AMD Ryzen

[Azdeus]

4 hours ago, Zoraptor said:

Same boat with me. Though I do have to say it's a sign of how good things have been on the AMD side when we are complaining about being limited to 2x cores, 1.5 node improvements, +20% IPC and +20% clocks across a single motherboard. Shame it won't be ~35% IPC improvement, but then the alternative has gone through 3 new motherboards, +400Mhz, -IPC, 2.5x cores, ++++ node and a hundred odd extra watts heat over the same period. Could definitely be worse.

The problem is now that 4000 series will (probably, gods of DDR5 permitting) be on an otherwise dead socket. Has to happen at some time, but it will put off some buyers; and the alternative may be extortionate RAM prices as phones grab all the new ddr5 and shaky gen1 memory controllers like the one I have that tries to brick my MB if I use XMP. You'd also have to expect that 3000 series procs will hold their value well too given that they'll be a terminal upgrade for a lot of people.

Yup, I haven't flashed mine though. My LEDs died on the MB so I feel very nervous about flashing, I don't want to brick the motherboard and then have to be without the computer for however long it'd take me to get a replacement. Atleast I'm lucky enough to be able to run XMP 2933 for the memory without any major stability problems, 3200 is worse and the computer sometimes won't boot and returns a OC error.. But yeah, it definetly could've been worse.

I'm not to worried about DDR5, as Keyrock says I think it'll be a year or two into DDR5 that it'll be worth going for them.

 

[AwesomeOcelot]

Rumour of better binned versions of 3600X, 3800X, 3900X, with the XT suffix. Not surprising, we've seen this a lot. Are they clearing inventory before a Zen 3 announcement, or is Zen 3 going to arrive a bit later than rumoured?

Edited May 23, 2020 by AwesomeOcelot

[Zoraptor]

Probably formalising the node improvements as a response to Intel's 10ks- later fabricated 1000 and 2000 series would hold better overclocks at lower voltages than the ones near release date. Combine that with the chiplet design and you can easily massage the performance upwards a bit on a mature node. Zen3 will almost certainly have the same staggered release schedule the other releases have had, so some models won't be expected until 2021 anyway as inventory builds in.

[Keyrock]

Ryzen 4000 Vermeer chips in B0 Stepping according to Igor's Lab. 

https://www.igorslab.de/en/ryzen-4000-vermeer-im-b0-stepping-factual-market-ripe-cezanne-with-vega-andvan-gogh-with-navi-one-step-back/

Igor is a well known leaker with a good track record. If true it means that the chips are on the cusp of mass production, as B0 Stepping signifies an essentially finished product getting tested before mass production, similar to a release candidate in software development. If they start mass production soon then Q4, as has long been rumored, seems the likely availability time window.

Edited June 29, 2020 by Keyrock

[Blodhemn]

I guess I'll choose the 4000 series for my needed upgrade. Don't care about missing out on DDR5, it's just ram anyway. Rather choose lots of DDR4 at a better price. Only question - should I go with a 6 or 8 core CPU for whichever few future games I might play? 6 cores still plenty or just better to go ahead and choose 8? Maybe just waiting for a price drop in an 8 core 3000 is the best choice...

[Keyrock]

25 minutes ago, Blodhemn said:

I guess I'll choose the 4000 series for my needed upgrade. Don't care about missing out on DDR5, it's just ram anyway. Rather choose lots of DDR4 at a better price. Only question - should I go with a 6 or 8 core CPU for whichever few future games I might play? 6 cores still plenty or just better to go ahead and choose 8? Maybe just waiting for a price drop in an 8 core 3000 is the best choice...

First we need to see what the prices will be for the different models. We still don't know if AMD will keep the core counts the same or increase them on each model. For example, we don't know if the R5 4600 (if that is what it will be called) will be 6 core/12 thread like the 3600 or perhaps 8/16. If the price difference isn't too big I would splurge for the 8/16 model simply because both the PS5 and xXxBOX Series xXx are 8/16 CPUs and developers tend to optimize their games to the lowest common denominator (i.e. the consoles).

