wait. If Pascal is only 6 months away, it makes sense to wait but if it releases next year in Q4 it would translate to a long wait. Also, Nvidia is not known for great pricing so I may be looking at a $1000 card unless AMD also releases something good.

 

[Insomniator]

I really doubt 'pascal midrange' is going to destroy 980TI's in performance in early 2016.

If you are gaming at 4k, and can sell your 780's for ~400, 980TI would quite an upgrade. I do wonder if that 2500k will keep up though. You can probably find 4k/SLI benchmarks with CPU's compared.

 

[RussianSensation]

I think it has PCI-e 2.0, I would need to look it up online. x16 on 1 and x8 on the other.

The higher the resolution/AA, the lower the FPS, the lower the demand on PCIe bandwidth for a graphics card.

"Contrary to intuition, the driving factor for PCI-Express bus width and speed for most games is the framerate, not resolution, and our benchmarks conclusively show that the performance difference between PCIe configurations shrinks at higher resolutions. This is because the bus transfers a fairly constant amount of scene and texture data for each frame. The final rendered image never moves across the bus except in render engines that do post-processing on the CPU, which has gotten much more common since we last looked at PCIe scaling. Yet the reduction in FPS due to a higher resolution is still bigger than the increase in pixel data even then." ~ Source

When a lot of PC gamers upgrade to a newer PCIe 3.0 x16 platform from PCIe 1.1 x16 or 2.0 x8, they see a big performance increase and think their mobo was holding them back. In almost all cases, what was holding them back was the slower CPU/IPC and lack of Hyper-threading/sufficient CPU cores.

In practice, that means for a Fury X/980Ti level card, PCIe 2.0 x8/x8 SLI/CF at 4K would only be a 3-4% performance hit.

Source

Definitely. I am looking to upgrade my Mobo and CPU, possibly a i7-6820k (if one does come out) next year. I don't do much beyond gaming on this machine so haven't seen the need to upgrade but the 2500k is definitely showing it's age now.

From what I've read, 6820K is just Broadwell-E and Skylake-E won't be out until 2017.

The Witcher 3 and GTA V are extremely demanding at 4K. If you want to max all settings and AA, a single 980Ti isn't sufficient.

Unfortunately, a single 980Ti is also not a very good upgrade for the games you listed over 780 SLI.

or GTA V where 980Ti ~ Titan X

I guess it depends if you want to upgrade now or wait until you do a full overhaul to i7 6700K/6820K + Pascal SLI.

Also, keep in mind that while at 1080P, you would be somewhat CPU limited with a 2500K @ 4.5Ghz, at 4K, you are almost 100% GPU limited. Overclocked i5 2500K would be at least as fast as a 4670K here and there is no way dual 980Ti SLi would be hitting those minimum FPS at 4K.

 

[Seba]

If you can look at a pair of GTX 980 Ti and consider it as a stop gap solution, I'll say go for it.

 

[RussianSensation]

A couple more benchmarks at 4K for TW3 and GTA V from TechSpot.

It's really up to you if you want to upgrade now or play at reduced settings and do a full system overhaul. If I were in your shoes I am not sure I'd go for 980Tis right now because their prices have barely fallen since June 1, 2015. I would feel like why didn't I just upgrade 5 months ago? Maybe at this point just wait for Pascal and do a full system rebuild in 1H of 2016. I will say though that if you have the $, then go for it since we still don't have a concrete date as to when Pascal will launch in 2016. You are kinda in between GPU generations and towards the end of the Maxwell generation so it's tough to say if this is a good time to upgrade to dual 980Tis.

Pricing remains to be seen. Essentially what you are claiming is Nvidia and AMD will be forced to reduce their profit margins when 16nm FF hits by possibly 50%. We'll see if competition alone will do that. So far it hasn't been shown as both companies are pricing big die at $650 on the very cheap and mature 28nm.

I believe they will end up meeting in the middle. Nvidia and AMD will take a cut in gross margins and increase their prices as well.

