PC Upgrade (Motherboard & CPU)

Yeah, Superposition is almost purely GPU limited and CPU has a negligible impact on score.

Anyway, you didn't tell me what score you were getting, but if I recall what my own 1080 Ti's got, I would guess something in the ballpark of 9k.

I can crack 9k on my RX 6800 XT at ~130w total board power, and with my 6900 XT (because it's the same thing, but with more shaders, so doesn't need to be clocked as high) at about 120w. My RTX 3080 isn't far off, but has a higher power floor, thanks to relatively hungry GDDR6X.

My next GPU will probably be an RTX 4090, and if I had to hazard a guess based on current leaks, rumors, and speculation, there probably wouldn't be any point in targeting such a low score, but it could be done with sub-100w.

My score is just over 9000 yes, just ran the benchmark all the way through & got 9024 (67.5fps average).

I've been trying to grab a screenshot from GPU-Z with some useful numbers, this is about as good as I got:
7FNRclJ.gif

The problem of course is that it stops the benchmark when it loses focus.
 
I'm thinking a more useful comparison for power consumption might be to use the game with a capped framerate that others can replicate on their own hardware.

I'm sat docked up on deck on a carrier with framerate capped at 60fps:
5two1yk.jpg


Running the game at 4k (GPU at 55% load):
WIBjNhm.gif


And at 1080p (GPU at 33% load):
nvSArbH.gif



Board power draw is the 12th stat down from the top in GPU-Z. Hope this is useful if anyone wants to compare it to their own hardware :)
 
Did some testing in Superposition with my air cooled RTX 3080. Turns out the power floor for that part in that test is about 160w. That's at the minimum memory clock I can set and a core clock I know is completely stable at the lowest voltage (700mV) I can set in software. It also scores 10.5k in Superposition. I can clock it even lower, low enough for ~9k, but since I cannot reduce voltage, power only decreases linearly with clock speed, and none of the non-GPU core power goes down at all, so there is very little to show from further reduction.

My Navi21 parts do scale down to about the 130w range, partially because they are smaller GPUs with less power hungry memory, but also because I have considerably more control over them in software.

Anyway, once you hit a certain threshold, you need to actually trim away hardware to reduce power further, but if it's within the range the larger parts scale well at (a bit below half of stock power) then the larger part (of a given generation) is probably more efficient.
 
Did some testing in Superposition with my air cooled RTX 3080. Turns out the power floor for that part in that test is about 160w. That's at the minimum memory clock I can set and a core clock I know is completely stable at the lowest voltage (700mV) I can set in software. It also scores 10.5k in Superposition. I can clock it even lower, low enough for ~9k, but since I cannot reduce voltage, power only decreases linearly with clock speed, and none of the non-GPU core power goes down at all, so there is very little to show from further reduction.

My Navi21 parts do scale down to about the 130w range, partially because they are smaller GPUs with less power hungry memory, but also because I have considerably more control over them in software.

Anyway, once you hit a certain threshold, you need to actually trim away hardware to reduce power further, but if it's within the range the larger parts scale well at (a bit below half of stock power) then the larger part (of a given generation) is probably more efficient.

How about thinking a different way about this: Is there a modern, current card that is equivalent to my 1080ti (that we have all the power usage numbers for) that could produce equivalent performance but at a lower power consumption? I'm thinking maybe your line of reasoning is hampered by your having top-end cards.

You can see in the second GPU-Z screenshot in my post above that at 1080p my whole PC was doing not much more than idling, similar to what you describe trying to match my card running flat out.


As far as my next card is concerned the plan is to get a brand new 3080ti around the time the 4090 launches, which is what I did when I got my 1080ti. Not read up much about the 40-series, price will be the main factor on whether I get one though.
 
How about thinking a different way about this: Is there a modern, current card that is equivalent to my 1080ti (that we have all the power usage numbers for) that could produce equivalent performance but at a lower power consumption? I'm thinking maybe your line of reasoning is hampered by your having top-end cards.

