News AMD Ryzen 3 5300G May Be the Ultimate APU for 1080p Gaming

[Admin]

An AMD Ryzen 3 5300G eBay listing shares benchmark results.

AMD Ryzen 3 5300G May Be the Ultimate APU for 1080p Gaming : Read more

 

[webtourist]

when when when they will put them on the market ???

 

[InvalidError]

Why no launch yet? The simplest reason would be too much of a backlog on other more profitable 7nm parts. There may also be some contractual obligations to catch up on Series-S/X and PS5 SoC shipments getting in the way of launching 7nm desktop APUs.

 

[digitalgriffin]

AMD is making too much money on the other parts. Plus the silicon differences between the desktop variants and mobile is minimal. And laptop demand is a high priority for AMD right now. They want to maintain their growth momentum with intel's IGP catching up.

I'm guessing laptops will get the lion share of these APUs. But I would still love to have one <$200 that's faster than my 3400G

 

[twotwotwo]

Guessing there'll be some desktop APUs because some chips don't hit the voltage/frequency curves required to be U or H chips but work when you throw enough watts at them. But those might go into a few OEM-only systems, like apparently happened with 4xxx, if they don't produce enough of them to keep them even intermittently available at retail.

And maybe the general public doesn't get more APU toys until the supply issues let up (which doesn't look imminent) and then they're still probably pricier than the 3xxx gen. To be fair about pricing, these should perform a lot better than the Zen+ APUs, so not necessarily a bad deal if they cost somewhat more.

I would really like to be wrong! I got a little ASRock box with a 3200G years ago and the modern equivalent would still be nice. Small, quiet, good bang for buck, runs Linux happily, minimal fuss.

 

[tharkis842]

This is cool, don't get me wrong, but honestly, if manufacturers aren't going to put out low end GPU's to fill the segments left open by the 3060/6700, then APU's like this need to be more robust to fill that space. Most of the focus in the GPU market seems to be at the Enthusiast level and above anymore. Currently, in my daughters PC, there's not much reason to "upgrade" from her current 1050 Ti. There isn't anything else that has such a low power but performs as well, which kinda makes me sad. These "refreshes" haven't really added much over the years.

 

[InvalidError]

if manufacturers aren't going to put out low end GPU's to fill the segments left open by the 3060/6700, then APU's like this need to be more robust to fill that space.

APUs can't compete with something like the 1050Ti which has more than double the dedicated VRAM bandwidth, they are not even on par with the GT1030. Right now, APUs would need GDDR6 or HBM2E on-package and double transistor budget allocated to the IGP to replace anything but the bottom-end dGPUs.

DDR5 will roughly double the peak bandwidth available to IGPs and we may see IGPs approaching the performance of a GTX1650. By the time DDR5 becomes affordable though, 1650-level performance may not be worth much unless you are only playing older games at 1080p60 mid-ish details.

 

[mrv_co]

It must be super frustrating for the people at AMD coming up with all these great designs and not being able to get them to market en masse due to the ongoing manufacturing bottlenecks.

 

[InvalidError]

It must be super frustrating for the people at AMD coming up with all these great designs and not being able to get them to market en masse due to the ongoing manufacturing bottlenecks.

I read somewhere that Apple has dumped some of its 7nm WSPM in favor of 5nm and 3nm with AMD and Microtek picking most of those up, so AMD may have a few thousand extra 7nm wafers per month to play with in the near future. Every little bit helps.

 

[tharkis842]

APUs can't compete with something like the 1050Ti which has more than double the dedicated VRAM bandwidth, they are not even on par with the GT1030. Right now, APUs would need GDDR6 or HBM2E on-package and double transistor budget allocated to the IGP to replace anything but the bottom-end dGPUs.

DDR5 will roughly double the peak bandwidth available to IGPs and we may see IGPs approaching the performance of a GTX1650. By the time DDR5 becomes affordable though, 1650-level performance may not be worth much unless you are only playing older games at 1080p60 mid-ish details.

I'm aware that they can't compete..that's what I'm complaining about. The tech is available and advanced enough, the power draw is low enough to make it happen for an SOC, especially an SOC that would go into a desktop or low power gaming laptop. Memory bandwidth is only one part of the equation. Core counts and frequency are just as important, GPU core counts for APU's have been stuck for years now. And like I'd mentioned, for a lower power system, to not have an upgrade path for going on 5 years now is incredibly disappointing.

The point I was making, is that the low end segment for stand alone GPU's isn't being advanced any where near the rate that top tier and mid level cards are, and since they're not advancing like the rest of the GPU market, there needs to be a viable low power option for people who either want to focus on power draw/heat/noise, or having a mini/micro build that doesn't need an add in card if nVidia/AMD aren't going to produce for that segment.

 

[InvalidError]

GPU core counts for APU's have been stuck for years now. And like I'd mentioned, for a lower power system, to not have an upgrade path for going on 5 years now is incredibly disappointing.

Shader count for IGPs cannot increase without sufficient DRAM bandwidth to support it. Affordable DDR4 that would make sense to put in a budget system has only gotten about 33% faster since DDR4 got introduced, so IGPs cannot get a whole lot faster. GDDRx on the other hand has gotten about twice as fast over the same time period.

While the tech to put GDDRx or HBM on the CPU package exists, it would require either that the motherboard provides the necessary power rails to support on-package DRAM though the socket or that the CPU does on-package voltage regulation. AFAIK, there are no signs of such considerations in the AM4 pinout so on-package DRAM likely isn't an option, neither is it on LGA1200 where most of the extra pins can easily be explained by the extra HSIO lanes and extra power/ground for PCIe 4.0.

The next decent step up in IGP performance will come when affordable DDR5 becomes readily available and gives IGPs another 30-50% more bandwidth to work with.

 

[tharkis842]

Shader count for IGPs cannot increase without sufficient DRAM bandwidth to support it. Affordable DDR4 that would make sense to put in a budget system has only gotten about 33% faster since DDR4 got introduced, so IGPs cannot get a whole lot faster. GDDRx on the other hand has gotten about twice as fast over the same time period.

While the tech to put GDDRx or HBM on the CPU package exists, it would require either that the motherboard provides the necessary power rails to support on-package DRAM though the socket or that the CPU does on-package voltage regulation. AFAIK, there are no signs of such considerations in the AM4 pinout so on-package DRAM likely isn't an option, neither is it on LGA1200 where most of the extra pins can easily be explained by the extra HSIO lanes and extra power/ground for PCIe 4.0.

The next decent step up in IGP performance will come when affordable DDR5 becomes readily available and gives IGPs another 30-50% more bandwidth to work with.

I understand what you're saying, and I appreciate the information you're providing. However, moving architectures from Vega to Navi alone would provide a pretty substantial increase in performance at a similar power envelope. Add onto that the performance increase from DDR4 to DDR5 when it's fully adopted, and we'd have an iGPU that can easily top the segment. I'm hoping this is the direction AMD is moving with iGPU's since the move to Navi hasn't happened yet, maybe they're waiting for DDR5 for a larger jump in performance?

 

[TheJoker2020]

I wonder if this delay is because of a change of manufacturer 🤔

Or will Samsung not make anything before their 6mn or 5nm production is ready and will produce AM5 stuff.? TBH, I do not know, haven't been following that story closely, anyone have any info on this.?

 

[InvalidError]

However, moving architectures from Vega to Navi alone would provide a pretty substantial increase in performance at a similar power envelope.

Maintaining the power budget does not do anything about RAM bandwidth, a Navi or RDNA2-based IGP would still only have 64GB/s of total RAM bandwidth to share between the CPU and IGP minus all of the bandwidth lost to the IGP and CPU causing active row changes all over the memory space. Caches and other structures in the IGP variant of Navi or RDNA would get neutered just like the 5300G only has 8MB of L3$ instead of 32MB per CCX for normal Zen3 parts, so Navi/RDNA IGP scaling may not be anywhere near as good as it is on discrete counterparts either.

Next consideration after that is that newer architectures have more features even when the effective number of shaders remains the same, which then translates into a larger die and higher manufacturing costs without a die shrink. The 4000-series APUs based on Zen2/Vega were 150sqmm, the 5300G likely bumps that to 170+sqmm between the Zen3 core logic and larger L1+L2+L3 caches. Upgrading the IGP on top of that would likely break 200sqmm and make it significantly more expensive.

Do you know what else AMD can make with approximately the same 200+sqmm that would be necessary to integrate a meaningfully better IGPs in Zen3 APUs? RX6600/6600XTs (236sqmm) that would likely earn AMD better net income per wafer than APUs with the added benefit of foisting most direct customer support costs and liabilities onto the AIB partners.

Having APUs nearly as big as GPU dies would be a little awkward in the middle of a global GPU shortage.

 

[tharkis842]

Maintaining the power budget does not do anything about RAM bandwidth, a Navi or RDNA2-based IGP would still only have 64GB/s of total RAM bandwidth to share between the CPU and IGP minus all of the bandwidth lost to the IGP and CPU causing active row changes all over the memory space. Caches and other structures in the IGP variant of Navi or RDNA would get neutered just like the 5300G only has 8MB of L3$ instead of 32MB per CCX for normal Zen3 parts, so Navi/RDNA IGP scaling may not be anywhere near as good as it is on discrete counterparts either.

Next consideration after that is that newer architectures have more features even when the effective number of shaders remains the same, which then translates into a larger die and higher manufacturing costs without a die shrink. The 4000-series APUs based on Zen2/Vega were 150sqmm, the 5300G likely bumps that to 170+sqmm between the Zen3 core logic and larger L1+L2+L3 caches. Upgrading the IGP on top of that would likely break 200sqmm and make it significantly more expensive.

Do you know what else AMD can make with approximately the same 200+sqmm that would be necessary to integrate a meaningfully better IGPs in Zen3 APUs? RX6600/6600XTs (236sqmm) that would likely earn AMD better net income per wafer than APUs with the added benefit of foisting most direct customer support costs and liabilities onto the AIB partners.

Having APUs nearly as big as GPU dies would be a little awkward in the middle of a global GPU shortage.

Sounds like you're making a lot of excuses for AMD, honestly. No, 6600's would not be a better use of wafer, especially when most of AMD's sales come from the low power market. Seriously, it's like you just want to be right. RAM bandwidth would "stay the same", architectural improvements have already been seen with the current gen consoles over previous gen. I guess that's a huge loss for AMD though, let's get to foisting already.

 

[InvalidError]

No, 6600's would not be a better use of wafer, especially when most of AMD's sales come from the low power market.

AMD hasn't released new retail APUs since Zen+ based 3xxxG models. The Ryzen 3100 and 3300X had next to no availability beyond the first few weeks. AMD has bailed out of the low-end retail CPU segment nearly three years ago. It clearly isn't a priority market segment.

RAM bandwidth would "stay the same", architectural improvements have already been seen with the current gen consoles over previous gen. .

Sounds like you are implying that the PS5 achieved double the PS4P's GPU performance by architectural improvements alone despite the PS5 having 16GB of GDDR6 delivering 448GB/s of bandwidth vs only 8GB of GDDR5 delivering 214GB/s for the PS4P. That's an almost linear increase in GPU-power relative to VRAM bandwidth, no miraculous architectural scaling there.

 

[tharkis842]

AMD hasn't released new retail APUs since Zen+ based 3xxxG models. The Ryzen 3100 and 3300X had next to no availability beyond the first few weeks. AMD has bailed out of the low-end retail CPU segment nearly three years ago. It clearly isn't a priority market segment.


Sounds like you are implying that the PS5 achieved double the PS4P's GPU performance by architectural improvements alone despite the PS5 having 16GB of GDDR6 delivering 448GB/s of bandwidth vs only 8GB of GDDR5 delivering 214GB/s for the PS4P. That's an almost linear increase in GPU-power relative to VRAM bandwidth, no miraculous architectural scaling there.

Sounds like you shove numbers and nomenclatures at people because you assume you know everything and everyone else knows nothing. Talk about full of yourself, seriously, get over yourself. Also, making large leaps to "put words in my mouth" is a pretty tired tactic from people who've run out of "argument". People these days don't know how to have a conversation, it's all about proving who's more 'correct'. And no, they haven't abandoned the low power market, but I guess all the CPU's they're releasing and have released don't mean anything. ONLY those using iGPU's.

No one implied the new consoles achieved their performance gap from architectural achievements alone, but spin it however you want, clearly you need the ego boost.

 

[InvalidError]

No one implied the new consoles achieved their performance gap from architectural achievements alone, but spin it however you want, clearly you need the ego boost.

Most of the performance improvements wouldn't have been possible without more cost-effective access to faster VRAM and massive increase in cache sizes that wouldn't be practical on a low-cost APU.

 

[tharkis842]

Most of the performance improvements wouldn't have been possible without more cost-effective access to faster VRAM and massive increase in cache sizes that wouldn't be practical on a low-cost APU.

You clearly have no idea what you're talking about. You think you do, but you don't.

I guess those 5000 series APU's gained performance with black magic alone.

Yeesh, this guy, seriously.

 

[InvalidError]

I guess those 5000 series APU's gained performance with black magic alone.

The 5000G worked around the limited memory bandwidth by bumping cache size, which is isn't exactly cheap either.

 

[logainofhades]

The 5000G worked around the limited memory bandwidth by bumping cache size, which is isn't exactly cheap either.

They improved latency as well, with the reconfigured cache arrangement, where all the cores, in a single chiplet share the same cache pool, instead of being split up.