Question Recommend me a low to mid level graphics card for "light" gaming (new or used)

[Lil'John]

I am doing older games like Battletech with Roguetech upgrade(ie memory hog), Shadowrun Returns, and Warhammer 2. I have ZERO interest in FPS games or multiplayer type games. First half of it is kind of a lie since I've broken into Mechwarrior 5

My system is a bit on the lower end: I am using an Intel 12100F with 32GB ram. My graphics card is a used to me Asus 1050ti.

I'm kind of brand agnostic for chips but I can't recall that I've ever gone AMD. Same thing with graphics card assembler but I've mostly done ASUS.

So what should I be looking for in the new or used market? I'm considering a budget of ~$250... I might stretch it to $300.

 

[The red spirit]

First of all, electromigration is a function of time, temperature, and power usage. In mining, the smart miners try to hit their efficiency "sweet spot", which means lower temps and power levels than gamers, who often do the opposite, and overclock.

Perfectly aware of that, but miner unlike gamer still keeps GPUs running for way longer than gamer and it's not liek their gains are tremendous, they still run at 60s or so, therefore time trumps anything that is saved by undervolting and underclocking. Not to mention that this is ideal case scenario, many don't do anything (probably more than half) or just tighten vRAM timings. Therefore you can only get either heavily used card or nearly failing card. Either way borderline e-waste.

BTW you can read about why chips fail here:

https://en.wikipedia.org/wiki/Reliability_(semiconductor)

https://en.wikipedia.org/wiki/Transistor_aging

Also desktop chips are only made to work so long:

https://en.wikipedia.org/wiki/High-temperature_operating_life

Basically 5 years at maximum 50C ambient temperature. But they don't expect you to hammer that hardware 24/7 for 5 years straight.

Meanwhile, the rest of components on card are still subject to all of these:

https://en.wikipedia.org/wiki/Failure_of_electronic_components

And yeah, temperature, voltage, current don't do it any favors, but more than anything else, time.

Remember that "New World" game that was burning up 3090s? Mining certainly wasn't doing that, or you would have heard about it in the news instead.

Now tell me how many people used those cards for BOINC. I frankly smell your bias here. Also it was faulty design from eVGA (maybe some others too) and any higher load would have killed them anyway. It's just that more people game with them, than doing some GPGPU work.

Second, you mention "ball joints", which I assume that you are referring to the BGA solder bumps used to affix the GPU package to the PCB Assy. Thermal stress is what damages those. Which gaming sees a lot of, due to the fact that the GPU loads is highly variable during gaming, but mostly constant (and thus lower thermal stress, less heating / cooling cycles) for mining.

Variability isn't necessarily a killer. Macs from 2009 died often, because poor material for GPUs was chosen (by nVidia). Also gaming isn't exactly all that variable load and productivity tasks can also be variating too (also mining, becasue nicehash software tends to have a lot of 0-100% load task, when it switches algos, which it often does or when searching for work). Anyway, this is still a problem as recently as with R9 290X. I think it was the latest GPU to have this particular type of failure after prolonged usage. I would say that gamers' 4 hours in game is better for card's longevity rather than 24/7 lower variating (arguably) load

 

[VirtualLarry]

And yeah, temperature, voltage, current don't do it any favors, but more than anything else, time.

I strongly disagree with your premise that time, and not voltage or temps, is the dominant factor in electromigration phenomena. If that were true, then why do semiconductor mfg's have a TEMP-dependent lifespan de-rating curve/chart for component? Also, ever researched "HOT-e effect" WRT electromigration.

Furthermore, I assert that time is the LEAST dominant effect for electromigration and thus shortened component lifespan, by the number and age of legacy PCs that are still out there.

 

[The red spirit]

Most electronics have a bathtub curve for failure rate. If it was being used for mining for a year it's pretty likely it would have failed during that time. Once they get beyond that earlier failure period the likelihood of failure goes down considerably until it gets extremely old. Posters here have all kinds of old tech that still runs fine despite being over a decade old and well past the useful lifespan of the hardware because the performance isn't up to today's standards.