Any modern card that matches the 1080 ti (an RTX 3060 ti or RX 6700 XT or better) does so at lower power consumption, and the power for these cards to deliver that performance falls as the tier of card increases (at 1080 ti performance, an RX 6800 will need less power than an RX 6700 XT), right up until the biggest biggest GPUs...which can't easily be set to 'only' 1080 ti performance without being below their sweet spot.

As far as my next card is concerned the plan is to get a brand new 3080ti around the time the 4090 launches, which is what I did when I got my 1080ti. Not read up much about the 40-series, price will be the main factor on whether I get one though.

Prices have fallen dramatically on current gen cards, and probably have a bit further to go, but are likely nearing bottom for non-secondhand parts. Shouldn't be a supply crunch any time soon, so it's probably safe to wait.

4000 series prices shouldn't be inflated as there are no supply side issues, but I'm still expecting to spend well over a thousand dollars for a RTX 4090.
 
One thing you should keep in mind
if you plan to go intel 12 series, you'd have to either make the move to Windows 11 or disable the e-cores
With an AMD you can keep using on Win10
It's not that a 12th gen Intel CPU won't work on Windows 10 at all.

It's that on Windows 10 it might fool the OS into putting "needs performance" work on the efficiency cores. So, yes, you could just disable them in BIOS, or you could do a little research and find how to pin important programs, like EliteDangerous64.exe, to just the P-cores.

Ref: https://ph.news.yahoo.com/intel-12t...s-10-11-streamers-multitasking-045701919.html
 
Anyway, I thought this thread would be a good read because I too have an i7-7700k and am looking to upgrade.

I've priced out an i5-12600k DDR4 system, not thinking that my needs require an i7 any more, given how many cores the i5 CPUs now have. I also priced out the same system, but with DDR5 (so changed motherboard and RAM), it's about £50 more for that. I might spec out a current-gen AM4 AMD system as well to compare, but really I'm waiting for the AM5 full announcement and third-party benchmarks, plus pricing to see what my choices really are.

I don't think anyone mentioned it yet, but for this sort of thing PC Part Picker is very useful. At least in the UK (and it has options for many other countries) it has all the suppliers you'd trust, plus some you might never have heard of. You can set price alerts per specific item, and look up historical price data on anything as well. UK site is https://uk.pcpartpicker.com/

And about a 'standard test' wrt Elite Dangerous. If you want to be sure about the CPU load then I'd go in Solo (because any other players can really change the result, especially if you arrive with someone else already running the instance) and dock in one standard type of orbital station.

Maybe just pick a specific one in Sol to make it easier for anyone not out in the black to use the exact same scenario.

Anyway, sit there for a good 10 minutes, compare both in the hangar and up on the pad. If you're going to use a Fleet Carrier than I'd use any system with literally only a single star, and thus the FC would be orbiting it, nothing else to complicate things.

For on-foot comparisons you can then just disembark there and go to the concourse. Surface on-foot would be trickier to find a standard test, because any given EDO settlement can end up shut down, with or without scavengers, on fire or not... too many variables.

Wouldn't it be great if Frontier just actually made a standard demo for us to test performance ?
 
For on-foot comparisons you can then just disembark there and go to the concourse. Surface on-foot would be trickier to find a standard test, because any given EDO settlement can end up shut down, with or without scavengers, on fire or not... too many variables.

Wouldn't it be great if Frontier just actually made a standard demo for us to test performance ?

The tutorials are repeatable, and some of them are pretty demanding on hardware.

About the only thing they are lacking is networking overhead (obviously) and situations with enough NPCs to be representative of a CZ when it comes to CPU load.
 
