News Upscaling can't save the world's slowest 'modern' GPU — FSR doubles performance on GT 1030, but titles still barely playable

[Neilbob]

At least the 1030 was priced somewhat accordingly, quite unlike the next **30 card they released, which was the GTX 1630.

That thing could still conceivably considered 'modern', but manages to make the current crop of overpriced 8GB excretions seem like terrific value. Even the review right here says that it manages to make the Radeon 6400 look good. If that's not an achievement, I don't know what is.

 

[bit_user]

It does not, only the 5060 Ti comes in both.

Oh, sorry. You're right. I got them confused.

I wouldn't be surprised if we saw a 12GB 5060 Super when nvidia feels like using higher capacity memory IC though.

True.

The 5060 does undermine the 5050 in terms of value/performance though.

That was my main point. The main argument I see for keeping both is if the RTX 5050 has some margins they can cut to reach lower market tiers or deal with more challenging market conditions. However, this being Nvidia, they hate to just cut prices without cutting specs. So, I'm still not sure there's not too much crowding.

Also, pretty sure I saw a recent news article predicting a rise in GDDR6 prices. That might impact the supply of RTX 5050's. So, maybe RTX 5060 is there as a hedge against that.

 

[thestryker]

That was my main point. The main argument I see for keeping both is if the RTX 5050 has some margins they can cut to reach lower market tiers or deal with more challenging market conditions. However, this being Nvidia, they hate to just cut prices without cutting specs. So, I'm still not sure there's not too much crowding.

Also, pretty sure I saw a recent news article predicting a rise in GDDR6 prices. That might impact the supply of RTX 5050's. So, maybe RTX 5060 is there as a hedge against that.

They've made so many anti-consumer moves what I think the 5050 exists for when the 5060 Super comes out. The way I see it there's no chance of nvidia charging less for more VRAM even though the performance should be the same (maybe a tiny TDP bump) so I could see a 5060 Super being ~$320 and them discontinuing the 5060. Then if they're able to keep the 5050 priced around $250 it's a "better" value than today.

 

[JohnyFin]

This is a joke? Exists something like technology threshold when impossible means impossible.

 

[rambo919]

Funny thing is how this thread is an exercise that will be repeated again in the future because people believe "8GB VRAM is all you will ever need"..... and probably at the time believed "2GB VRAM is all you will ever need".......

 

[rambo919]

They've made so many anti-consumer moves what I think the 5050 exists for when the 5060 Super comes out. The way I see it there's no chance of nvidia charging less for more VRAM even though the performance should be the same (maybe a tiny TDP bump) so I could see a 5060 Super being ~$320 and them discontinuing the 5060. Then if they're able to keep the 5050 priced around $250 it's a "better" value than today.

Ignoring the costs of the hardware which itself is a case of the consumer gets assumed..... and ignoring the shuffle games they play with the names of the tiers. People are stupid and keep buying regardless because of the catharsis involved in overspending, frogs loving the heating water personified.

The pattern is clear for each generation, the highest end card is for enthusiasts who have no problem dealing with issues, the lowest end card is exploitation because it will barely last for any length of time. The only difference now is the mid level has become vague.

This is true for everyone but Nvidia is the most guilty here. Garbage and overly expensive hardware exists because someone will buy it.... otherwise a lot of laptops would not exist.

Frankly I am kinda impressed that some people still use 1030's.....

 

[bit_user]

Funny thing is how this thread is an exercise that will be repeated again in the future because people believe "8GB VRAM is all you will ever need"..... and probably at the time believed "2GB VRAM is all you will ever need".......

Certainly nobody here is saying "8GB is all you will ever need". I did say I think devs will be stuck supporting it, for a good while. If I were buying a card for gaming and could afford to go 16 GB, I would.

It would be interesting to see how far back in time you have to go, in order to find respectable build guides telling people to go for a 2 GB card. Even then, I'm sure they didn't say it like it's all someone would ever need.

FWIW, I have an old HD 7870 (mid-range card), on a shelf, that has 2 GB. It launched in March 2012. So, that's a data point.

 

[thestryker]

The pattern is clear for each generation, the highest end card is for enthusiasts who have no problem dealing with issues, the lowest end card is exploitation because it will barely last for any length of time. The only difference now is the mid level has become vague.