Survivorship bias here, nothing more. Also mining load is 24/7, gaming is 4 hours per day. Miner in year keeps GPU on for more than 6 times longer (mostly because hardly any of us game for 4 hours straight everyday for 365 days). 2 years of mining and it maybe borderline e-waste.

 

[Lil'John]

If you run it on a mid-end to high-end 4K TV, they can usually upscale 1080p to 4K without discernible quality loss. That way, you could turn your settings to max and reduce the load on your GPU.

I don't have the specific model but I am running an Acer UHD 4K2K monitor for a few years.

 

[The red spirit]

I strongly disagree with your premise that time, and not voltage or temps

Which makes no sense, because you still use voltage and still are above ambient temperature anyway. Sure it takes longer for time to do its job, but it's a time that's the killer in the end. Not like you can keep using GPU in 0 kelvin environment with 0 volts.

If that were true, then why do semiconductor mfg's have a TEMP-dependent lifespan de-rating curve/chart for component? Also, ever researched "HOT-e effect" WRT electromigration.

Because time isn't the only variable, but it's important one. You extend life of GPU by undervolting, perhaps by factor of 2 and it doesn't matter a single bit, because you use it 6 times longer (24/7 mining vs 4 hours per day gaming or more realistically more like 2-3 hours of gaming and certainly not everyday, making 6 times assertion very optimistic).

Furthermore, I assert that time is the LEAST dominant effect for electromigration and thus shortened component lifespan, by the number and age of legacy PCs that are still out there.

lol but did they mine coins for years? If they did, they would be already dead and scrapped.

BTW this is MTTF of semiconductors, called Black's formula (https://en.wikipedia.org/wiki/Black's_equation):

A is a constant
j is the current density
n is a model parameter
Q is the activation energy
k is Boltzmann's constant
T is the absolute temperature in K

Temperature is just one variable, time too, but temperature remains just one part of this wear out model, meanwhile time affects almost every parameter.

 

[In2Photos]

I strongly disagree with your premise that time, and not voltage or temps, is the dominant factor in electromigration phenomena. If that were true, then why do semiconductor mfg's have a TEMP-dependent lifespan de-rating curve/chart for component? Also, ever researched "HOT-e effect" WRT electromigration.

Furthermore, I assert that time is the LEAST dominant effect for electromigration and thus shortened component lifespan, by the number and age of legacy PCs that are still out there.

I agree with Larry. If heat and voltage were not the dominant factor in chip life then why would data centers have such large cooling requirements? Why not let them run hotter and save on energy costs?

PS. Typing this while my i7-920 from 2009 has been running nearly 24/7 since it was built.

 

[VirtualLarry]

Survivorship bias here, nothing more. Also mining load is 24/7, gaming is 4 hours per day. Miner in year keeps GPU on for more than 6 times longer (mostly because hardly any of us game for 4 hours straight everyday for 365 days). 2 years of mining and it maybe borderline e-waste.

There are mining rigs with R9 390, RX 470/480/570/580 (Polaris), as well as NVidia Turing (10-series) that have been mining since day-1 purchases, which are still running great (or were, until ETH stopped being minable).

Like I said, miners consider their cards like chickens that produce golden eggs.
So they treat they well. THey don't run them into the ground, if they know enough not to.

 

[The red spirit]

There are mining rigs with R9 390, RX 470/480/570/580 (Polaris), as well as NVidia Turing (10-series) that have been mining since day-1 purchases, which are still running great (or were, until ETH stopped being minable).

Like I said, miners consider their cards like chickens that produce golden eggs.
So they treat they well. THey don't run them into the ground, if they know enough not to.

And now I wonder how many of those cards are already replaced. No way none of them died, absolutely no way. Particularly no way that r9 390s didn't die early, they were particularly finicky even stock.

 

[igor_kavinski]

I don't have the specific model but I am running an Acer UHD 4K2K monitor for a few years.

LG 4K TVs use an Nvidia GPU for upscaling. At least, my C8 does.

 

[VirtualLarry]

LG TVs use an Nvidia GPU for upscaling.

Source?

 

[igor_kavinski]

Source?

I just checked my "About this TV". Sadly, it seems Nvidia's copyright notice was removed in the last update. Maybe they were using some proprietary Nvidia tech and switched to opensource?