[Blodhemn]

1 hour ago, Keyrock said:

First we need to see what the prices will be for the different models. We still don't know if AMD will keep the core counts the same or increase them on each model. For example, we don't know if the R5 4600 (if that is what it will be called) will be 6 core/12 thread like the 3600 or perhaps 8/16. If the price difference isn't too big I would splurge for the 8/16 model simply because both the PS5 and xXxBOX Series xXx are 8/16 CPUs and developers tend to optimize their games to the lowest common denominator (i.e. the consoles).

Yeah - should be another month? I'm going with an 8 core and a ton of RAM - the two things that have locked me out of some games on this ancient rig.

[Keyrock]

52 minutes ago, Blodhemn said:

Yeah - should be another month? 

Probably a little longer. I expect the Zen 3 chips will release in November.

The scuttlebutt is that Nvidia Ampere will release in September and AMD's Zen 3 CPUs and their Big Navi GPUs will release in November. This is all speculation, mind you, coming from sources with proven track records, but not from Nvidia nor AMD themselves.

Edited August 10, 2020 by Keyrock

[Blodhemn]

1 hour ago, Keyrock said:

Probably a little longer. I expect the Zen 3 chips will release in November.

The scuttlebutt is that Nvidia Ampere will release in September and AMD's Zen 3 CPUs and their Big Navi GPUs will release in November. This is all speculation, mind you, coming from sources with proven track records, but not from Nvidia nor AMD themselves.

More time to piece together the rest of the build. 

[Keyrock]

[Zoraptor]

AMD ruining my predictions of 5nm 5000series PCIe5 DDR5 AM5 by deciding to rationalise their confusing and nonsensical CPU naming scheme.

(Zen3 announcement 8 Oct with desktop being 5000 series not 4000 series; RDNA2 announcement 28 Oct for anyone wanting a summary)

Edited September 9, 2020 by Zoraptor

[AwesomeOcelot]

I hope I don't have to wait long for Zen 3. I want to start building ASAP.

[Keyrock]

6 minutes ago, AwesomeOcelot said:

I hope I don't have to wait long for Zen 3. I want to start building ASAP.

Well, you gotta wait a month for the announcement and then likely another week or two for availability. Some of the rumors I've been hearing are almost too good to be true, so I'm looking forward to finding out for real.

[Zoraptor]

Wonder how long it takes to get the chips packaged up and distributed. Huawei is out of TSMC come the end of the month, so there will be a lot of capacity suddenly free especially with 5nm continuing to come on line. They might actually have decent availability for both launches, for once.

[Keyrock]

2 minutes ago, Zoraptor said:

Wonder how long it takes to get the chips packaged up and distributed. Huawei is out of TSMC come the end of the month, so there will be a lot of capacity suddenly free especially with 5nm continuing to come on line. They might actually have decent availability for both launches, for once.

To add more good news, Samsung recently posted 980 Pro SSD specs on their site utilizing their own new memory controller, faster than the phison E16 (previously the only PCIe4 memory controller). They promptly got taken down (I guess someone jumped the gun). Not to be outdone, Phison just announced their E18 memory controller with absolutely ludicrous speeds (roughly double that of top of the line PCIe3 SSDs). These aren't available yet, but they should be very soon, so now PCIe4 actually means something for people with X570 motherboard, like me.

[Azdeus]

4 hours ago, Keyrock said:

Well, you gotta wait a month for the announcement and then likely another week or two for availability. Some of the rumors I've been hearing are almost too good to be true, so I'm looking forward to finding out for real.

I'm not holding my breath really when it comes to RDNA2s performance, Igorslab.de predicts between 3070 and 3080. What is most depressing to me is the time of their "release" and the fact it usually takes a couple of months for board partners to produce their own cards from reference material and such, and as of right now, no AIB has recieved a Navi card to build for. Which means that it's likely only AMD reference cards will be available, and.. well AMD has never produced a good reference cooler, so to get a Big Navi with good cooling I'd have to wait until 2021 to build a new rig, and I ain't got no patience for that. I've got vacation coming up for CP2077, so it seems my first nvidia card since the TNT2 is a go.

I'm kind of sad about that.

[Blodhemn]

8 hours ago, Keyrock said:

Well, you gotta wait a month for the announcement and then likely another week or two for availability. Some of the rumors I've been hearing are almost too good to be true, so I'm looking forward to finding out for real.

Tell me more about these rumors..

[Keyrock]

5 hours ago, Blodhemn said:

Tell me more about these rumors..

20% IPC increase over Zen 2. If true AMD would take the IPC lead and it might be enough to overcome Intel's clock speed advantage and take the outright gaming lead. That number seems too high, I think 10% is more reasonable. It would be awesome if true, though. I guess we'll find out soon.

Edited September 10, 2020 by Keyrock

[Zoraptor]

+15% IPC and +10% clocks plus 8 core ccxs and way better inter ccx latency were what I've heard. Zen1-> Zen2 was +15% IPC and ~+25% clockspeed, with a full node shrink, so I'd agree that rumours are probably overstating things.

(Zen2 actually has an IPC lead in desktop already, albeit in large part due to Intel being stuck on skylakes for 4 years. Intel's gaming lead is due to its frequency lead and less latency with ringbus. Intel has the slight IPC lead on mobile with icelake, but way worse efficiency, core count and clock speeds there; though tigerlake may improve some of those significantly)

[Keyrock]

I'll be curious to find out if AMD managed to further improve the Infinity Fabric, given how sensitive the Zen processors are to memory speed. Zen 2's IF is rated at 1800 MHz, making 3600 MHz DDR4 the ideal choice (DDR = double data rate, hence 1800 X 2 = 3600). I had heard legends of people overclocking Zen 2 IF to 2000 MHz or even 2100 MHz, but 1900 MHz was the number most people could reliably reach. If Zen 3 IF were to, by default, top out at 2000 MHz, for example, that in itself would be a significant boost to overall performance.

It's partly this reason why I've held off on buying RAM for my upcoming build, since I don't yet know what the sweet spot will be for Zen 3.

Edited September 10, 2020 by Keyrock

[Azdeus]

I'm going to be looking at getting a 1 ccx cpu, similar to the 3700.

[Zoraptor]

1 hour ago, Keyrock said:

I'll be curious to find out if AMD managed to further improve the Infinity Fabric, given how sensitive the Zen processors are to memory speed.

IF frequency is meant to be decoupled from RAM speed in Zen3? Grain of salt, I cannot remember where I saw that, I do remember it being described as potentially like the old Northbridge/ FSB overclocking. There are certainly IF 'dividers' added in the new AGESA per 1usmus.

 

[Blodhemn]

14 hours ago, Keyrock said:

20% IPC increase over Zen 2. If true AMD would take the IPC lead and it might be enough to overcome Intel's clock speed advantage and take the outright gaming lead. That number seems too high, I think 10% is more reasonable. It would be awesome if true, though. I guess we'll find out soon.

20% would most likely mean more power draw? 10% is more in line with standard improvement, which should keep the midrange models at 65w, I'm assuming/preferring.

[Keyrock]

17 minutes ago, Blodhemn said:

20% would most likely mean more power draw?

Not necessarily. An increase in clock speed is often accompanied by an increase in power draw and heat generated. An increase in IPC (instructions per cycle) doesn't necessitate an increase in power draw. Furthermore, advancements in power efficiency can often offset clock speed increase power draw.

Edited September 11, 2020 by Keyrock

[Zoraptor]

First performance leaks for 5000 series have been found from CPUZ and AotS, showing 20-25% overall performance improvement from 3000 series. Still grain of salt for another week though.