I genuinely feel bad for anyone who can't buy a more expensive video card these days. I bought a GeForce 4 Ti 4600 for ~$400 (~$650 adjusted for inflation at around the time I got my 3080) on launch and then didn't buy high end again until buying my RTX 3080 12GB for ~$815. Everything I bought between those two cards were actual mid range which has now all shifted to $500+. The only cards below $500 people should realistically be buying are the 5060 Ti 16GB, 9060 XT 16GB and B580 (B570 would potentially be okay too) though pricing is a bit of a mess. It would be so much better for the market as a whole if 8GB cards were limited $200 and below parts.

Spoiler: a bunch of pricing stuff on cards I bought

$ paid at time ($ approx inflation to today)
FX 5900 XT: $188 ($322)
6800GT: $279 ($450)
7800GS*: $309 ($500)
9800GT: $130 ($200)
GTX 275: $215 ($320)
GTX 460: $235 ($347)
GTX 560 Ti: $250 ($365)
GTX 660 Ti: $310 ($433)
GTX 970: $340 ($460)
GTX 1660 Ti $225 ($278)

*I don't remember actually buying this card for myself as I bought two of them at different times and had a card die in there somewhere, but I picked the more expensive of the two.

 

[rambo919]

Certainly nobody here is saying "8GB is all you will ever need". I did say I think devs will be stuck supporting it, for a good while. If I were buying a card for gaming and could afford to go 16 GB, I would.

It would be interesting to see how far back in time you have to go, in order to find respectable build guides telling people to go for a 2 GB card. Even then, I'm sure they didn't say it like it's all someone would ever need.

FWIW, I have an old HD 7870 (mid-range card), on a shelf, that has 2 GB. It launched in March 2012. So, that's a data point.

The only thing I actually firmly remember is the 1050ti being widely praised as "4GB VRAM being all you will ever need for 1080p". Now it's actually held up very well but clearly that was wrong.

My point is that these beliefs have always come and gone, not forgetting the infamous 640k one. People seem downright terrible at predicting the future when it comes to any memory. The other current "perpetual" myths now include RAM timings and speed and no one ever learns the lesson of being forward looking instead of backward looking.

You constantly (perhaps not in this specific thread.... yet) see on these forums people in the delusional cope that "8GB VRAM is perfectly fine and will always be for 1080p"..... I cannot imagine why people would defend this obvious planned obsolescence in the face of how larger capacities are actually cheaper in isolation because of SSD's expanding production output and no amount of logic has been able to break through to them.

A big problem is probably how GPU manufacturers are trying to service AI and Desktop at the same time and their brains are getting wonkey from having to support wildly different form factors. AI rigs simply should be designed for high energy demand and you simply cannot do that on Desktop motherboards as they currently exist. The way PCIe is implemented must be completely rethunk.... if nothing else if things meander on as they are soon GPU's must get their own dedicated power supplies or regular PSU's must essentially double in what they can do which will probably have at least 3 times a price increase if you think about it.

That is assuming the hardware AI for individuals trend is not a bubble the way that crypto was. Even NAS is not yet really a consumer thing because of how ludecrously expensive and complicated it still is to start nvm maintain... and it's older. I have to wonder how long the water cooling trend will still last because that too is showing early signs of dying out.

 

[rambo919]

I genuinely feel bad for anyone who can't buy a more expensive video card these days.

I just get irrationally angry that people encourage others to buy hardware that will become outdated sooner than it should in the assumption that everyone has the money to constantly upgrade the way that they do.... nvm the mountains of ever expanding ewaste in the service of corporate greed.

The more money people have the less they realize its worth... and often objectively wealthy people whinge and moan about their supposed poverty because they compare themselves to even wealthier people.

We all have problems we have to work out I guess.

 

[bit_user]

My point is that these beliefs have always come and gone, not forgetting the infamous 640k one.

That's often taken out of context. The subject was around carving up the memory map, which was hard-limited to 1 MB. So, the actual statement was talking about devoting 640k out of like 1088k (it's slightly more than 1024, due to how 20-bit x86 addressing works). That remaining 440k also didn't go nowhere. Some of it went for BIOS, while some of it got reserved for memory-mapped devices, like video cards. Years later, when VGA hit the scene, it was much faster than it otherwise might've been, thanks to the 64k that had been set aside for it.

