Silent, Green replacement GPU

[seveej]

(Yes, another "what GPU for me" -thread, but with a twist).

Rig: MP 5,1, 64 GB RAM, Radeon HD 5770 (Original)

I've been very happy with this machine and quite happy with this GPU (Reasonably silent, good ports, no hassles).

Quite happy, but not always happy: I don't play a lot, but when I do, I'm surprised by the amount of "OOMPH" even games like Civilization V seem to demand (Obviously, I have not yet dared go for CIV 6).

I honestly don't know whether the problem is in the 5770's relatively little memory, or in the raw processing power, so...

This has anyway started me thinking, I'd look into upgrading the GPU.
As I'm not a HC gamer, and as I'd really only need the added kick for less than 5% of my uptime, I find myself frustrated. While there are loads of GPU benchmarks, issues like power consumption and noise level are seldom addressed. Likewise, I can't really rationalise spending huge amounts of money on a GPU.


I'd very much appreciate your recommendations on any drag-and-drop, fully supported GPU (pre flashed is okay) replacement, with special consideration to noise and wattage.

RGDS,

 

[666sheep]

Someone has recently complaining about Civ performance with R9 280X against 5770. Check resources usage while playing, because it could turn out that your CPU is the bottleneck here at first place.

 

[ActionableMango]

I have or had an HD5770, GTX680, and GTX980. None of them were noticeably louder than the others to me.

The fastest fully supported and flashable cards would be the GTX680 and AMD R9 280x. (The RX460 is newer and appears to be fully supported in Sierra, but is a slower card than those other two).

The RX470 and 480 are showing promise in Sierra. I can't say they are fully supported because you have to disable SIP and do a KEXT hack (and probably repeat for updates), but they are very, very close.

 

[seveej]

Someone has recently complaining about Civ performance with R9 280X against 5770. Check resources usage while playing, because it could turn out that your CPU is the bottleneck here at first place.

CIV 5, large map, light video options (1440x1050, no AA, all selections either at low or minimum) 20 turns of established savegame to test:
CIV5 uses between 2,5 and 2,8 gigs of RAM. About 50 gigs free.
CPU utilization between 115% and 320% - so can utilize multiple cores, and (looking at Activity monitor's CPU utilization bars) seems to spread the load quite evenly. Seems though not to be hyperthreading, as load is only on the physical cores (1,3,5,7,9,11,13,15) not on the alternating virtual cores.

At no stage ever, is any resource utilization (visible in ActMon) close to peak. Memory there's plenty of, CPU never close to maxed out, Hard drive (SSD, SataII) bandwidth during gameplay never more than 10MB/s.

While the savegame is painfully slow to load (75 secs), the actual gameplay is tedium itself: recentering the map (clicking minimap) results in 10 seconds of "silent thinking", before it actually refocuses and starts drawing the map, which also takes some seconds (which in actual gameplay happens all the time as the view jumps from one unit/city to the next). The redrawing of the map seems to produce a tiny "bump" in CPU utilization, but never anything significant.

Running on the basic 5,1 Octo (2,4 Ghz), I'm quite sure the game would like a CPU speed boost, but what I fins most interesting is that going from Normal Game View to Strategic View does in no way alleviate the horrendous slowness...

RGDS,
P.S. Running in Native fullscreen (2560x1600) is not significantly slower, but buggy - after a couple of minutes, the game stops loading the visuals for the entire visible area ...

 

[h9826790]

CIV 5, large map, light video options (1440x1050, no AA, all selections either at low or minimum) 20 turns of established savegame to test:
CIV5 uses between 2,5 and 2,8 gigs of RAM. About 50 gigs free.
CPU utilization between 115% and 320% - so can utilize multiple cores, and (looking at Activity monitor's CPU utilization bars) seems to spread the load quite evenly. Seems though not to be hyperthreading, as load is only on the physical cores (1,3,5,7,9,11,13,15) not on the alternating virtual cores.

At no stage ever, is any resource utilization (visible in ActMon) close to peak. Memory there's plenty of, CPU never close to maxed out, Hard drive (SSD, SataII) bandwidth during gameplay never more than 10MB/s.

While the savegame is painfully slow to load (75 secs), the actual gameplay is tedium itself: recentering the map (clicking minimap) results in 10 seconds of "silent thinking", before it actually refocuses and starts drawing the map, which also takes some seconds (which in actual gameplay happens all the time as the view jumps from one unit/city to the next). The redrawing of the map seems to produce a tiny "bump" in CPU utilization, but never anything significant.

Running on the basic 5,1 Octo (2,4 Ghz), I'm quite sure the game would like a CPU speed boost, but what I fins most interesting is that going from Normal Game View to Strategic View does in no way alleviate the horrendous slowness...

RGDS,
P.S. Running in Native fullscreen (2560x1600) is not significantly slower, but buggy - after a couple of minutes, the game stops loading the visuals for the entire visible area ...

Is that just CIV 5 utilise up to 320% CPU? or total CPU usage up to 320%. It's so different. If that's total, may be CIV 5 can only use up to 100%, and the other 230% were from other background process.

