Zen 4 Ryzen 6000

If anybody can wait 'til 2022... go ahead. My only problem is that I am bound by SP3 and do not intend to upgrade to SP5 or Epyc Genoa.
 
If anybody can wait 'til 2022... go ahead.
It'll be 2022 before we can even get Zen 3 purchased from shops.

I've given up on upgrading my Haswell to a Zen 3 platform, I'll probably use the budget on an OLED + next gen console, and then look at Zen 4 if it even reaches shops during 2022.
 
It'll be 2022 before we can even get Zen 3 purchased from shops.

I've given up on upgrading my Haswell to a Zen 3 platform, I'll probably use the budget on an OLED + next gen console, and then look at Zen 4 if it even reaches shops during 2022.
5600X and 5800X are super easy to get right now, and have been for months. I was passing on them usually once or twice per week in my quest to get a 5950, which I ended up getting in Dec.

the real challenge is getting a 3080 - even tried camping out at microcenter a couple times and got squat there.
 
5600X and 5800X are super easy to get right now, and have been for months. I was passing on them usually once or twice per week in my quest to get a 5950, which I ended up getting in Dec.

the real challenge is getting a 3080 - even tried camping out at microcenter a couple times and got squat there.
Both of them are bad choices for a long term platform I'm planning.

And since I'm doing the whole setup, I'll also need that nice GPU to go with it, not a 3060 or something. Which would be remotely possible after we go past H1 2021, but then it's too late. And still going to be above MSRP.
 
Both of them are bad choices for a long term platform I'm planning.

And since I'm doing the whole setup, I'll also need that nice GPU to go with it, not a 3060 or something. Which would be remotely possible after we go past H1 2021, but then it's too late. And still going to be above MSRP.
If you’re really going long term on the build, the 3090 isn’t all that hard to get. I’ve passed on them multiple times as well. I’ve seen quite a few 5950s roll through at microcenter, and have seen people getting their 5950 B&H pre orders as well. A 5950 with 3090 could probably be put together in a week of shopping, maybe less.
 
5600X and 5800X are super easy to get right now, and have been for months. I was passing on them usually once or twice per week in my quest to get a 5950, which I ended up getting in Dec.

the real challenge is getting a 3080 - even tried camping out at microcenter a couple times and got squat there.
They still haven't delivered all the 3080s that were ordered in september in my country. One I know got his frist card in october and the second late december. I got mine early december. He ordered within 5 minutes, I ordered within 30 minutes. Another guy I know ordered 4 days after launch and will probably get his card around easter. They closed for orderes late september and still haven't opened up for new ones for the 3080. The 3090 has been in stock several times though, but it is just too expensive IMO. I do think there is a decent amount of cards being shipped though cause the site of the largest etailer was having major issues with their site due to lots of people ordering. Problem is most likely miners and that a lot of people were sitting on 4 year old cards due to 2000 series being underwhelming in the price/performance ratio for owners of 1070 and higher.

Think the AM5 platform will most likely be really good. AMD has been struggling with their memory controller performance so DDR5 should be a massive improvement. The memory performance can be a big bottleneck on the 5000 series so if they manage to get that on par or better than intel then I think they will be able to get almost 10% just from that. Will be interesting to see how the demand is though as a lot of people I know just purchased 3000 or 5000 series or are looking to purchase 5000 series and they don't upgrade more than maybe every 3-4 years.
 
The Zen 3 refresh is supposed to be on AM5, as a way of testing a fairly known chip on an unknown platform ( that's what I heard anyway ).
A Zen 3 Refresh on AM5 would require a new IO die, but could potentially work with Zen3 dies, so that may work.

I'm still stuck in the mindset that new memory controller requires a new tapeout, but in reality with chiplets, AMD can pump out a new IO die and leave the existing compute elements.
 
A Zen 3 Refresh on AM5 would require a new IO die, but could potentially work with Zen3 dies, so that may work.

I'm still stuck in the mindset that new memory controller requires a new tapeout, but in reality with chiplets, AMD can pump out a new IO die and leave the existing compute elements.
Yeah, the only requisite would be that the io die would need to be compatible with both cores, but since it's a refresh anyway, they could modify the core slightly to ease compatibility.
 
Yeah, the only requisite would be that the io die would need to be compatible with both cores, but since it's a refresh anyway, they could modify the core slightly to ease compatibility.
Well the Zen 2 and Zen 3 compute cores paired with the same IO die, so there must be some sort of magic that AMD is doing to make the IO Dies and Compute cores agnostic to eachother.
 
Could really get used to this increase in IPC. Makes a huge difference in the realm of simulation games and games modded far past what legacy engines were ever intended to support.

I'm just imagining a Civ 4 game with every modded civilization in a realistic Earth map. Anyone who has tried that, even today knows just how much of a loading simulator that can be.

Or if you like grand strategy games like Stellaris. You could have an insanely sized galaxy with numerous civs.

I believe the CPU has long been a bottleneck for storage performance as well. I'd be curious if better IPC equates to fastest game load times etc...

Exciting stuff!
 
They probably can't get anymore speed inside the CCR, so the new DDR5 -lanes, which utilize the current IO-structure has to have 3-5 times per clock cycle to read the data from L3-caches and transfer the data thru IO-lanes to the DDR5. However there might be a physical limitation of how much data you can push thru one IO-lane (DDR4 3200 MB/s, but I have heard that DDR5 will go all the way up to 9600 MB/s in the later stages), so if you can't speed up the data throughput in the single IO-line, then you have to dedicate maybe 2 or even 4 IO-lines for one DDR5 -stick, so in EPYC Genoa -mobo that has 8 or even 16 memory channels to utilize all the 16 memory sticks simultaneously, might take up about 64 io-pins for the DDR5 memory traffic alone (if the memory lanes cannot be made any faster). Atleast, the 2 io-lines per stick is supported by the fact that each stick has its own memory controller that can read simultaneously from the DIMM-forward part (the first xx pins prior to gap) and from the DIMM-backend part (xx pins after the gap). However the data length is divided in one simultaneous reads of 4 bits from front and back, when earlier it was either 8 bits from the front OR back, so this could mean that the data is transferred thru separate IO-channels, because it does not make any sense to combine the two separate 4 bit striplets into a combined 8 bit, when the read/write is utilized by a different threads of the processor to do completely different things. Memory stick level separate memory controllers might also eliminate the need to have any memory controller whatsoever on the mobo, if the memory traffic from the 16 different sticks connects directly to their dedicated 32 or 64 or whatever amount of IO-pins in the processor.
 
Last edited:
I believe the CPU has long been a bottleneck for storage performance as well. I'd be curious if better IPC equates to fastest game load times etc...
Maybe a part but storage can be faster with actual CPU. Check the Optane.

1613745117009.png
 
Back
Top