Monday, December 27th 2021

AMD Navi 24 GPU Powering RX 6500 XT Built on 6nm

AMD's first GPU built on the N6 (6 nm) silicon fabrication process isn't some big RX 7000 series behemoth, but the smallest chip from the Navi 2x GPU family, codenamed Navi 24. Based on the same RDNA2 graphics architecture as the rest of the RX 6000 series, the Navi 24 physically packs 1,024 stream processors across 16 compute units (8 WGPs), and on the RX 6500 XT, reportedly comes with 4 GB of memory across a 64-bit wide memory bus. The chip also packs a tiny 16 MB Infinity Cache. VideoCardz scored the first renders of the upcoming Radeon RX 6500 XT and RX 6400, which are based on the Navi 24. The RX 6500 XT features a full-height, 2-slot board design that uses a simple aluminium monoblock fan-heatsink. The RX 6400, on the other hand, is not just low-profile (half-height), but also single-slot.

Update Dec 28th: Unless we're mistaken, the SMDs near the PCIe interface in those renders seem to suggest that the GPU features a PCIe x4 interface. This should offer sufficient bandwidth for a GPU in this segment, and should help lower the pin-count of the GPU, as well as board costs.
Source: VideoCardz
Add your own comment

94 Comments on AMD Navi 24 GPU Powering RX 6500 XT Built on 6nm

#51
Ellertis
ModEl4I'm not too optimistic about these, even with higher clocks than 7nm (let's say 2847MHz for RX 6500XT) we will probably have regression to performance/$ regarding SRPs compared to $159 1650 super and $169 RX 5500XT 4GB!
If RX 5500XT 4GB is 100% in 1080p performance probably we will have something like below:
RTX 3050ti 8GB 139% $219-249 SRP
GTX 1660super 6GB 127% $229 SRP
RTX 3050 8GB 116% $179-199 SRP
RX6500 4GB 116% $199 SRP
GTX 1660 6GB 114% $219 SRP
GTX 1650super 4GB 101% $159 SRP
RX5500XT 4GB 100% $169 SRP
RX6400 4GB 94% $169 SRP (if not OEM only)
Also there is a possibility for a RTX 3050 4GB at probably -$20 vs the 8GB version that will further spoil the performance/$ image of RX 6500/6400!
Price/performance was degrading for the last 5-6years,only Navi 23 brought something better than rx 470 in this metric
Posted on Reply
#52
QuietBob
Vayra86These low-end AMD GPUs smell like Huang's leather jacket. Not his new one either, but the one he's been wearing the past decade while leading.
Made my day! :roll:
Posted on Reply
#53
aindriu
If this card isn't available through the AMD website price is going to go up hours or days after launch, an extra 2GB RAM would have helped but it would also attach the miners.
Posted on Reply
#54
Dranzule
u2konlineWhy bother to make these cards that have 64bit bus lol. stop it please. and 4gb? sigh. There's plenty of cards you can buy on the market that's better. Even my WX 4100 is faster than that card.
I'd argue Infinity Cache can make up for it. These are low-medium 1080p cards, not ultra 1080p ones. Considering that faulty dies are also a thing and that these are theoretically meant to succeed cards such as the 1050ti/1650 Super, they would be priced accordingly.
We know they won't though.
Posted on Reply
#55
Ellertis
DranzuleI'd argue Infinity Cache can make up for it. These are low-medium 1080p cards, not ultra 1080p ones. Considering that faulty dies are also a thing and that these are theoretically meant to succeed cards such as the 1050ti/1650 Super, they would be priced accordingly.
We know they won't though.
Infinity cache helps with bandwidth, but it doesn't compensate for the lack of vram
Posted on Reply
#56
Dranzule
EllertisInfinity cache helps with bandwidth, but it doesn't compensate for the lack of vram
I was talking about the bus. Regardless, 4gb VRAM isn't good, but should be decent. Anymore would make these cards attractive to mining as they're super efficient. They're also meant to be as cheap as possible.
Posted on Reply
#57
Garrus
Kind of hilarious the Xbox Series S is a better investment for gaming than PC gaming now. The Series S will beat the RX 6500 XT. Sad.
Posted on Reply
#58
Ellertis
GarrusKind of hilarious the Xbox Series S is a better investment for gaming than PC gaming now. The Series S will beat the RX 6500 XT. Sad.
In terms of hardware sure, but everyone knows how those companies make money, subscription services and games. At the end the pricing isn't that impressive considering they're making money afterwards
Posted on Reply
#59
Chrispy_
EllertisThere is still algos like Etc and Rvn that use less than 4gb and are profitable
Profitable? Only if you ignore the purchase cost of the card and the mining rig it's going into; RVN barely pays for the electricity in most countries and that's the most profitable altcoin to mine after ETH.

In places where the electricity is ridiculously cheap, a 6600XT (which is literally twice the card of the 6500XT) makes about $0.75 a day, and costs $650 to buy. That's an ROI of 30 months and assumes (incorrectly) that you got your mining frame, PSUs, 6-slot motherboard, CPU, RAM, SSD, internet connectivity, and space to house it all for free. The profitability considers ONLY the GPU's own power consumption vs energy cost.
DavenDepends on the clock speed over everything else.
Absolutely, but we have to guess that clock speeds won't be that different to current entry-level RDNA2 offerings. The 6600XT runs at ~2.5GHz and with the smaller process node the 6500XT might break 3GHz. Unlikely but not impossible.

Let's just say it runs at 4GHz though - it won't matter because the reduced bandwidth and cache put an upper limit of how fast it can be; The 6600XT with twice the cache and twice the bandwidth is already running into cache and bandwidth issues at 1440p; It's a 1080p card.

Half the bus and half the cache means that the 6500XT may already be bandwidth-limited at just 1080p - increased clock speeds just means more wasted cycles are completed before the data arrives to process, there's not really any extra performance to be had if the GPU is starved.
Posted on Reply
#61
HD64G
Great for budget gamers and bad for miners. The combo that might allow it being sold on normal price (close or under $200).
Posted on Reply
#62
watzupken
DranzuleI'd argue Infinity Cache can make up for it. These are low-medium 1080p cards, not ultra 1080p ones. Considering that faulty dies are also a thing and that these are theoretically meant to succeed cards such as the 1050ti/1650 Super, they would be priced accordingly.
We know they won't though.
Agree at this point, but as games become more demanding, that "finite" cache is not going to be enough. It will hit your VRAM and the narrow memory bus will weigh the card down further. I rather get/ recommend a GTX 1650 Super if I get the chance.
HD64GGreat for budget gamers and bad for miners. The combo that might allow it being sold on normal price (close or under $200).
I actually don't think this is good value for anyone. You may save some money upfront, but you will quickly run into performance issue in the not so distant future. And if you look at the pricing of the RX 6600, I think this will go for at least 250 on paper, and potentially higher when it hits the market.
Posted on Reply
#63
Bomby569
HD64GGreat for budget gamers and bad for miners. The combo that might allow it being sold on normal price (close or under $200).
No, because there are no gpu's, especially news, so this will be just in short supply as any others.
Posted on Reply
#64
Ellertis
Bomby569No, because there are no gpu's, especially news, so this will be just in short supply as any others.
Considering they're very small dies, the more people will finally have a new gpu the less pressure the market will suffer
Posted on Reply
#65
Bomby569
EllertisConsidering they're very small dies, the more people will finally have a new gpu the less pressure the market will suffer
this is a brand new node, the defect rate is probably higher then any other gpu, that's not going to help one bit to get a lot of this out, and that's probably why they are doing it first with a gpu like this
Posted on Reply
#66
Ellertis
Bomby569this is a brand new node, the defect rate is probably higher then any other gpu, that's not going to help one bit to get a lot of this out, and that's probably why they are doing it first with a gpu like this
6nm existed now for a while, just as a reminder they already ship a lot of 5nm chips. I believe it has even better yields(6nm), cause I've heard the defect density is the same as on 7 nm but because the transistor density is increased the effective yield rate is even better
Posted on Reply
#67
HD64G
6nm is an enhanced 7nm process with a bit better performance/W and a bit higher density, so yields should be the same with 7nm and surely better than 5nm.
Posted on Reply
#68
RJARRRPCGP
What?! No hardware encoder?! That would be like their version of a GT "non-X" GeForce, just like the GT 640! Where I had to drop the recording resolution, just for the game to be playable!
Reminds me of my FX 8350 with GeForce GT 640 rig that I built in 2015. (with Sabertooth 990 FX R 2.0)
Posted on Reply
#69
sillyconjunkie
Switch to a new process on a low-end part? What happens to partially functioning cores? Backwards, wasteful and very unlikely.
Posted on Reply
#70
RJARRRPCGP
watzupkenThis RX 6500 and 6400 is the ultimate gimp with meagre amount of cache, and narrow memory bus. While AMD can argue that the card will run games at a specific resolution and settings, but I suspect for unoptimised games, the performance will tank. And I really don't like to when AMD is artificially limiting overclocking and thus, limiting the potential of the graphic cards.
The first part: I really hope this isn't "another ATI Radeon 9000 Pro", where only one benchmark or a few performed better than the previous gen or roughly the same as the previous gen. (the Radeon 8500 series)

And for the artificial overclock limiting, that's why I had to "jailbreak" a Radeon RX 5600 XT.
Posted on Reply
#71
Ruru
S.T.A.R.S.
GarrusKind of hilarious the Xbox Series S is a better investment for gaming than PC gaming now. The Series S will beat the RX 6500 XT. Sad.
Agree, it has a great price/performance ratio especially if you are playing with a 1080p TV. I still guess that even these lower-end cards are going to expensive (over 200 EUR/USD) with the current situation.
Posted on Reply
#72
mechtech
"Update Dec 28th: Unless we're mistaken, the SMDs near the PCIe interface in those renders seem to suggest that the GPU features a PCIe x4 interface. This should offer sufficient bandwidth for a GPU in this segment, ....."

"64-bit wide memory bus"

Almost seems contradictory in a way ;)
Posted on Reply
#73
Ellertis
mechtech"Update Dec 28th: Unless we're mistaken, the SMDs near the PCIe interface in those renders seem to suggest that the GPU features a PCIe x4 interface. This should offer sufficient bandwidth for a GPU in this segment, ....."

"64-bit wide memory bus"

Almost seems contradictory in a way ;)
How is that contradictory?
Posted on Reply
#74
Ruru
S.T.A.R.S.
mechtech"Update Dec 28th: Unless we're mistaken, the SMDs near the PCIe interface in those renders seem to suggest that the GPU features a PCIe x4 interface. This should offer sufficient bandwidth for a GPU in this segment, ....."

"64-bit wide memory bus"

Almost seems contradictory in a way ;)
By bandwith it means how many PCIe lanes it has... not the memory bus.
Posted on Reply
#75
mechtech
EllertisHow is that contradictory?
lol oooops read that as ver4, instead of x4, which would make way more sense.

was a busy morning with the kids
Posted on Reply
Add your own comment
Dec 18th, 2024 02:04 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts