Tuesday, August 16, 2016

Upgrade plans for my desktop PC

The Small Form Factor PC that I built last year is quite fast for almost all the things I do with it. But I love upgrading my PC just to experience what the latest and the greatest hardware has to offer. Sadly, things do not usually live up to my expectations, but that is a different story.

This story is about the potential upgrade path this particular PC offers.

Sadly, my options aren't vivid. The PC is built in such a way that a tangible increase in performance requires a substantial amount of spending, which sometimes defies the reasoning behind "building" a desktop PC.

Let me explain.

Video card

Let's look at the video card first, because as you probably know, I am a gamer. The video card is the most important component in a gaming PC. But, I am not the typical gamer you would find elsewhere, but that is for another discussion.

Right now, I have an overclocked MSI GTX 970 Gaming 4G in my PC. To see a tangible increase in performance, I have to at least get a GTX 1070 and overclock it, which would give me 50% performance gains over my current card. But it would cost me close to $700 if I want to buy locally. $100 or so cheaper if I buy it from outside Australia, but it will make claiming for warranty painful. The GTX 1080 would give me twice the performance but it would cost me a whopping $1000 if bought locally! I am not going to spend that sort of money on a single piece of computer hardware, probably ever! But the GTX 1080 can be bought for less than $900 if I look to import it from outside.

I just have to see if the upcoming AMD Vega will be fast and affordable. But, why would AMD price it cheaply if it is fast? I probably might have to skip this generation altogether so that the next generation cards will give me a lot more performance for about the same price. For example, if the GTX 1170 is about 2.5 times faster than the GTX 970 for about the same price as the GTX 1070 today, then it would be worth holding out for it. Is it wishful thinking?

I sure would try to sell the GTX 970, but it would be difficult to demand more than $250 for it. The bottom line is, I will have to pay the deficit of $450 for 50% FPS boost, or $650 for 100% FPS boost. Economically, the GTX 1080 makes sense, but I feel either of these price tags are too much.


I have a 500GB Samsung 850 Evo SSD as my only drive in my PC. I think I made a mistake when I bought it in the regular 2.5” form factor. I should have gone with a M.2 drive instead. As my PC is very small, it would have made the internals of the PC much cleaner, because you don't need SATA data cables or power cables then. Even for a full ATX board, I recommend that you get a M.2 drive nowadays. I think it was the JPY 3,000 or so premium that put me off. I always look at the performance per $$$ and this always bite me back. Always!

While there are much faster M.2 drives out there in the market these days compared to then, they sadly cannot flex their muscles if plugged into the M.2 port on my Z97 motherboard since the M.2 port is wired to the PCI-E 2.0 x2 port. This limits the bandwidth to just 10Gbps or about 900Mbps, which is not a huge improvement over SATA. To get the best performance out of these new NVMe SSDs, you either have to own go Skylake on the consumer end or a Haswell-E or Broadwell-E on the prosumer end. The NVMe SSDs like the Samsung 950 Pro are compatible with my current board but there is no point paying a lot of money if I cannot get full money's worth.

That said, the gains these fast SSDs offer for the normal consumer is hardly noticeable, so it is not worth investing in them without knowing where the bottleneck is. The bottleneck I believe is in the NAND flash. Despite improving the sequential read/write speeds substantially, the 4K random/write speeds have not improved much. Perhaps we will see things moving a lot with Intel/Micron 3D XPoint technology. But if I'm building a PC today, I still won't go below a 950Pro.

The bottom line is, upgrading the SSD won't give me much of a speed gain.


These three components go in together unfortunately, because to change one, you need to change the other two as well. For example, to switch to a Core i7 6700K CPU from my Core i7 4790K, I would have to buy a motherboard based on the Intel Z170 chipset and along with DDR4 RAM. Since my CPU is pretty much the fastest CPU supported on my current platform, I cannot upgrade the CPU in a cost effective manner. There is no real reason just to upgrade the motherboard or RAM keeping the CPU where it is at, so I am stuck there as well.

The 6700K is not substantially faster to justify spending $300 or more on the upgrade, assuming I can sell my existing stuff. There is the option of upgrading to the HEDT platform too, but that would add another $250 to the total. Unfortunately, I would lose 10% performance most of the times because the HEDT platform is one generation behind the consumer platform. If all six cores can be utilitsed, then there will be about 30-35% performance gain (because of the 90% IPC loss), assuming you could overclock it to the same clock speed as an overclocked 6700K. The 6700K will generally overclock 200-300MHz higher, so the advantage will be even less. 25% would be a decent estimate.

The bottom line is, upgrading the CPU requires $300 or more, but it will only give me 10%-15% performance increase in CPU bound applications, like video encoding. It will not contribute to gaming at all, because my bottleneck is with the GPU. That's where my focus should be.

So, now what?

So, by all of the above, what I understood is that if I am to upgrade my PC, it has to be the GPU first. But there is one problem. My GPU choices are seriously limited because of the 267mm graphics card length limitation of the Silverstone SG13 case. Should I first upgrade the case then? Gosh, this ain't going to end!

No comments:

Post a Comment

Related Posts Plugin for WordPress, Blogger...