GPU Buying Guide - (Read first post)

So, I spotted an NVIDIA GeForce RTX 3070 for $1283, but the specs say it needs a 650W power supply. I’m currently rocking a Corsair at 620W, so I'm wondering if that's gonna be an issue. Plus, honestly? That card feels like total overkill just for 1080p gaming. At 44, I'm basically ready to hang up the controller and enter my gaming retirement anyway.

Forget those spec sheets; they’re just rough guidelines, much like how not every 600W power supply is created equal. You can find 600W units designed for $50 and others that only pull $167 for 🙂
. The power supply you (or rather, "you all" 😬) have is more than sufficient, though how much you're willing to shell out is a different story entirely. 😉

Even the RTX 3060 isn't a bad buy—it'll perform about 60-100% better depending on the game. You can usually snag one for around $733

It’s not overkill for the long haul, and don't sweat it if you already have a 620W Corsair

Of course, it all depends on what you're playing. Honestly, if you aren't hitting the latest AAA titles, that card might be way more power than you actually need

I’d stick with the Intel Core i5-10400, and honestly—as far as GPUs go—it really just comes down to how much you're willing to shell out. In my humble opinion, if you grab an NVIDIA GeForce RTX 3060 for around two thousand bucks, you're looking at a massive jump from that old GTX 1060—like, almost double the performance! Plus, that setup should keep you going for a few more years easily.

Besides, with mid-range cards, you can always play around with the graphics settings and details in your games. There are tons of websites out there suggesting which specific settings you can dial back to get a decent boost in frames without really losing any visual quality.

I personally play Madison Square Garden on an Intel Core i5-10400 and a GTX 1060 at a pretty steady 60 FPS. Before I even started playing, I made sure to follow those online guides—there are so many of them anyway—so I don't waste energy stressing over how the game "might" look on ultra settings.

With an NVIDIA GeForce RTX 3060, you can tweak the details just a little... and then you'll be cruising past 100 frames without a single worry.

The RTX 3060 is only about 60% faster than the GTX 1060, which isn't even close to double!

Alexander Moore67 said:The RTX 3060 is only about 60% faster than the GTX 1060—that’s not even close to double!

Wait, seriously?
I was watching some clips on > and honestly, it looked like the frame rate jump was way more than just 60%... maybe I'm seeing things, but it felt much bigger than that!

Lawrence Phillips said:Are you serious?
I was watching some clips on YouTube and it looked to me like the frame rate difference was way more than just 60%.

You are. The gap ranges from about 60% up to double, depending on which title you're looking at.

Seeing the Radeon RX 6600 mentioned alongside the Radeon RX 580 actually got me thinking about whether pairing them together might make sense for my work in DaVinci Resolve, especially since I see the Radeon RX 6600 priced at $726 for SD workflows. I don't think the PCIe x8 limitation will be much of an issue, as having two cards running in parallel would likely limit them to x8 mode anyway, and when I tested the x8 speeds with my Radeon RX 580, I didn't notice any real performance hit, though I suppose with two cards, things might change. The memory bandwidth seems to lean slightly toward the Radeon RX 580 given its wider bus despite the lower clock speeds. My bigger concerns are more practical: I'll have two cards with relatively high TDPs sitting right on top of each other with barely a CMOS battery's width of space between them, and I'm wondering if a 600W EVGA SuperNOVA will actually hold up, considering I'm running a Ryzen 7 3700X...
I did look into the GeForce RTX 3060 12GB, but after running some of my own tests, it was only about 1.5x faster than the Radeon RX 580 in DaVinci Resolve, which felt a bit strange to me... especially since the Radeon RX 6600 shows a similar 1.5x boost in DaVinci Resolve based on the GPU effects score:

Radeon RX 6600: https://benchmarks.pugetsystems.com/....php?id=116150
Radeon RX 580: https://benchmarks.pugetsystems.com/...w.php?id=82699

However, these specific benchmarks for the GeForce RTX 3060 don't really align with what I saw during my own testing:

GeForce RTX 3060: https://benchmarks.pugetsystems.com/....php?id=117066

And then there is that whole confusing mess regarding the choice between the 8GB and 12GB versions of the 3060; the difference in Video RAM speed is massive, and we all know how much DaVinci Resolve craves fast Video RAM.

So, if I go ahead with this $733 investment, I could potentially see nearly double the performance, but if that doesn't pan out, I'll just be left with a 1.5x improvement and a Radeon RX 580 I'll need to sell...

What do you all think? Any advice or suggestions?

Interesting idea, but I’m not sure if it’s actually worth your money. First off, from what I can see, that power supply only has two 6+2 pin connectors. How many does that Radeon RX 580 actually need you to plug in? Some models demand both a 1x8 and a 1x6 pin setup. It might be fine without overclocking, but check your specs first. Make sure you even have enough connectors to run both cards. Even if you do, you'd be pushing that power supply right to its absolute limit.
Also, you realize you won't actually get more Video RAM, right? Doubling up on GPUs doesn't increase your Video RAM capacity; instead, both cards just handle the same task simultaneously, which makes things faster.
Furthermore, DaVinci Resolve doesn't support Crossfire; it uses its own specific dual GPU system. Honestly, I haven't a clue how it works under the hood. I did dig around and found some advice here.

https://forum.blackmagicdesign.com/v...?f=21&t=134479

The guy here says it's better to just buy one powerful card rather than pairing up two weaker ones. I don't know. Maybe it would work, but I fear you'll end up with more headaches than if you had just invested in a single, stronger card.

I've been working with DaVinci Resolve for quite a while now, so I'm not under any illusions about suddenly having an endless pool of memory at my disposal—that’s definitely not what I'm aiming for here. My Radeon RX 580 only requires a single 8-pin connector, which means I'm already set up and ready to go on that front. According to my math, I should probably be looking at a power supply of around 510W, so this 600W 80+ Bronze unit should handle things just fine, don't you think?

Spending somewhere in the neighborhood of $700 to $800 on a single GPU just to see a twofold increase in performance feels like a much more expensive route than most people realize...

I can't say much about specific workloads, but I can talk power draw.

I'm running a Radeon RX 580; my brother has a 6600XT.

Mine sits at a 185W TDP by default, and overclocking pushes it higher.
But, if you undervolt to 925-950mV and underclock to 1150MHz, it pulls about 100W at full load during Ether mining. Left the Video RAM at stock.
Obviously, you could go even lower on the underclock/undervolt and sit comfortably around 120W.
If your software is more memory-intensive than GPU core-heavy, you shouldn't lose much performance.

As for the 6600XT, it’s a very low consumer—around 150W stock, and the RX 6600 is even less...
Basically, you can easily get those two cards to pull a combined 300W, and each uses a single 1x8pin connector.

The PSU shouldn't be an issue. I just don't know if the software is pushing a Ryzen 7 3700X to full load at the same time...

wiredotter14 said:I can't really speak to the specific performance for your exact workload, but I can certainly talk about power consumption.

I'm running a Radeon RX 580, while my brother uses a 6600XT.

Mine has a default TDP of 185W, and if you start overclocking, it goes even higher.
However, if you try undervolting to around 925-950mV and an underclock of 1150MHz, it pulls about 100W at full load during Ether mining, provided you leave the Video RAM at stock settings.
Of course, you could push that underclock or undervolt a bit further and settle comfortably around 120W.
If your software happens to be more sensitive to memory bandwidth than the actual GPU core, you shouldn't see much of a performance hit.

As for the 6600XT, it’s actually quite efficient, pulling about 150W at stock, and the Radeon RX 6600 pulls even less than that...
So, you should easily be able to tune both cards to pull a combined total of under 300W, and since each has a single 1x8-pin connector, it's manageable.

It shouldn't pose any issues for your power supply, though I am curious if your software pushes the Ryzen 7 3700X to its limit at the same time...

It does happen sometimes—if the system needs to decode an input format that the GPU doesn't support, the CPU has to step in, otherwise the GPU handles almost everything and the CPU stays relatively idle. But even then, we're talking under 100W at full load, and there isn't much point in overclocking besides using PBO.

There isn't much sense in overclocking the Radeon RX 580 because the gains are so marginal, and a heavier undervolt tends to make it unstable under heavy loads like Speed Warp, so I just stick to stock. The 6600XT is pretty hard to find right now anyway, and while the 6650XT pulls about 175W, the Radeon RX 6600 sits at around 130W...

And what about the heat situation, given that you have two cards stacked right on top of each other? My Radeon RX 580 is hitting just over 80C when it's fully loaded.