That's not his claim at all. Notice he said that GTX680 outperformed 580. GTX680 was a mid-range card priced at $500 which meant NV's profit margins were still very high. At 2560x1600 4XMSAA, on day 1 680 beat 580 by 35%.

The move from 560Ti to GTX680 at 1920x1080 8xMSAA shows a 94% increase in performance (new node + new architecture).

If we look at 4K benchmarks at TPU, and apply a conservative 1.70X increase rather than 1.94X increase to the 980 as was the case with 560Ti ->980, it still ends up easily outperforming the GTX980Ti

4K
980 @ 64 x 1.70 = 109%
980Ti sits at 80%
http://www.techpowerup.com/reviews/MSI/GTX_980_Ti_Lightning/23.html

If Pascal delivers 70% increase instead of the 94-100% increase of Fermi->Kepler, the successor to the 980 should still be 30-35% faster than the 980Ti at 4K. NV could price 980's successor at $549-599, make $$$ hand over fist too since it would be selling mid-range Pascal for $550+

 

[ddogg]

Nice info there. I actually completed the witcher 3 and GTA 5 at 4k. I had to keep most settings at high, a few at medium and hairworks off and got decent playable fps (35-40). GTA 5 was medium settings and a few at high. Good thing is in 4k you can turn off AA and nothing looks jagged so it was an enjoyable experience. I don't mind turning settings down a little but with each new title release it seems that 4k is getting more demanding on my rig.

 

[RussianSensation]

Nice info there. I actually completed the witcher 3 and GTA 5 at 4k. I had to keep most settings at high, a few at medium and hairworks off and got decent playable fps (35-40). GTA 5 was medium settings and a few at high. Good thing is in 4k you can turn off AA and nothing looks jagged so it was an enjoyable experience. I don't mind turning settings down a little but with each new title release it seems that 4k is getting more demanding on my rig.

I still remember $700 780Ti's performance level migrating to a $330 970 SKU in just 10 months. Since 5 months already passed from 980Ti's launch and since Pascal = 16nm node + new architecture + faster GDDR5X or HBM, imho, everything is aligned for Pascal to deliver one of the biggest increases in perf/watt and price/performance. Maxwell's 970 made 780Ti look like money sunk but that's nowhere near as dramatic as to what Kepler did to Fermi.

GTX580 (last gen's 40nm flagship) < new 28nm node + new architecture 660Ti/670/680/780/780Ti. Even the GTX660 was almost as fast as the GTX580!

780Ti was 2X faster than the 580.

While there are no guarantees that we'll have a Big Pascal in 2016, it's absolutely guaranteed that the best $650 card in 2016 will be much faster than the 980Ti, have better feature set, more VRAM and most importantly focus on driver optimization.

NV is going to want gamers to upgrade. Focusing only on perf/watt won't be enough (i.e., 980Ti's performance in a 165W TDP won't be enough of a selling point to spend $600 on such a product). That means it's logical that NV's Pascal will either bring way better performance than 980Ti at similar prices and/or it will bring 980Ti's level of performance at lower prices. This is how the GPU industry has worked for decades and there is little reason to believe Pascal/Arctic Islands will be different.

 

[ddogg]

I snagged my 780s when the Ti came out. They dropped from 650 new to about 400-450 used
Looks like I may just have to suck it up for a bit longer at medium settings or game on my 2nd monitor which is 1440p and do a complete overhaul when Big fat Pascal or Arctic Islands comes out.

 

[happy medium]

game on my 2nd monitor which is 1440p

Second monitor?

Oh ok , good for you, stick with the 780's then.

 

[Timmah!]

If you can look at a pair of GTX 980 Ti and consider it as a stop gap solution, I'll say go for it.

oh well, i envy the people who can look at pair of high-end GPUs as a stop-gap solution :-D

anyway, if you intend to get Pascal next year, dont buy those 980Ti´s.

 

[Timmah!]

