Question Speculation: Ryzen 3000 series pricing

[bsp2020]

There has been a lot of post back and forth about what the pricing and core counts will be of Zen2 based Ryzen processors. So, I think it's time to ask people to vote for what they believe AMD's high end AM4 product stack will look like at what prices. I think there are mainly two different opinions about what the high end AM4 offering will be like. So, please, vote for your favorite.

Edit: The extra comment I put at the end of the last choice does not make sense. Please, ignore it.
"They are both too cheap. AMD will go after market share at all cost."

 

[Topweasel]

Thats more or less what I was thinking. 12-core should be more than enough for AMD in round 1 (before Intel responses) especially if clocks and IPC go up as well.

But if they're going all in then we'll certainly see a 16-core variant at ~500-600USD. The top 12c will cost more than the 2700X regardless which route they choose (expecting 350-400USD MRSP).

People keep questioning whether AMD goes all in but if they want max value for a 16c product it makes sense to launch that early maybe even before everything else. AMD probably made an extra 10's of millions with the 2-3 Million Ryzen 7's they sold its first month available by waiting to release the 6c and 4c Ryzen's. A 16c AM4 Ryzen will have no better value then they time they can have it available not only ahead of Comet Lake but even ahead of lower sku's in the product line. Release it early or don't release it at all imho.

 

[Timorous]

People keep questioning whether AMD goes all in but if they want max value for a 16c product it makes sense to launch that early maybe even before everything else. AMD probably made an extra 10's of millions with the 2-3 Million Ryzen 7's they sold its first month available by waiting to release the 6c and 4c Ryzen's. A 16c AM4 Ryzen will have no better value then they time they can have it available not only ahead of Comet Lake but even ahead of lower sku's in the product line. Release it early or don't release it at all imho.

I agree. It also puts far more pressure on Intels response because if the plan was to launch a 10c CPU then that might compete closely enough with a 12c Ryzen 3 for people to stay with Intel but a 10c CPU vs a 16c CPU on what will be a more mature platform is a much harder sell unless Intel reduces their price which impacts their margins and has a potential knock on effect for future investment.

I think while AMD has the chance to reduces Intels revenue they need to take it, it makes it harder for Intel to compete going forward and makes it easier for AMD to get improved OEM support which will offer far more profits in the long term than squeezing every penny now.

The only reason to not offer a 16c AM4 chip straight away is if 64c EPYC2 demand is high and their 8c yields are pretty low meaning they are constrained. The thing is though even with a defect rate of 0.5/sq cm they can still manufacture something like 1159 8c dies per wafer at a yield of 67.69%. That is a lot of 8c chips available even at a really low yield.

 

[Zucker2k]

I'm going to stick my neck out and say I don't think AMD is ever going to release a 16 core chip on desktop until they can pair it with DDR5. Dual channel DDR4 isn't going to cut it for the kind of monster chip most are expecting this chip to be. unless the strategy is to release a bandwidth starved chip for this generation, and sell DDR5 as the upgrade incentive in 2020 with the release of Ryzen 3000.

 

[DrMrLordX]

A 16c CPU on AM4 with DDR4-4266 will be about the same as an 8c CPU on AM4 with DDR4-2133. There is a JEDEC spec for 1y DDR4-4266, so we might actually see affordable DIMMs in that range before the launch of DDR5. I hope.

 

[Thrashard]

Saw this article pop up today. Reaching almost 6000mhz DDR4. This ball park number is where I'd figure DDR5 comes to play. Will be interesting in next few years

AMD Matisse CPUs and 5,800MHz DDR4 memory modules referenced in AIDA64. Wait… what?

https://www.pcgamesn.com/amd/amd-matisse-cpus-3rd-gen-ryzen-aida64-5800mhz-memory


>

 

[Shivansps]

I'm going to stick my neck out and say I don't think AMD is ever going to release a 16 core chip on desktop until they can pair it with DDR5. Dual channel DDR4 isn't going to cut it for the kind of monster chip most are expecting this chip to be. unless the strategy is to release a bandwidth starved chip for this generation, and sell DDR5 as the upgrade incentive in 2020 with the release of Ryzen 3000.