People seem downright terrible at predicting the future when it comes to any memory.

I want to see an actual quote, from a reputable source, about video memory. Because, I think you're playing fast-and-loose with the facts, here.

A big problem is probably how GPU manufacturers are trying to service AI and Desktop at the same time

How is this a problem for VRAM capacity? If not for AI, there's no way Nvidia would've put 32 GB on the RTX 5090! It would've been another 24 GB card, like the RTX 4090.

AI rigs simply should be designed for high energy demand and you simply cannot do that on Desktop motherboards as they currently exist. The way PCIe is implemented must be completely rethunk.... if nothing else if things meander on as they are soon GPU's must get their own dedicated power supplies or regular PSU's must essentially double in what they can do which will probably have at least 3 times a price increase if you think about it.

Power at the wall becomes a limiting factor, for some. Cooling is another huge problem, if you want to scale up PC power consumption.

 

[rambo919]

That's often taken out of context. The subject was around carving up the memory map, which was hard-limited to 1 MB. So, the actual statement was talking about devoting 640k out of like 1088k (it's slightly more than 1024, due to how 20-bit x86 addressing works). That remaining 440k also didn't go nowhere. Some of it went for BIOS, while some of it got reserved for memory-mapped devices, like video cards. Years later, when VGA hit the scene, it was much faster than it otherwise might've been, thanks to the 64k that had been set aside for it.

Granted

I want to see an actual quote, from a reputable source, about video memory. Because, I think you're playing fast-and-loose with the facts, here.

It's something I remember from the time in almost all the reviews I read.... one might be PCGamer (wasn't yet the total disaster it is now) I dunno it's way too long ago to remember that specifically.... and I really don't care enough to spend hours trying to find anything. I was specifically trying to choose what new GPU to buy so this stuck in my mind.

In 3-5 years time I am confident you will see common claims of "16GB VRAM is all you will ever need for 1440p" if not 1080p.

How is this a problem for VRAM capacity? If not for AI, there's no way Nvidia would've put 32 GB on the RTX 5090! It would've been another 24 GB card, like the RTX 4090.

A technical problem that was mentioned by graphics card makers in recent years in defense of their insistence that 8x speed is enough for 8GB models, the more memory you put on the card mobo the more attention you have to pay to lanes.... so the more VRAM you put on a graphics card the more expensive it becomes. This is part of why high VRAM models are so much more expensive.

Power at the wall becomes a limiting factor, for some. Cooling is another huge problem, if you want to scale up PC power consumption.

Well the way PSU's are manufactured the more power they consistently output the more difficult and thus more expensive they are to do. On the other hand..... power bills are not getting cheaper so how many people can operate a PC that dims the house lights if you turn it on?

Air cooling has if not overtaken by now reached enough parity with water cooling... especially since both basically cost the same now for the same effect with water cooling having a very low lead.... at least for most people. If you can boast about being 5-10 celsius cooler than the next product over that is much less expensive.... you won't win that budget race.

 

[thestryker]

I just get irrationally angry that people encourage others to buy hardware that will become outdated sooner than it should in the assumption that everyone has the money to constantly upgrade the way that they do.... nvm the mountains of ever expanding ewaste in the service of corporate greed.

I totally understand this. I used to upgrade because I was young and it was something to have fun with. I gave friends cards or put them in spare boxes as when I was replacing them they were still good (barring a dead card, but I only remember that once). I ended up keeping my GTX 970 for 5 years or so and I just don't think an 8GB card today is going to have that experience whereas 12GB could and 16GB almost assuredly will. I think the 9060 XT 16GB is probably the best sub-$400 card we've seen since the RTX 20 series and it's still got a $350 MSRP which is a steep cost of entry for a lot of people and can't be bought at it anyways.

I was having this discussion with a friend the other day that if someone can't afford the 9060 XT 16GB they're probably best buying used. The thing that stands out about this is that we're not talking about doing so to get something dirt cheap. Used 3070s when I'd looked were going for around $250 and they're faster than everything under $380 that's new so if you're stuck with 8GB VRAM anyways why not.

