7 May 2020

Untitled hardware post...


Since there's not much going on with tech at the moment (although I do have a short post in the works regarding RDNA 2 rumoured die sizes) I thought I'd shift gears to a general hardware post.

Mass market VR is just around the corner...


First up there's the news that the number of VR users has significantly increased in the wake of Half Life:Alyx. I think this is great and I'd love to play the game to experience it but really the cost of VR is still too high (and quality long-form games too sparse) for mass market appeal.

I recently upgraded my PC in preparation for the next 10 years (I'll get onto that below), and it wasn't cheap for a pretty low-end gaming system - approximately €640. The Oculus Rift S costs around €400 and the Valve Index costs around €550... and those are just for the headsets, not even thinking of the specialised controllers that actually help to bring VR games to life.

My €640 system probably cannot run VR games at an acceptable level of performance but aside from that we then get into a discussion of what "acceptable performance" really means.

However, there was one enlightening comment on that news post at Ars Technica that prompted me to write this portion of my post and it's a refrain I've heard before:
It's just that $1000 is overstating it a bit. An entry level headset is around $200, and you can play a great deal of VR games on a 580. If you don't have a gaming machine at all then jumping into PC gaming with a VR setup will be costly, but for an average person with an mid range GPU then the jump to VR would be a few hundred. Pricy still I know, $200 isn't nothing, but not out of reach of many people.
Maybe my standards are higher but I don't agree with this argument. Of course, this directly relates to personal thresholds relating to that discussion about what is acceptable performance.

For me, it's like saying that you can buy a GTX 1050 and play any game at 1080p without issue - you don't need to spend €200-300 on a graphics card. Technically, the argument is correct. You can do that... but what sort of experience would you expect and what underlying system do you have to support that piece of hardware? And how do those relate to your performance thresholds?

I don't think that a current choice of cheap headset is the way to go for VR as the reviews I've seen have all said that it's better to go for the more expensive and fully-fledged implementations (not a cheap $200 headset) and I'm not really convinced that an RX 580/GTX 1060 will be sufficient for good quality experiences, especially going forward. People are already complaing about the earlier headsets' quality (especially PSVR and the first Oculus devices) with respect to their DPI and refresh rates... or their discomfort. Could I really recommend such a low-end experience?

PC Gamer recommends an RTX 2070 or RX 5700 XT at high performance in Half Life Alyx... Valve only states minimum specs - an R5 1600 combined with an RX 580/GTX1060!

I actually have quite a high tolerance for dropped frames, stuttering and popping-in of scene details since I primarily game on a console and my PC has always squarely been mid-range and I would not accept the performance drop to the level provided by a GTX 1050! So I definitely wouldn't accept a low-end performance in VR where drops in performance will not only affect enjoyment but also potentially induce physical discomfort, if not illness!

This is the crux of the problem - VR requires more performance for an acceptable level than a static screen does. 60 fps on a monitor is perfectly fine, allowing the RX 5600 XT/GTX 1660 to play Half Life:Alyx without issue. However, many people feel that 60 fps just doesn't cut it and that 90 fps is required for a truly smooth and comfortable experience. That puts this tentpole game experience in the realm of the RX 5700 XT and RTX 2070 and maybe the RX 5700/RTX 2060, though those will come with a lot of dips in fps.

Moving back to the comment above, yes, you can just about scrape together a gaming PC for €500 but I would not suggest it or promote it. Realistically, you would want to spend €750-1000 in order to get a setup that is worth the money. Cutting back a bit, you can put together a decent build but is it VR-ready? 

A mid-range current CPU (R5 3600), and low-ish end GPU (RX 5500 XT or GTX 1660) together cost around €400 depending on the deals you can get, a 512 GB (ideally NVMe) SSD for boot and a not terrible case will cost another €70-ish* and 16 GB of 3200 Hz DDR4 will set you back around €70* as will a not-expensive motherboard. Then you need a PSU and various sundries (you sometimes need to buy cable extensions or adapters) will cost around €60-100 and a windows licence can be had for €20-100 depending on what and where you buy from.
*It's important to note that prices have increased beyond these figures over the last few months and I expect them to keep rising until the end of the year...
This build is assuming you already have a monitor, adequate keyboard and mouse and isn't including shipping and is at the low end of those estimated ranges above... and it's already €690. Realistically, this build is not possible except in the most extremely favourable circumstances (e.g. free shipping and deep discounts).

While this build would be more than adequate for 1080p gaming and ripe for a GPU upgrade in a couple of years time, it isn't going to be overclocking much because the motherboard and PSU are probably at voltage and wattage capacity at those prices. Space on that SSD will also quickly fill up with games, applications and general files. Looking back up at those Half Life:Alyx requirements, it's also not going to give a great VR experience. So, again, I disagree with this sentiment - this is a very capable low-mid range PC but you still could do with spending more on a better GPU in the first place and that's going to be pushing the total up to €850, minimum.