If Pascal delivers 70% increase instead of the 94-100% increase of Fermi->Kepler, the successor to the 980 should still be 30-35% faster than the 980Ti at 4K. NV could price 980's successor at $549-599, make $$$ hand over fist too since it would be selling mid-range Pascal for $550+

I still remember $700 780Ti's performance level migrating to a $330 970 SKU in just 10 months. Since 5 months already passed from 980Ti's launch and since Pascal = 16nm node + new architecture + faster GDDR5X or HBM, imho, everything is aligned for Pascal to deliver one of the biggest increases in perf/watt and price/performance. Maxwell's 970 made 780Ti look like money sunk but that's nowhere near as dramatic as to what Kepler did to Fermi.

GTX580 (last gen's 40nm flagship) < new 28nm node + new architecture 660Ti/670/680/780/780Ti. Even the GTX660 was almost as fast as the GTX580!

780Ti was 2X faster than the 580.

While there are no guarantees that we'll have a Big Pascal in 2016, it's absolutely guaranteed that the best $650 card in 2016 will be much faster than the 980Ti, have better feature set, more VRAM and most importantly focus on driver optimization.

Agreed about everything else, but 30-35 percent perf increase is hardly "much faster". Especially when its probably half year later at best. Now if it was supposed/expected to be 2x faster (which in the end may as well be, who knows), then by all means...that is definitely worth the wait. 1/3 not so much IMO.

 

[tential]

oh well, i envy the people who can look at pair of high-end GPUs as a stop-gap solution :-D

anyway, if you intend to get Pascal next year, dont buy those 980Ti´s.

Why? As long as you have significant savings, buy the 980tis, sell at a 50 dollar loss on each card. You pay far more on the phone bill each year. So 100 for a year worth of 980ti sli isn't bad at all.

 

[Silverforce11]

I really doubt 'pascal midrange' is going to destroy 980TI's in performance in early 2016.

Why would you doubt that? History has proven that to occur at almost every node transition.

On top of that, this is a uarch revolution as well.

It's like the 480/580 jump to the 680. The 680 was a mid-range small chip, it was faster than big Fermi, used much less power and cost similar.

The same thing happened for AMD at the last node jump, the 7870 was a very small chip, it end up faster than GTX 580 and AMD's 6970.

Node jump + new uarch = big time performance leap.

That's why I said anyone who wants to folk out over $1500 (close to $2000 for water cooled setups) for 2x 980Ti now is nuts when next-gen mid-range will make these current stuff so obsolete.

These chips also were reported to taped out awhile ago, they are also reported to not use HBM2 as that's reserve for the big-chips, so there should be no issue putting a small chip with GDDR5/X to market towards the end of H1 2016.

Re the pricing, I just don't see NV or AMD trying to sell a mid-range chip for $1K... but let's not give them ideas! :0

 

[tweakboy]

Looks like 980 it is .....but in spring watch out

 

[Timmah!]

Why? As long as you have significant savings, buy the 980tis, sell at a 50 dollar loss on each card. You pay far more on the phone bill each year. So 100 for a year worth of 980ti sli isn't bad at all.

Cause its wasting 1500 EUROs on things, which will be obsolete maybe half year from now (if his intention is to get Pascal then). Maybe he has money to burn, but its still waste of them.

And its not like he is on some low-end GPU right now. He has 780 SLI ffs. Those used to be top of the chain not so long ago.

 

[ddogg]

oh well, i envy the people who can look at pair of high-end GPUs as a stop-gap solution :-D

anyway, if you intend to get Pascal next year, dont buy those 980Ti´s.

Haha - I don't really upgrade all that often. CPU, mobo every 4 years or so. GPUs every 2nd generation or so. So when I do, I tend to go all out and get the best available (not including stupid purchases like the TitanX or Z). Only reason I was considering this is because my cards do struggle unless I drop settings. I'll just wait like people have advised and get the top-end when they do release sometime next year.

 

[thesmokingman]

Imo, it's better to invest or plunk down cash at the leading edge than at the trailing edge. Buying tis now will most definitely be at the trailing edge. I'd hate to be staring at the posts of everyone buying into 16nm FF and HBM2 while you're stuck on yesterdays news.