CPU update, GPU update or both??

xaephod · Mar 24, 2015

Always looking for the update that will last me 3 years (Previous builds were FX-60, then Q6600) and now I5-2500K with a 7970 ghz. I'm a gamer and Mainly play Planetside 2. The rig gets 60fps with everything maxed. I wonder, will I get smoother gameplay from an update? I realize that even though I get 60fps, there is probably some fluctuation while playing which can lead to choppiness.

Will an update make the gameplay more seamless? If so, Which to update or all? If yes, to what? Remember, I look for that sweetspot 2-3 year update cycle. Wait for that next chip from Intel? Skylake?

Thanks!

p.s. How about another 7970 ghz for crossfire?

Kevin_Smith · Mar 25, 2015

xaephod :

If you had a 120hz monitor or 144 one, you could probably get a lot smoother gameplay experience when you reach above 60 fps. I'm sure a gtx 970 and an i5 of this generation, will be enough to last you a few years.

Icaraeus · Mar 25, 2015

Even a Titan X will only last until mid-late 2016 before it'll be massively outdated.

paulbatzing · Mar 25, 2015

Icaraeus :

For playing Planetside 2? Obviously there will be faster GPUs each year, but playing on 1080p and not going crazy on the AA the OPs GPU should be fine... Any GPU bought this year will be fine for a while.

To the OP: Do you experience "choppiness", or is this just something you expect? Have you run some benchmarker while you play, to see how far the FPS actually dips? Your GPU is on par with the 280X (see http://www.tomshardware.com/reviews/gaming-graphics-card-review,3107-7.html ), and looking at benchmarks (e.g. http://forums.videocardz.com/topic/188-planetside-2-performance-test/ ) the game saturates well below your configuration, so for that game alone, I would not update.

Generally, if you want a noticable increase, you should not go below at HD7990 or Titan X to get at least a 30% increase in performance, but the first of these is an (old) multi GPU sollution (which scales differently, depending on CFX support), and the second is 1000$. I would wait at least to the next generation if i were you. Getting a second 7970 ghz might be good, but then you need to find one, and they have been out of production for a while.

When it comes to your CPU, there is even less a point in upgrading as long as it is only for gaming.

Icaraeus · Mar 25, 2015

What's so special about Planetside 2? It's an unoptimized game. Next year GPUs will be between 4-10x faster than current tech, not to mention there'll be stacked VRAM, DDR4, HBM, NVLink, PCI-E 4.0 and other things.

paulbatzing · Mar 25, 2015

Icaraeus :

He asked specifically if he should buy new to increase his performance in Planetside 2, and his configuration should not have ANY problems at 1080p no matter what he does. So if the question is "should I update to get a better performance in planetside 2" the answer is a clear no.

I doubt a 10x increase in performance from the next generation. Look at past node changes, it will probably be more like 2x increase (and that is optimistic) in the budget and enthusiast class cards (GTX1060 and GTX1080 and the like, whatever they will be called), and a modest 1.8x increase for the most expensive cards (GTX1080ti or TITAN Y or whatever they will call them). If it is more I will be pleased, but VERY suprised.

From the things you mention: NVLink will not be implementet in consumer class cards next year (or even in the near future), it is a feature designed for many-gpu super computers. PCI-E 3.0 is not saturated by any GPu as of now, so PCI-E4.0 will not yield any increase for games. HBM and stacked VRAM are the same thing, but they only yield much improvement in scenarios where you are memory speed limited, which is not a big problem in 1080p. And DDR4 will only give you an improvement if you buy memory that is actually faster then the DDR3 you use today, which is not available today. So none of these technologies will massively improve gaming performance for anybody.

I do agree with you that right now is not the time to buy a GPU if one can wait, and the OP clearly can, but the difference will not be as big as you state.

Icaraeus · Mar 25, 2015

Ten times faster in FP32/64 scientific computations. NVLink is actually designed for Nvidia Pascal and the new CPU architecture. HBM and stacked VRAM are not the same thing. HBM is a new memory standard that has up to 2TB/s bandwidth, while stacked VRAM allows what it says it is. Pascal is designed for ultra HD and 5K. DDR4 is 2x faster than DDR3 dependant on the stick.

paulbatzing · Mar 25, 2015

Icaraeus :

Well, you are quoting Nvidias markelting numbers for per-core performance increase on a pascal card. They will certainly not release a consumer card for 600$ GPU with 10 times the performance of the GTX980. Just to compare: the GTX 980 (~4600 GFLOPS) has about 5 times the performance of the GTX 280 (~900 GFLOPS), which are 6 generations apart. If there really will be a 10x increase in ppc on the board, then they will probably release a board with 1/5th of the number of cores or something like that. If only to compensate for low yield of the production in a new node. So ok, maybe the per core performance will increase this much, the gaming performance of consumer cards will not.

For NVlink to work, you will need a CPU and a motherboard that supports it. As of now, there is no indication that intel has any plans in abandoning PCIE and supporting a proprietary link for Nvidia only cards. And if this happens, it will fragment the mainboard market so you have to choose either a NVLink motherboard, or a PCIE one. They certainly have partners that will provide these features, but I serisously doubt that any consumer CPU and motherboard will ever have this tech. The future is more likely in some unannounced standard that is GPU producer agnostic, but that is pure speculation.

HBM stacked memory is, as you say, a new standard. It is also the stacked VRam standard that will be used for GPUs in the forseable future, so what other stacked VRam standard you are talking about, I don't know. Please enlighten me.

While there are the first DDR4 sticks that are faster then DDR3, I still think that there are few scenarios where increased system memory speed has any impact on game performance. You would have to be heavily CPU limited, and that just is not the case with any modern CPU today in most games. Maybe that will change with directX12, but I doubt the change is that big. Just look at this article: http://www.tomshardware.com/reviews/memory-bandwidth-latency-gaming,3409.html , the conclusion holds true as of today...

Icaraeus · Mar 25, 2015

10x faster at FP32/64 calculations, not gaming performance. There will surely be NVLink capable motherboards to support Pascal and Intel generally doesn't talk about their new CPUs much until a couple months before release. Pascal is late 2016. HBM is 9x faster than GDDR5 and will be able to be stacked.

paulbatzing · Mar 25, 2015

Icaraeus :

Yes, 10x more Floating point operations per second per core in average is what Nvidia claim. Which I doubt. And, as I tell you, even if that is the case, there will not be any consumer card that will perform 40000 GFLOPS in sale for 600$ next year. There has been a 5x increase in number of floating point operations in the last 8 generations of Nvidia gpus, and there will not be that big of an increase now.

Well, if you think we will see socketed GPUs next year, good for you. I seriously doubt that. See http://www.anandtech.com/show/7900/nvidia-updates-gpu-roadmap-unveils-pascal-architecture-for-2016

To pull off the kind of transfer rates NVIDIA wants to accomplish, the traditional PCI/PCIe style edge connector is no good; if nothing else the lengths that can be supported by such a fast bus are too short. So NVLink will be ditching the slot in favor of what NVIDIA is labeling a mezzanine connector, the type of connector typically used to sandwich multiple PCBs together (think GTX 295). We haven’t seen the connector yet, but it goes without saying that this requires a major change in motherboard designs for the boards that will support NVLink. The upside of this however is that with this change and the use of a true point-to-point bus, what NVIDIA is proposing is for all practical purposes a socketed GPU, just with the memory and power delivery circuitry on the GPU instead of on the motherboard.

NVIDIA’s Pascal test vehicle is one such example of what a card would look like. We cannot see the connector itself, but the basic idea is that it will lay down on a motherboard parallel to the board (instead of perpendicular like PCIe slots), with each Pascal card connected to the board through the NVLink mezzanine connector. Besides reducing trace lengths, this has the added benefit of allowing such GPUs to be cooled with CPU-style cooling methods (we’re talking about servers here, not desktops) in a space efficient manner. How many NVLink mezzanine connectors available would of course depend on how many the motherboard design calls for, which in turn will depend on how much space is available.

[...]

With all of that said, while NVIDIA has grand plans for NVLink, it’s also clear that PCIe isn’t going to be completely replaced anytime soon on a large scale. NVIDIA will still support PCIe – in fact the blocks can talk PCIe or NVLink – and even in NVLink setups there are certain command and control communiques that must be sent through PCIe rather than NVLink. In other words, PCIe will still be supported across NVIDIA's product lines, with NVLink existing as a high performance alternative for the appropriate product lines. The best case scenario for NVLink right now is that it takes hold in servers, while workstations and consumers would continue to use PCIe as they do today.