Half Life:Alyx, requires a decent CPU (while Valve lists the R5 1600 as minimum spec), hardware reviewers prefer the 1600 AF or R5 2600 as the minimum CPU required. At this price point, you're only dropping around €30-50 from the cost of an R5 3600 (and this was before the current issues really began) so not much of a saving, really. An R5 2600 + 2060 Super, as that site recommends for an average performance, would set you back around €500. This just isn't viable for the average gamer that's looking to upgrade their PC as well as get into VR gaming.

VR's proponents, by and large, are people with either 1) a large amount of disposable income or 2) people with an already above-average PC build available to them. Spending as much money on a VR setup as you would for a low-to-mid-level build just shows how disconnected these people are from the average gamer's situation.


Flash data reads are incomparable with DDR or GDDR rates...

RAM on GPUs...


With the released specs of the two new consoles from Microsoft and SONY I've seen a few commentators* speculate about the future of PC gaming - specifically gaming hardware. The scuttlebutt is that the fast SSDs and unified memory of the consoles will bring about a new paradigm in hardware design on the PC in order to "keep up" with the consoles.
*I forgot to save the youtube videos and links but if you look a bit you might find some speculation.
The two speculative trends that I've seen floated were NAND flash installed on consumer-grade GPUs (like the datacentre GPUs that have been released by AMD) and games requiring 32 GB standard as the system DDR RAM instead of 8-16 GB as we currently have to contend with.

Needless to say, I am skeptical of both of these potentials.

First off, even before the current price increases on NAND and RAM and the pinch on supplies a 256 GB SSD was not cheap... at least not by a percentage comparison with a graphics card. And by "not cheap" I'm of course referring to actually decent, PCIe Gen 3 or Gen 4 NVMe devices - not a SATA II device. My best SATA II SSD is able to achieve a 0.50 GB/s transfer speed whereas my Gen 3 NVMe SSD can manage 2.02 GB/s read speeds. It sounds fast but my 16 GB of RAM can manage 37 GB/s... which, again, is nothing compared to GDDR5 or 6 transfer rates.

Having to replace an SSD (or two) over the course of a graphics card's lifetime is quite an expense and doesn't really add anything better than what we have available in the three-tier SSD/system RAM/VRAM paradigm. 

This brings me to my second point: How many games will be loading more than 13.5 GB of data for a game to read (the Series X's theoretical maximum stored in the GDDR6)?

I can answer that fully with a big, fat, none!

Worse than that, the consoles won't even be loading in that much data per frametime, just a fraction of it. The current setup in PCs is more than fast enough for this. If anything, mid-to-high-range 6-10 GB VRAM graphics cards on PC already have more memory performance available to them than the APUs in the next gen consoles when combined with 8 GB system RAM and a 7200 rpm HDD.

Realistically, 16 GB system RAM is still going to be the maximum required for games over the next generation but I do expect the amount of VRAM on graphics cards to stabilise at 8 GB - even down to the low-end cards. While there have been leaks pairing the RTX 3070 with 8 GB RAM, I would be surprised if the RTX 3060 would come with less than that because 6 GB just doesn't cut it anymore as far as I'm concerned - even if you're going to be running at 1080p or 1440p.


I'm actually really happy with these results... despite messing up the boost clock on the CPU. D'oh!


The build...


So moving onto my strange build. 

I have had my prior PC for 10 years now but with some upgrades along the way. I've spoken about it before but let's just list the specs so you can appreciate the level of improvement over what came before:
  • Phenom II X4 955 BE, 4 C/T @ 3.2 GHz
  • ASUS motherboard M4A785TD-M EVO
  • 16 GB DDR3 2133 MHz RAM
  • 256 GB Samsung 850 Evo SSD boot drive
  • GTX 1060 6 GB
  • ATX 650W PSU
  • Unknown manufacturer case
I was pretty happy with this build because it ran pretty much everything I wanted at 1080p. However, I began to experiment with video encoding and found that the poor CPU was just not up to the task - it lacks modern silicon that provides hardware support for various instruction sets which also impacts the abiliuty to run certain games.  

So it was time to upgrade... but it's a bad time to upgrade with prices of certain items creeping up due to the one-two gut punches of covid-19 and a console launch year. Coupled to this are several impending BIG hardware launches for PC (namely Navi 2.0, the RTX 30 series and both Intel's and AMDs' next generation of chips and chipsets), all of which are promising large improvements in performance (hopefully per unit currency, too!).

I couldn't wait another 6-10 months so I did what every person in that sort of situation does and compromised:
  • Ryzen 3 3200G 4 C/T @ 3.6/4.2 GHz base/boost
  • MSI B450-A Pro MAX
  • 16 GB DDR4 3200 MHz RAM
  • 500 GB Western Digital SN_750 w/ heatsink SSD boot drive
  • GTX 1060 6 GB 
  • 600W Corsair SF600 PSU 
  • NZXT 510 case
Now, I know what you're thinking, "this build is pretty low-end" and a bit random. However, I have a plan! I needed to upgrade but I wanted to be as prudent with my money as I could be... and that meant NOT updating right now. So, and maybe I'm crazy here, but I spent my money on the support hardware and skimped on the main items - the opposite of what people usually recommend.