 

[bit_user]

A technical problem that was mentioned by graphics card makers in recent years in defense of their insistence that 8x speed is enough for 8GB models, the more memory you put on the card mobo the more attention you have to pay to lanes.... so the more VRAM you put on a graphics card the more expensive it becomes. This is part of why high VRAM models are so much more expensive.

My theory is that PCIe speed doesn't directly have anything to do with VRAM. I know people have done benchmarks where some low-memory card is abnormally affected by limited PCIe speed, but I think those are corner cases. If you're running low on VRAM, more PCIe lanes or a newer bus standard won't save you from bad 1% framerates. Your best bet is to lower settings or resolution until the VRAM is no longer under pressure.

IMO, it's the faster cards that can justify faster PCIe connectivity. With lower-end cards, PCIe just ends up not being the bottleneck, as long as you don't overfill your VRAM capacity.

 

[Drefsab]

So the absolute bottom end of a generation of cards thats about to be phased out struggles with games now just like it did at launch and somehow this is news? The 1030 was a joke as a gaming card when it was released you were always better of putting the money into something else.

 

[rambo919]

My theory is that PCIe speed doesn't directly have anything to do with VRAM. I know people have done benchmarks where some low-memory card is abnormally affected by limited PCIe speed, but I think those are corner cases. If you're running low on VRAM, more PCIe lanes or a newer bus standard won't save you from bad 1% framerates. Your best bet is to lower settings or resolution until the VRAM is no longer under pressure.

Its not so much a speed as number of lanes thing. Its like driving on a thin as opposed to broad road.... you never really know when you are going to need more road to maneuver in.

I find the idea that the number of physical VRAM chips influencing optimal lane counts highly plausible but not enough of a hardware guy to know for sure.

IMO, it's the faster cards that can justify faster PCIe connectivity. With lower-end cards, PCIe just ends up not being the bottleneck, as long as you don't overfill your VRAM capacity.

And people get into vhement arguments as to how much speed you actually lose if you put a PCIe5 GPU in a PCIe3 slot. 1-10% in one use case can easily become 10-30% in another.

Everyone seems to make this personal for some reason I am just tired of all the drama.

 

[bit_user]

Its not so much a speed as number of lanes thing. Its like driving on a thin as opposed to broad road.... you never really know when you are going to need more road to maneuver in.

Bandwidth is bandwidth. Doesn't matter if it's half the lanes or half the speed per lane. It's the overall sum that matters.

And people get into vhement arguments as to how much speed you actually lose if you put a PCIe5 GPU in a PCIe3 slot. 1-10% in one use case can easily become 10-30% in another.

No, never. You're playing fast and loose with numbers, again.

Here, read this:

NVIDIA GeForce RTX 5090 PCI-Express Scaling

The new NVIDIA GeForce RTX 5090 is the first high-end graphics card that makes use of a PCI-Express 5.0 bus interface. Are you in trouble when trying to run it on PCIe 4.0? What about x8, like when an SSD is using up some bandwidth? We've also tested various PCI-Express 3.0, 2.0 and 1.1 configs...

www.techpowerup.com

The only time you might see numbers like what you mentioned is when you overfill your VRAM. And like I said, if you're relying on faster PCIe connectivity to swap stuff in/out, your 1% lows are going to suck. So, it's a false solution. Better off just dialing back settings until VRAM is no longer overfilled.

Everyone seems to make this personal for some reason I am just tired of all the drama.

You might be creating some of it.

 

[rambo919]

Bandwidth is bandwidth. Doesn't matter if it's half the lanes or half the speed per lane. It's the overall sum that matters.

Somehow I think physics probably has it's own opinion....

No, never. You're playing fast and loose with numbers, again.

Here, read this:

​

NVIDIA GeForce RTX 5090 PCI-Express Scaling

The new NVIDIA GeForce RTX 5090 is the first high-end graphics card that makes use of a PCI-Express 5.0 bus interface. Are you in trouble when trying to run it on PCIe 4.0? What about x8, like when an SSD is using up some bandwidth? We've also tested various PCI-Express 3.0, 2.0 and 1.1 configs...

www.techpowerup.com

​

The only time you might see numbers like what you mentioned is when you overfill your VRAM. And like I said, if you're relying on faster PCIe connectivity to swap stuff in/out, your 1% lows are going to suck. So, it's a false solution. Better off just dialing back settings until VRAM is no longer overfilled.

But that ignores the slowdown due to using newer gen cards on older gen motherboards.... which has nothing to do with VRAM.... unless VRAM capacity is somehow baked into PCIe generations?

You might be creating some of it.

I might autistically unintentionally trigger it sometimes... but often it already exists and I just point it out.

 

[bit_user]

Somehow I think physics probably has it's own opinion....

I'm just telling you what the data says. Each PCIe generation is double the speed of the previous. So, whether you step up/down a generation or double/half the number of lanes, it has the same effect.

But that ignores the slowdown due to using newer gen cards on older gen motherboards....

The article I linked covers this. They look at different speeds and generations. If this is a topic you care about (seems to be?), I think it'd be worth spending some time to at least look through the findings in that piece.

 

[rambo919]

Here, read this:

​

NVIDIA GeForce RTX 5090 PCI-Express Scaling

The new NVIDIA GeForce RTX 5090 is the first high-end graphics card that makes use of a PCI-Express 5.0 bus interface. Are you in trouble when trying to run it on PCIe 4.0? What about x8, like when an SSD is using up some bandwidth? We've also tested various PCI-Express 3.0, 2.0 and 1.1 configs...

www.techpowerup.com

Looking at those benchmarks it confirms what I asserted though.... there is one obvious problem with the testing in that they only used the top end of the generation when they should also have tested the bottom end and compared....

 

[bit_user]

Looking at those benchmarks it confirms what I asserted though....

No, not the specific numbers you gave.

there is one obvious problem with the testing in that they only used the top end of the generation when they should also have tested the bottom end and compared....

The card most sensitive to PCIe speed is the top-end card.

The only time a lower-end card could be more sensitive is if you overfill VRAM, but faster PCIe will not save you from bad 1% lows (i.e. stutters). It won't be a good gaming experience, no matter what. So, it's not worthy of serious investigation. It's much better to just avoid that situation by dialing back your settings.

 

[rambo919]

I'm just telling you what the data says. Each PCIe generation is double the speed of the previous. So, whether you step up/down a generation or double/half the number of lanes, it has the same effect.


The article I linked covers this. They look at different speeds and generations. If this is a topic you care about (seems to be?), I think it'd be worth spending some time to at least look through the findings in that piece.

You replied faster than I did. To spell it out... as predicted 1-10% roughly eventually translates to 10-30% as you proceed down the line. You just need to skip to the end to see this.

Also as I mentioned above they did not test x8 or low end hardware for that matter which would more starkly outline the problem.

 

[rambo919]

No, not the specific numbers you gave.


The card most sensitive to PCIe speed is the top-end card.

That could be true towards one extreme IF the game maxes out the hardware which I don't see in their tests.

On the other hand if hardware is slow in something at PCIe 5.... it would be more likely to be slow at lower generations.

The only time a lower-end card could be more sensitive is if you overfill VRAM, but faster PCIe will not save you from bad 1% lows (i.e. stutters). It won't be a good gaming experience, no matter what. So, it's not worthy of serious investigation. It's much better to just avoid that situation by dialing back your settings.

You are arguing with average FPS scores that contradict you.....

 

[bit_user]

You replied faster than I did. To spell it out... as predicted 1-10% roughly eventually translates to 10-30% as you proceed down the line. You just need to skip to the end to see this.

I've gone through several generations of these articles in quite a bit of detail, including this one. There's no place where "PCIe5 GPU in a PCIe3 slot" results in "10-30%" performance loss. You have to also cut the lane count, in order to see such a drastic performance reduction. Like I said, bandwidth is bandwidth.

Also as I mentioned above they did not test x8 or low end hardware for that matter which would more starkly outline the problem.

They certainly did try cutting lane count.

 

[bit_user]

You are arguing with average FPS scores that contradict you.....

You need to slow down and go through the charts in careful detail. Take some time and understand what they're saying. Reading the article text might help. They only support what I'm saying. There's no contradiction.

I can't help you any further. I've shown you where the water is. Only you can decide whether to drink.

P.S. usually, when people claim the data refutes what someone is saying, they cite a specific example. Not that I want to continue this exchange, but you would need to say which entries of which chart, on which page contradict what I said.