Hi everybody :)

Thanks for the recent video link you posted @Riverside (y) , I've bought / received the Thermalright frame and reviewing the recent video it seems it was the best choice compared to the others .;)
Question, to anybody really. I was looking at an ASRock Z690 Steel Legend, Intel Z690, S 1700, DDR4, SATA3, PCIe 5.0, 3x M.2, 2.5GbE, USB 3.2 Gen2, ATX motherboard yesterday on the web. Are ASRock motherboards about the same quality etc.as similar priced Asus / MSI / Gigabyte boards?
Any thoughts welcome. :)

Jack :)
 
Hi everybody :)

Thanks for the recent video link you posted @Riverside (y) , I've bought / received the Thermalright frame and reviewing the recent video it seems it was the best choice compared to the others .;)
Question, to anybody really. I was looking at an ASRock Z690 Steel Legend, Intel Z690, S 1700, DDR4, SATA3, PCIe 5.0, 3x M.2, 2.5GbE, USB 3.2 Gen2, ATX motherboard yesterday on the web. Are ASRock motherboards about the same quality etc.as similar priced Asus / MSI / Gigabyte boards?
Any thoughts welcome. :)

Jack :)

No opinion on the brand to offer but your spec list reminded me of something I learned about these z690 boards & their M.2 slots.

On mine the top M.2 slot is connected (optionally) directly to the CPU with 4x PCIe Gen 5 lanes, which is an important benefit for future NVMe drive speed potential. The others (mine has 4 total I think) are all PCIe gen 3 (via the z690 chipset) which is fine for current NVMe drive speed but you definitely want your boot drive in that gen 5 capable slot & check the bios box to have it communicate directly with the CPU. The top PCIe 16x slot still communicates directly with the CPU at full speed (16x gen 5) too, something I couldn't do with the z270 board I upgraded from.

The gen 5 lanes extra bandwidth for future (faster) drives is imo a worthwhile 'futureproofing' benefit, more so than gen 5 lanes for the GPU & was one of the reasons I went with intel over an AMD CPU & board.
 
DDR5 adds a bit to the cost, but it's worthwhile, especially if you intend to keep the platform a while.

I bought a DDR4 board principally to save money, as I wanted to be able to use my existing memory & that gave me a significant saving and I don't think memory performance was (or is) a bottleneck for me.

I agree that longer term DDR5 probably makes more sense if one has to buy memory at the same time as the motherboard/CPU anyway.
 
Yes, in my opinion. My previous gaming pc is an i5-2500k which is still running fine on an Asrock z68 pro3 gen3 motherboard. I built it in 2012 or 2013 and it still runs to this day.

My only issue with ASRock is that their firmware updates are sometimes delayed relative to other major brands. Hardware is almost always great and support is good. Had an issue with one of my boards not long ago and they made new beta firmware for me a day later.

Hard to judge reliability from only a few samples, so it's how support departments react to issues that I mostly pay attention to, in the absence of good statistics.

I bought a DDR4 board principally to save money, as I wanted to be able to use my existing memory & that gave me a significant saving and I don't think memory performance was (or is) a bottleneck for me.

I agree that longer term DDR5 probably makes more sense if one has to buy memory at the same time as the motherboard/CPU anyway.

There are a lot of games that are quite sensitive to memory performance and DDR5 usually has an appreciable edge over DDR4 in these titles on Alder Lake. I haven't don't any comparison between DDR4 and DDR5 in Odyssey, but given how it's responded to increasing main memory performance and adding cache on my other platfoms, I would be surprised if there was not an advantage. Some areas of Odyssey are CPU/memory limited at low enough frame rates that this could be more than an academic issue.
 
Not only did AMD announce their new AM5 platform last night:

  1. Source: https://www.youtube.com/watch?v=WcH_7xsYtUk
  2. https://wccftech.com/amd-ryzen-7000...ces-between-299-699-us-27th-september-launch/
with them going on sale 27th September...

... but it is now rumoured that Intel will be announcing their 13th generation CPUs on 27th September, with launch 20th October:


So I'll be holding fire on my upgrade until both those new lines are out with decent benchmarks/comparisons, and the pricing is clear.
 
Late 2022/early 2023 will be a good time for building a brand new PC. Just be aware that a new highend PC will not only be massively faster than everything we have right now, but will also require as much power as previously only tri or quad SLI systems used. I'll keep my current PC at least until Windows 10 support ends in 2025 however.
 
Back
Top Bottom