It's been 10 years since I've built in a new case and the amount of optimisations and quality of life improvements are staggering...
That translated to buying the cheapest CPU I could (that I felt comfortable buying as an interim upgrade) and re-using the GPU from my old PC (that one inherited a fanless GT 210 testing card I had lying around) in anticipation of buying a much better GPU and CPU some time next year when discounts are available and the maximum performance can be obtained for the AM4 socket and raytracing capability for the cheapest price.

It's true, I could have gone with something even less performant - e.g. the Athlon 3000G, but I still wanted to be able to utilise this PC in the interim before the upgrade I'm actually planning for. At the time I ordered the parts, the Ryzen 3 3100 and 3300X were not even rumoured to be coming so I wasn't even planning on waiting for those and, to be fair, I would have been spending more money for not much more performance. In my opinion, the 3300X is priced too closely to the 3600, making it pretty redundant and the 3100 costs 25% more than the 3200G that I ended up buying while having only a little more headroom for overclocking (though I couldn't even make this comparison at the time).

My reasoning for picking the 3200G also centred around the fact that it is a part that's designed with quite a lot of thermal overhead due to requiring heat dissipation from the iGPU cores. I figured that this would allow better and more stable overclocking if that was switched off, despite the Zen+ parts not being super overclockable. Plus, I figured it would have a higher resale value than the Athlon 3000G for when the time comes to upgrade to (hopefully) an R5 4600, pending a BIOS update from MSI.
UPDATE: And of course, within an hour or so of me posting this, AMD released an announcement stating that the B450 chipset will only officially support Ryzen 3000 and B550 would support Ryzen 4000.
If I'm honest, I'm quite disappointed if it's true as from the beginning of the year, rumours have been bubbling to the surface of B450 support for next gen Ryzens. It would mean that, officially, both B450 and B550 only support a single processor stack... regardless of AMD's position that the AM4 socket is multigenerational, it's the board that those sockets sit on that reallz matters. I just have to hope even more that a BIOS update is made available for my board. If not, I will wait for a cheap R7 3700X or 3800X to upgrade in a year or so. I'd be losing around 15% performance (according to estimates) but not that big of a deal. 
First up, I saw what the CPU was capable of at stock. I found that the natural tendency of the CPU was to hover around 3.7 GHz when under load - which had me slightly worried considering the boost clock is meant to be "up to" 4.2 GHz. I thought I'd lost the silicon lottery! But a little playing around  with Ryzen Master showed that I could easily attain an all-core 4.1 GHz clock. After a bit of stress testing, I went into the BIOS and set it up.

Unfortunately, you can see that I must have forgotten to change boost frequencies because when I was then testing back in windows, I saw that my boost clock was below my base clock. At any rate, I didn't have a chance to fix this issue or try higher base clocks as yet - but I am hopeful I can reach higher base clocks than the maximum, out of the box, single core boost frequency.

The GPU was a different story, I had to be much more choosy with how much extra voltage I was providing the card and how much extra frequency I was jumping by. Eventually, I settled on a small overclock to 2100 MHz on the core and 8401 MHz on the memory (GDDR5 on the GTX 1060). I reckon I can push it further but I reallz haven't had time to sit down and properly test it's all working as it should as when I benchmark, the GPU happily sits at 2100 MHz but as soon as I entered the game I was testing, it dropped to 1400 MHz - lower than stock. So I think there's something going on there... If you have any suggestions, let me know! I'm pretty new to overclocking in general but am especially not confident on overclocking GPUs.

And that's what I was looking for...

So coming back, full circle, to the primary reason for this upgrade happening right here and now - H.265 encoding of game footage - this upgrade is a significant success and timesaver. I no longer need to have the PC running 24 hours a day in order to keep up with footage generation and, quite frankly, the gaming performance isn't terrible either. I was averaging 74 fps in the Total War: Warhammer benchmark.

I suppose people will ask "Why not NVENC?". Well, that's because I need both high compression in order to get smaller file sizes. A 4 GB file compresses to around 300-500 MB with H.265 but with H.265 NVEC I was only saving around half of the original file size (i.e. ~2 GB). Sure, the encoding times were around 20 mins but I wasn't getting the file size I wanted.


Conclusion...


There is so much going on in tech at the moment that it's difficult to keep up. It's even harder to commit to a given build with releases pegged for the near future and stealth releases happening left and right. I really think that prices are going to be quite competitive for next generation CPUs and GPUs and, even if they're not, the prices of the current items will drop a bit to something more palatable to my budget. :)

If you've built a system recently, post it in the comments. Also, PSA, PCGamer are running a competition to win an RTX 2080 Ti decorated in Cyberpunk 2077 livery. They've also had a nice friendly system builder competition (for bragging rights) at $500, $1000 and $2000 budget levels (as usual, only considering the tower and no software in the budget). I sometimes get in the mood to see what other people can come up with and these are fun exercises, despite alot of people leaning into the same sorts of parts...

No comments: