Hacker News Reader

Comment by Tenoke | Hacker News Reader

Good way to spin having only 4GB vram into an 'anti-mining feature' I guess.

Of course, 4GB is typically considered the bare minimum for modern 1080p (let alone higher) gaming and this is just a budget card which was definitely not limited with any mining considerations.

View on HN

The fact that the entry-level card can run its clock 30% faster than the same line's top-end offering supports the story pretty solidly, IMO. Not supporting 8 GB of vram can't have given them enough breathing room to casually run the clock that fast, so this has to have been decided way up front.

Like the article says, it remains to be seen whether this will also make the card bad at gaming, but I think that AMD at least had a coherent plan here.

The RX 6500 XT is a worse card than the RX480 released 5 years ago https://i.redd.it/oyxyoxmch2a81.jpg .

This is kind of like Google and Youtube becoming worse because they have too many of the wrong kinds of users.

I'd expect otherwise actually.

RDNA is far more efficient at gaming. GCN was compute-focused (higher TFLOPs, but weaker in practice). Case in point, micro-benchmarks show that VRAM latency is ~100 nanoseconds on RDNA, but ~300 nanoseconds on GCN.

RX 6500 XT is aimed at roughly the same specs as the RX480, but should perform slightly faster in practice, along with Raytracing support, Infinity Cache and therefore faster VRAM latency (probably faster fill-rates and other microbenchmarks important to gaming), all at 107W instead of 150W.

---------

As such, Rx 480 would be a better mining card, while 6500 XT would be a worse mining card, but better video game card.

Its still stupid for AMD to market their cards like this, but they're not lying... (and whatever argument marketing wants to make... well... that's their job. It all looks stupid to me but I can't honestly call them out on it since it is true...)

-------

Turns out that video-gamers who want more uniformity and less jitter between frames want higher-clock speeds (notice the 2000+ MHz clock speed on 6500 XT) rather than raw TFLOPs (Rx 480 ran a slow clock of ~1000MHz but had many more compute units / SIMD units computing in parallel).

Width of GPU leads to more variance, because its harder to load-balance your parallelism. If you have a narrower GPU (fewer compute units) at higher speeds, its easier to reach more consistent frame times / less jitter.

Further, given the prices of RX480 and modern video cards, even if they were on par, if you can actually GET RX6500XT for $200 today, then:

1. It sucks from big picture perspective, as we are "back where we were 5 years ago"

but

2. It's great from pragmatic perspective, as we are otherwise today in "Wish I could afford a card from 5 years ago" / "there's no such thing as a budget GPU" stage.

If we look at the benchmarks of the 5500XT vs the RX480 I expect it to maybe be 5-8% faster if at all. AMD basically released this card 3 times now but this time they scrubbed the whole encoding feature set.

> AMD basically released this card 3 times now but this time they scrubbed the whole encoding feature set.

Do you not see the difference between GCN, RDNA, and RDNA 2?

At a minimum, this RDNA 2 6500 XT has raytracing cores that the two previous versions do not have. I'd assume that most video gamers care more about the raytracing feature (especially as more and more games are using it) than H265 encoding.

Remember: Raytracing instructions were invented on RDNA2. The 5500 doesn't have it, and the 480 doesn't have it.

-------

I mean, seriously. Would you rather have the existence of this card? Or the non-existence? If you want something faster, that's why the 6800 exists. This 6500 XT is for people who want a barely-gaming ready card at the cheapest possible price. If that's not your use case, then don't buy the card.

But every now and then, I build a computer for a child who doesn't play many video games, or for some uncle/aunt who doesn't know much about computers. I never like the idea of going iGPU only, because these people inevitably find at least one weird game that they like to play. So I like to put in a $100 to $200-class card in there. Not expensive enough to break the budget, but high enough that it'd perform decently on 10-year-old games and maybe a few indie-games.

Are you suggesting that you could use Raytracing at all in modern games on a low end card like the RX 6500 XT? If you would use it you would lose the marginal performance increase of the updated architecture. Also content creation is a big point for younger generation if it is livestreaming, recording clips or general gameplay. This will both not be doable with this card if you don't have a good cpu. There is no discussion that more cards are good in the market we have right now. The card is 50% over MSRP right now at most retailers.

> Are you suggesting that you could use Raytracing at all in modern games on a low end card like the RX 6500 XT?

Yes?

Minecraft Raytracing and Quake Raytracing are rather low-specs and probably would run on an RX 6500 XT. You might have to drop down to 720p but it probably would run.

Minecraft release 18 November 2011. Quake release June 22, 1996.

They may have frankensteined the engines to support ray tracing but these are not modern games at all.

Are you kidding? These are fully path traced games and they are extremely hard to render. RDNA2 cards are worse in raytracing titles than Nvidia cards. For example 6900xt is half as performant in Minecraft RTX compared to 3090(18FPS vs 40FPS) according to LTT tests(testing was in 4k). And according to random tests I saw on Youtube. RX6600 outputs around 6fps with RTX enabled in 1080p, and it's basically 50% stronger. Not sure about 720p, but I would be surprised if it 5500XT gets to 10FPS at that resolution.

> Minecraft Raytracing and Quake Raytracing are rather low-specs and probably would run on an RX 6500 XT.

[citation needed]

Strictly speaking, it will "run", but if it run well that's another thing entirely.

It was the reason why the meme for heavy game morphed to "but can it run Minecraft RT?" for a while.

That picture is misleading. It doesn't show that the 6500 has >2x the clock speed and lower power requirements.

2x the clock speed is crazy.

2x the clock speed is a combination of design tradeoffs and Moore's Law.

I don't think that's true - the GTX 1060, which was its' competitor, and matched it's performance pretty closely only had 4.375 TFLOPS of compute. Architecture matters

Do you have a more reliable source than a random Reddit image assemblage?

You can check it yourself on AMDs website.

Well is it cheaper than a 5 year old rx480?

It's the same MSRP.

"MSRP"

First partners selling the part in Europe already have a hefty 50% surcharge [0].

[0] https://videocardz.com/newz/amds-199-radeon-rx-6500xt-offici...

I meant once it's released will it cost me less to get a rx6500 or a rx480.

It looks like it is just a different card designed for a different purpose. High clock speed, more power efficient, and a lot smaller...and $200 (supposedly).

I have both those cards and your very wrong. The RX480 with 8gb still can’t match the RX6500 with 4gb. The RX480 was a great mid tier card, but it’s showing it’s age and the RX6500 is a good upgrade. It’s not drastically better, but enough to play most games 1080p and medium to high graphics. I was stuck at 1080p and low to medium with the Rx480.

Isn't it a power budget and heat dissipation thing? For example the Xbox Series X and Playstation 5 GPUs have pretty similar architectures, but the PS5 has fewer compute units at a higher frequency. Surely Microsoft would have loved to clock theirs as high as Sony.

PS5: 10.28 Teraflops, 36 Compute Units running at 2.23GHz (variable frequency)

XSX: 12.11 Teraflops, 52 Compute Units running at 1.825GHz (fixed)

> but the PS5 has fewer compute units at a higher frequency.

Yes, this seems to be what's happening here. Less power used on memory bandwidth, means more is available for other things. This card is not so much "bad at mining" and more like trying to design for a different niche that might be underserved by all the "good at mining" GPU's.

I love how Mark Cerny managed to spin Sony having to raise the clock frequency of the PS5 GPU to compete with a the XSX's TFlop specs as an advantage. GPUs tend to be memory-bandwidth bound - if you play around with raising the clock frequency by 10% from stock on most GPUs, you won't get 10% more fps, likewise if you drop the frequency by 10%, you won't lose 10% fps.

But with the power consumption scaling with the square of the frequency, the difference in power draw can be absolutely significant. That's the reason many people (me among them) chose to underclock their 5700XTs

My 1060 3GB can still run every game I throw at it (except Cyberpunk). MS Flight Sim runs great at high, for instance, especially after the performance update.

What FPS do you consider "great"?

Looks like ~60 (more in rural areas, less in cities) for 1080p low on that card... which isn't fantastic but better than I expected. The limiting factor in cities seems to be the VRAM as the 1060 6 GB version can bench about the same flying through New York City bumped up to medium and shows using nearly 5 GB of VRAM while doing it.

For a flight sim, 45fps. Forza Horizon 5 runs at 60+ on high settings (despite telling me I need more memory, it doesn't crash). I mostly play CS, which runs at 200+ (yes I know it's an older game, Doom Eternal also runs at 60+ on high).

4GB ram is not enough for a good 1080p gaming card. You'd most likely only be able to play esport titles on the 6500 XT. I went from a 3GB 1060 to a 6GB 5600 XT and it makes the extra frame buffer makes a massive difference as new games are getting very demanding to run, even on 1080p.

Question for the experts in the comments:

I'm playing around more with deep learning as of late. Are the same technical specifications one would look for in a 'miner-friendly' GPU the same ones I'd look for in a GPU for Neural-net computations?

Reply on news.ycombinator.com