It's interesting, though; this Radeon RX 6600 stays quite cool on the bottom side, specifically regarding the PCB:

https://www.computerbase.de/2021-10/...review-test/3/

You can't really push a heavy undervolt without dropping the clock speed. But if you dial that clock down below 1200MHz—specifically on a Radeon RX 580—you can shave off a ton of voltage.

It’s probably a similar story with the Radeon RX 6600.

That setup with two cards is way too close... obviously it isn't ideal, and I wouldn't bet money on it holding up well under a full load.

How does the 6650XT actually perform?
Maybe it makes sense to just dump the Radeon RX 580 and pick up a 6650XT instead?

Either way, I wouldn't obsess over the stock TDP. It's better to look at what you can actually achieve and the performance gains once you find the right undervolt.

I probably wouldn't bother messing around with undervolting unless I actually started running into performance issues.
The Radeon RX 6600 isn't really offering much of a leap for me, and what I'm truly looking for is something that delivers at least twice the speed of my Radeon RX 580 for around $450. If I were just looking at someone else's benchmarks, the GeForce RTX 3060 12GB would seem like the perfect fit, but after running my own tests, I've realized it's not quite there. My next move would likely be picking up an Radeon RX 6700 XT, which would definitely provide more than double the performance for roughly $1400, especially if I can manage to sell my Radeon RX 580 for about $333, which would bring my total right to that $470 mark...
That said, I'm not exactly in a rush to sell anything right now, though I do wonder if waiting too long will just mean watching the prices continue to slide further away from me...

I'm not sure you'll get $333 for a Radeon RX 580. Honestly, $700-$267 feels more realistic.

Grabbing a 6700XT for $1400 isn't much of a steal when the Radeon RX 6800 is already $1560 and offers better Video RAM and higher bandwidth.

If you hold out for a few months, I bet the Radeon RX 6800 (new or used) will drop down to the $3000-$1167
range.
How does it actually stack up performance-wise?

Well, looking at the data from Puget Systems—though I’ve started having my doubts about whether their methodology is always the gold standard, even if they are pretty much the only reliable source left—the Radeon RX 6800 is easily at least three times faster than the Radeon RX 580, but I really don't expect such a massive price drop until they're almost entirely cleared out of stock. I managed to track down a 6700XT for $1233 locally, or I could grab a 3400 on Amazon, which is roughly 2.5 times faster.

P.S. Why do I feel like Puget isn't entirely relevant? I think they're great for comparing AMD cards against one another, but when you pit AMD against Nvidia, things get tricky. They benchmark by rendering in H.264 using the GPU's hardware encoder rather than the CPU and then logging the total render time, but in my own experience, NVENC is drastically faster than the VCE, even on a much weaker Nvidia Quadro used for compute tasks, which naturally makes the AMD scores look worse. Honestly, though, it doesn't matter much to me since I never actually export my final projects in H.264 or H.265; I stick to formats that rely on the CPU.

The Radeon RX 6800 was hovering around $600 for the last couple of months, and you could snag a 6800XT for about $700 in the States. Now that Black Friday is over, prices have jumped by maybe $100.

With the 7900XT hitting the market at, say, $1100, I’m expecting the 7800XT to land around $850 and the RX 7800 to sit near $750.
That means the previous generation's prices absolutely have to drop...

It’s possible AMD will pull the plug on the old cards relatively quickly if manufacturing the new ones becomes cheaper, but either way, those older prices need to fall. Otherwise, the value proposition is just going to be a disaster.

I’m feeling a little mischievous today, just diving deep into the specs for the Intel Arc... I mean, looking at 256-bit memory bandwidth, 512 GB/s speeds, and 17 Teraflops, all for under $450... 😵

Teraflops don't mean much when you're comparing different architectures, especially between rival brands. But yeah, generally speaking, this card should have been in the GeForce RTX 3070 class—maybe even better if it had a 16GB option—but... it just isn't.

Supposedly it handles encoding and decoding well enough, but otherwise, it’s a total lottery depending on which app or game you're running....

🙂

Yeah, I’ve been looking into Puget Systems, though I tend to take their benchmarks with a grain of salt—the Intel Arc performance just hasn't been quite what I was hoping for. I find myself circling back to that idea of running two RX series cards at once... if I decide to dial back the clock speeds and voltage, the power draw from the GPU itself (not the whole board, mind you) actually drops below 100W while keeping temperatures steady around 70C, all without losing more than maybe half a frame per second. It really comes down to whether I want to go that route or just wait for a high-end card to see a significant price drop.