(emphasis added by me) So as I said, feel free too believe whatever you want, but there is zero indication that NVlink will be coming to anything consumer-related anytime soon.

I do not doubt that HBM is faster, I just don't place as much emphasis on VRam speed as you do. VRam speed has not been a limiting factor in 1080p gaming for a long time, and I doubt it will be anytime soon. I just don't think this will make much of a difference for real life performance.

All of this is very off topic, as we both seem to agree that the OP should not replace his GPU now, so maybe it is fine to shut this discussion down now?

geetings
Paul

Icaraeus · Mar 25, 2015

Pascal is not designed for 1080p though, it's designed for ultra HD and 5K. Bandwidth becomes a limiting factor in high resolutions.

paulbatzing · Mar 25, 2015

Icaraeus :

Just as Kepler was designed for 4k, which doesn't mean that the cards actually delivered a good gaming experience at 4k. Pascal will probably have the possibility to connect screens at up to 8k (although as of today there is no connector to use 8k at 60 hz, so either it will be a multi DP sollution, or a new standard will come along before that), but that does not mean games will be playable in 8k. But yeah, the high end pascal cards will probably stay in the 60 fps range on 4k, so you can expect something a bit better than the titan x on next generations x80 card. Just don't expect 10x the titan x, that will not happen.

xaephod · Mar 25, 2015

Great discussion!
So do you think upgrading to a 4K monitor would be the start of the upgrade path?
I figure the current GPU cannot support 4K seemlessly. Is 4K that much more awesome for gaming? Never seen a live display. For now, I'm at 1920x1200 on a 28" monitor.

Icaraeus · Mar 25, 2015

Titan X can do 4K but it's very expensive and will be outdated next year. Unless you can afford both 4k monitor and Titan X then 4k isn't really worth it.

paulbatzing · Mar 26, 2015

xaephod :

Not the way to go right now. Next generation might offer a reasonable GPU for 4k gaming, but I doubt it will be below 650$, and even then it will not deliver 60fps consitently. I think 4k for entusiasts is still 2-3 years away, and the improvement is not so great anyway. If you feel that you need a higher resolution, rather go with 1440p at 144 Hz, then you get the low latency and higher frequency, and if the screen is not to big, you also get a much better PPI then you have today. It comes at a price though.

If I were you, I would wait to upgrade until at least next fall. If you do not want to wait that long, wait at least until the summer, then there will be a new ATI GPU (380x if I am not mistaken) and probably a 980ti (Titan X performance for 200$ less), so prices on the rest of the market should go down. But as I said, if you want more power, CFX is not so shabby either, and your GPU is still quite good.

Icaraeus · Mar 26, 2015

How do you know it will not deliver consistent framerate? Pascal is the biggest jump in hardware in years. It's best to wait until Pascal.

paulbatzing · Mar 26, 2015

Icaraeus :

We have discussed this at length, so let's just say we disagree. Nvidias marketing is certainly not the same as a benchmark, so I will believe the leap in performance when I see it. My prognosis: A gtx 1070 ~ 10% slower then the titan X, a gtx 1080~30% faster then the titan X, then a couple of month later a Titan Y (or something) with ~50%-70% better performance than the titan X. The titan X is still not consistently over 60 fps in all games, so there is no reason to expect the next generation to be with next years games. They might be, and then I will be delighted, but all of this is pure speculation, unless you have a pascal card to test at home.

Just to clarify: Pascal will most certainly deliver consistent framerates, I just said I do not belive we will have them consistently above 60 fps next generation in 4k. 4k is the same as driving 4 1080p displays at the same time, and as long as there is no card that can deliver 240 fps consistently, faster memory does not help (slow memory acts as a bottleneck, fast memory will relieve the bottleneck, but not make the image calculation faster).

Anyway, OP: Waiting is a good strategy for you, especially with such a new card.

CPU update, GPU update or both??

Distinguished

Honorable

Honorable

Reputable

Honorable

Reputable

Honorable

Reputable

Honorable

Reputable

Honorable

Reputable

Distinguished

Honorable

Reputable

Honorable

Reputable

Share this page