I believe LG had a short love affair with the Tegra SoC and maybe some form of it they licensed for upscaling?

 

[Mopetar]

Survivorship bias here, nothing more. Also mining load is 24/7, gaming is 4 hours per day. Miner in year keeps GPU on for more than 6 times longer (mostly because hardly any of us game for 4 hours straight everyday for 365 days). 2 years of mining and it maybe borderline e-waste.

It's only survivorship bias in that miners who've had a card for a year have found most of the cards with dud components or other problems that were going to fail or require an RMA regardless of who bought them. Unless you're worried about someone selling you a card they know is defective, I wouldn't be worried.

Two years of mining for something like ETH isn't going to kill a card. It's operating under a relatively lighter load than almost anything else you'd do with it. There are people here that run their CPUs 24/7 and don't have problems with them because they aren't cranking up the voltages or otherwise running their components outside of normal operating parameters. The same silicon that goes into consumer GPUs is going into data center products that are also running 24/7 and often under heavier loads. Those aren't failing horribly as a result. Why do you think cards used for mining are categorically different?

 

[The red spirit]

Why do you think cards used for mining are categorically different?

Much higher chances of buying something that will work for a while and then die. The rest I explained before.

 

[maddie]

BTW this is MTTF of semiconductors, called Black's formula (https://en.wikipedia.org/wiki/Black's_equation):

A is a constant
j is the current density
n is a model parameter
Q is the activation energy
k is Boltzmann's constant
T is the absolute temperature in K

Temperature is just one variable, time too, but temperature remains just one part of this wear out model, meanwhile time affects almost every parameter.

This part intrigues me. Just not seeing it. Where is time part of every parameter?

Not A
Not j
Not n
Not Q
Not k
Not T

What am I missing?

 

[The red spirit]

This part intrigues me. Just not seeing it. Where is time part of every parameter?

Not A
Not j
Not n
Not Q
Not k
Not T

What am I missing?

First, let's start with MTTF. It means Mean Time To Failure, so here's your first clue. Basically this whole formula should tell you how long it takes to reach failure, therefore, time affects every single thing here.
Then, let's dig deeper. Notice how you first have material properties and geometry (A) divided by current density. More current density in same silicon will shorten it's lifespan, but besides that, the worst thing is that exp part, meaning exponent, which means material properties of silicon and current density will lead to failure by factor of exponent. Now Q is activation energy, in semiconductors meaning voltage (not threshold voltage, but maximum power state voltage, meaning voltage required for achieving high clock speed). And then you divide it by Boltzman's constant multiplied by temperature in Kelvins (which barely variates as 0C is around 270K, idling GPU is at around 310-320K and at full load it may be at 340K-380K), meaning that voltage itself and alone will affect lifespan of semiconductor much more than temperature ever will and in some amount of time. I won't calculate it, because I'm really awful at physics myself, but it's pretty obvious too, undervolted card will only get like 10-20% voltage reduction and perhaps loss of 10 kelvins, which is like 2% reduction. So even considering the fact that it's an exponent, you may double the lifespan of your HW, but if you keep it on for 6 times longer, then it just means have to have a seriously aggressive undervolt and underclock to make your card's wear and tear similar to gamer's, which is not going to happen, because performance is crucial and you just end up with card that are at best (meaning that user cared to undervolt and keep them well cooled) are like 2 years old, but already are as worn out as 6 year old cards and that gets seriously worse very quickly. If person never undervolted them and kept them cool, card that is 2 years old will already have been worn out like 12 year old card. That's a big problem, because absolute minimum MTTF for desktop hardware is 5 years. Now, that MTTF number is probably calculated in 24/7 type of usage, because predicting average usage is way too subjective, but still, my point that you may buy way too worn out hardware for its age stands.

 

[maddie]

First, let's start with MTTF. It means Mean Time To Failure, so here's your first clue. Basically this whole formula should tell you how long it takes to reach failure, therefore, time affects every single thing here.
Then, let's dig deeper. Notice how you first have material properties and geometry (A) divided by current density. More current density in same silicon will shorten it's lifespan, but besides that, the worst thing is that exp part, meaning exponent, which means material properties of silicon and current density will lead to failure by factor of exponent. Now Q is activation energy, in semiconductors meaning voltage (not threshold voltage, but maximum power state voltage, meaning voltage required for achieving high clock speed). And then you divide it by Boltzman's constant multiplied by temperature in Kelvins (which barely variates as 0C is around 270K, idling GPU is at around 310-320K and at full load it may be at 340K-380K), meaning that voltage itself and alone will affect lifespan of semiconductor much more than temperature ever will and in some amount of time. I won't calculate it, because I'm really awful at physics myself, but it's pretty obvious too, undervolted card will only get like 10-20% voltage reduction and perhaps loss of 10 kelvins, which is like 2% reduction. So even considering the fact that it's an exponent, you may double the lifespan of your HW, but if you keep it on for 6 times longer, then it just means have to have a seriously aggressive undervolt and underclock to make your card's wear and tear similar to gamer's, which is not going to happen, because performance is crucial and you just end up with card that are at best (meaning that user cared to undervolt and keep them well cooled) are like 2 years old, but already are as worn out as 6 year old cards and that gets seriously worse very quickly. If person never undervolted them and kept them cool, card that is 2 years old will already have been worn out like 12 year old card. That's a big problem, because absolute minimum MTTF for desktop hardware is 5 years. Now, that MTTF number is probably calculated in 24/7 type of usage, because predicting average usage is way too subjective, but still, my point that you may buy way too worn out hardware for its age stands.

MTTF is a time value, but time is not in the factors contributing to MTTF. Why are you claiming that "time affects every single thing here."? It does not. What should be said is that everything here affects time to failure.

 

[The red spirit]

MTTF is a time value, but time is not in the factors contributing to MTTF. Why are you claiming that "time affects every single thing here."? It does not. What should be said is that everything here affects time to failure.

Time is literally the unit that MTTF calculates.

 

[maddie]

Time is literally the unit that MTTF calculates.

Yes, and also true, time is not in the factors that are used to calculate it, which you keep claiming, wrongly, You are calculating a time value using non-time factors, with the variables being, materials used, temperature and current density.

 

[In2Photos]

Time is literally the unit that MTTF calculates.

Exactly. It is the output of the calculation. None of the input values are time based or affected by time.

 

[The red spirit]

Yes, and also true, time is not in the factors that are used to calculate it, which you keep claiming, wrongly, You are calculating a time value using non-time factors, with the variables being, materials used, temperature and current density.

Yeah, I goofed up a bit, but that formula let's you calculate time to failure state. Time is variable if you want to get wear out percentage. My point that time is the most meaningful metric still stands. You still don't want a miner's card.

 

[Shmee]

I think the argument here about mining cards is a bit OT. We should probably get back to helping recommend cards for the OP.

I would recommend a used RX 5700XT in that range. Or a New RX 6600 (XT) or 6650XT series card on sale. I am considering selling one of my 5700XT cards myself. Check out the AT FS/T section.

 

[Hans Gruber]

I am deflated with the GPU market. I still makes no sense. I would suggest a 1660super/Ti or 3060 or 6700xt around black Friday.

 

[maddie]

For light gaming, a RX 6600 for slightly over $200 with 55-60% the performance of a RX 6900XT @ 1080p seems unbeatable.

 

[blckgrffn]

For light gaming, a RX 6600 for slightly over $200 with 55-60% the performance of a RX 6900XT @ 1080p seems unbeatable.

Sub $200 offerings are creeping in. Excellent perf/watt too.

 

[Leeea]

Like I said, miners consider their cards like chickens that produce golden eggs.
So they treat they well. THey don't run them into the ground, if they know enough not to.

I think your typical hobby miner was like that. People like you.


But, most mining cards were part of businesses. We all saw the warehouses, and there just was not enough labor on hand to just clean the dust off. Those folks are going to just work like every other business, max profit this quarter, we will worry about next quarter next quarter.

Yea, there are exceptions, but we call them exceptions for a reason.

They also are not going to reflash the bios before reselling now. Just spray them down so they look nice and sell them off.

And sadly, that is the vast majority of cards on the used market right now. Corperations shedding inventory.