Also, even it can use more than one core, it doesn't mean that all calculation can use more than one thread, the game may still suffer from single core performance limitation.

In my observation, when a single thread process demand 100% from the CPU, the CPU can distribute the load to different core at different time. They can never really working in parallel, but may make the monitoring software looks like the CPU is doing some multi thread calculations.

e.g.

At Time T=0, the CPU use core 0 to finish the 1st 20% of that single thread demand.
At Time T=1, the CPU use core 1 to finish the next 20% calculation.
At Time T=2, core 2 is in used.....
At Time T=3, core 3
At Time T=4, the CPU use core 4 to finish the demanded process.

However, the monitoring software may cover the whole period from T=0 to T=4. So, it display all core 0 to core 5 has 20% utilisation.

It sounds strange, but that's how it works

 

[Fl0r!an]

P.S. Running in Native fullscreen (2560x1600) is not significantly slower

This is a very good sign that you're CPU limited. When you're GPU bound, the performance should decrease almost linearly with resolution.

As @h9826790 has explained, one thread in your game could be running @ 100%, limiting your total performance. From my experience with OS X games & drivers this isn't unlikely at all.

 

[seveej]

Is that just CIV 5 utilise up to 320% CPU? or total CPU usage up to 320%. It's so different. If that's total, may be CIV 5 can only use up to 100%, and the other 230% were from other background process.

No, at max (especially when loading a save) Civ5's CPU utilization may hit 300+ %
During regular gameplay, it's around the 100%, going up to 150-160 when stressed...

Also, even it can use more than one core, it doesn't mean that all calculation can use more than one thread, the game may still suffer from single core performance limitation.

In my observation, when a single thread process demand 100% from the CPU, the CPU can distribute the load to different core at different time. They can never really working in parallel, but may make the monitoring software looks like the CPU is doing some multi thread calculations.

e.g.

At Time T=0, the CPU use core 0 to finish the 1st 20% of that single thread demand.
At Time T=1, the CPU use core 1 to finish the next 20% calculation.
At Time T=2, core 2 is in used.....
At Time T=3, core 3
At Time T=4, the CPU use core 4 to finish the demanded process.

However, the monitoring software may cover the whole period from T=0 to T=4. So, it display all core 0 to core 5 has 20% utilisation.

A-ha. That explains it., because that's (almost) exactly what the utilization may look like...

I guess you are also saying I should rather opt for a pair of X5672, X5679 or X5680 (even X5690) instead of upgrading the GPU.

EDIT: Looking at the Geekbench 64-bit Single core scores, my current CPU gives 2000 flat, while the toughest BTO processor (3,33 Ghz hexa) gives 2745 - a 37% difference. Only.
Quite puny compared to the Top-End iMacs' 4400+

So, in order to play games, i Need an iMac or a rMBP with a discrete GPU? Swell.

RGDS,

 

[ActionableMango]

That graph doesn't look CPU limited at all. The two cores with the largest load are only at 1/3.

 

[h9826790]

That graph doesn't look CPU limited at all. The two cores with the largest load are only at 1/3.

That graph can means CPU limiting. Please see my attachment. The test is programmed to use single thread only. And end up few cores working at low %. It was CPU limiting in my case.

https://forums.macrumors.com/threads/silent-green-replacement-gpu.2011731/#post-23853963
[doublepost=1478204336][/doublepost]

So, in order to play games, i Need an iMac or a rMBP with a discrete GPU? Swell.

A gaming PC or a console is actually a better choice.

 

[ActionableMango]

That graph can means CPU limiting. Please see my attachment. The test is programmed to use single thread only. And end up few cores working at low %. It was CPU limiting in my case.

https://forums.macrumors.com/threads/silent-green-replacement-gpu.2011731/#post-23853963
[doublepost=1478204336][/doublepost]

A gaming PC or a console is actually a better choice.

That's a good explanation. Is there a way to accurately tell if I'm CPU limited or not?

 

[Flint Ironstag]

So, in order to play games, i Need an iMac or a rMBP with a discrete GPU? Swell.

RGDS,

No, because in your first post you indicated that noise is a concern. iMacs and laptops will spin up their fans to annoying levels. It's usually 1 of 2 scenarios with those machines:

1. use SMC (or whatever flavor it is these days) fan control to crank your fans up to 100% while playing for best performance - which equals noise

2. let the mac regulate its temperature during game time, which often results in the fan noise varying - which is even more annoying and you may experience throttling.

Ha! This is one instance where a nMP might be good for you. The thing is silent. Get a 6 or 8 core with d700s if it's also a good fit with your day job.

 

[orph]

hay i play civ 5, it only ever uses 2-3 cores never more ie 200-300%
(most the time in game it seems to be around 190-200% and only 300% on loading maps i think)
im on a GTX660 (3.3ghz cpu) and have all settings set to max.

for civ 6 steam says your ok to run at low settings
http://store.steampowered.com/app/289070/

you can always boot in to windows

if you want a gpu a gtx 770 is worth a look (as steam recommends it) or if you can move on to osx10.12 then a RX 460 or RX 470