I suspect the I/O may be hidding a dual channel DDR5 controller, considering the dates.... But i doubt am4 has the enoght pins to make it "am5" in a similar way AMD did AM2/AM3 in the past.

 

[DrMrLordX]

Saw this article pop up today. Reaching almost 6000mhz DDR4. This ball park number is where I'd figure DDR5 comes to play. Will be interesting in next few years

That's pretty surprising. It's hard to get 1x DDR4 past DDR4-3600 for a stock bin (there seems to be a lot of these modules now, running CAS19 and selling for cheap). A lot of DDR4 is still selling in the 2400-2933 range. Intel systems still default to DDR4-2666 on DDR4 that supports it. DDR4-5800 is more than twice that speed. It also makes my "awesome" DDR4-4400 look kind of lame in comparison . . .

 

[Topweasel]

I suspect the I/O may be hidding a dual channel DDR5 controller, considering the dates.... But i doubt am4 has the enoght pins to make it "am5" in a similar way AMD did AM2/AM3 in the past.

AMD is very likely to move to LGA for future platforms considering how holding out for the sake of the OEM's has held back the feature set on AM4. I am guessing the next socket is closer in size to Intel's 2011.

 

[NostaSeronx]

DDR5 is mostly compatible with anything DDR4. Both are 288-pin 0.85mm pitch. They have not moved the notch from DDR4 to DDR5. Other than the curved connector, DDR5 can be put in DDR4, so there might be hybrid DDR4/DDR5 slots. Which have a non-curved connector(Legacy DDR4) and a curved connector(Superset DDR5).

 

[Topweasel]

DDR5 is mostly compatible with anything DDR4. Both are 288-pin 0.85mm pitch. They have not moved the notch from DDR4 to DDR5. Other than the curved connector, DDR5 can be put in DDR4, so there might be hybrid DDR4/DDR5 slots. Which have a non-curved connector(Legacy DDR4) and a curved connector(Superset DDR5).

While I guess it would be nice if they could go with a AM4+ for DDR5 in the future. I get the distinct feeling that AMD would want to keep their desktop CPU's as flexible as possible and they don't really have a pin to spare. They have implied or outright said AM4 would be supported up to DDR5, I think they said that specifically because they will want to dump the socket for a LGA one at that point and leave them open to upping things like PCIe in future IO dies. Can't do that without a few extra Pins lying around. At minimum getting the full 32 lanes would have been nice. We don't know whats currently in the IO die, but again leaving room for change in the future would be nice. In the end 4 generations on one socket would be amazing, even more so if they continue to support newer archs with older io's for the older socket.

 

[Mopetar]

The thing is though even with a defect rate of 0.5/sq cm they can still manufacture something like 1159 8c dies per wafer at a yield of 67.69%. That is a lot of 8c chips available even at a really low yield.

You must be calculating something wrong as AMD should receive that many dies. The TSMC 7nm process uses a 300mm wafer if that's where your calculation is off. The percentage of good dies wouldn't change, just the number per wafer.

I'm going to stick my neck out and say I don't think AMD is ever going to release a 16 core chip on desktop until they can pair it with DDR5. Dual channel DDR4 isn't going to cut it for the kind of monster chip most are expecting this chip to be. unless the strategy is to release a bandwidth starved chip for this generation, and sell DDR5 as the upgrade incentive in 2020 with the release of Ryzen 3000.

It really depends on the workload. If it's something that isn't particularly memory intensive, it won't make too much of a difference.

Someone with a ThreadRipper could probably do some testing using a few different applications to see how what kind of performance impact there is.

 

[Shivansps]

While I guess it would be nice if they could go with a AM4+ for DDR5 in the future. I get the distinct feeling that AMD would want to keep their desktop CPU's as flexible as possible and they don't really have a pin to spare. They have implied or outright said AM4 would be supported up to DDR5, I think they said that specifically because they will want to dump the socket for a LGA one at that point and leave them open to upping things like PCIe in future IO dies. Can't do that without a few extra Pins lying around. At minimum getting the full 32 lanes would have been nice. We don't know whats currently in the IO die, but again leaving room for change in the future would be nice. In the end 4 generations on one socket would be amazing, even more so if they continue to support newer archs with older io's for the older socket.

And tecnically if they launch AM5 for DDR5 that would be true, that dosent mean it couldt have retro-compatibility with Ryzen 3000 AM4 cpus if they have the DDR5 controllers and enoght pins. But dont get me wrong, this is unlikely. I do belive that is possible that the I/O die to have the DDR5 controller, but DDR5 support would need new package and non-compatible socket. But im petty sure it can be done with the same amount of pins since DDR5 is still has the same amount of pins.

 

[beginner99]

Saw this article pop up today. Reaching almost 6000mhz DDR4. This ball park number is where I'd figure DDR5 comes to play. Will be interesting in next few years

Before I read your post I wanted to reply about bandwidth with a 16-core and that it would be okish if AMD invested heavily into the memory controller so that it allows much higher frequencies. Maybe that is exactly what they did. And obviously this would decrease latency and help a lot with gaming.

I suspect the I/O may be hidding a dual channel DDR5 controller,

I very much doubt that. ddr5 is very much in it's infancy and and even if it releases end of 2019/early 2020 not really great for consumer platform. These ddr5 modules with 5200 mhz have absurdly high CAS latencies which would make it slower on latency level than ddr4. And latency matters more for desktop than bandwidth. I go as far as to say even ryzen4000 series in 2020 will use ddr4 while server will move to ddr5. And that is also the beauty of the chiplet design. They can still use the same chiplet on both platforms.

 

[Thrashard]

When are we going to get these Superman Crystals ? They do exist

 

[VirtualLarry]

DDR5 is mostly compatible with anything DDR4. Both are 288-pin 0.85mm pitch. They have not moved the notch from DDR4 to DDR5. Other than the curved connector, DDR5 can be put in DDR4, so there might be hybrid DDR4/DDR5 slots. Which have a non-curved connector(Legacy DDR4) and a curved connector(Superset DDR5).

I sincerely hope that DDR5 and DDR4 are not physically DIMM- and socket-compatible, especially if they run at different voltages.

I can just imagine the nightmares that we might have seen, compatibility-wise, had DDR2 and DDR3 DIMMs and sockets been physically directly-compatible, during the transition from AM2 to AM2+ to AM3 to AM3+. Sure, they might be pin-compatible, with a mixed DDR2/DDR3 controller on the CPU, but that's still no reason to make the DIMMs, that run on different voltage ranges, to be physically compatible, including the notch position.

Won't someone please think of the ease of system integrators!

Edit: In hindsight, depending on the voltage ranges of DDR5 versus DDR4, maybe it wouldn't be the end of the world. SATA 1, SATA II, and SATA6G, all share compatible ports, and you can mix-and-match devices of all three types (more or less, compatibility between SATA6G and SATA 1 might be spotty), it does bring flexibility. But I believe that the I/O voltage for all SATA ports is the same, or negotiable somehow.

 

[NostaSeronx]

I sincerely hope that DDR5 and DDR4 are not physically DIMM- and socket-compatible, especially if they run at different voltages.

They aren't DIMM compatible. However, most of DDR4<->DDR5 is the same. Making a interoperable DIMM supporting DDR4/DDR5 wouldn't be difficult.
Both DDR4/DDR5 are 288-pin for DIMMs, where as DDR3/DDR4 was 240-pin to 288-pin for DIMMs and 204-pin to 260-pin for SO-DIMMs.

If Intel does UniDIMM2, then it could be to spec for both DDR4 and DDR5 RAM support. While not breaking compatible with either.

 

[zinfamous]

They are launching the 8 core first, not a 12 core. Lisa said so in her interview.

yup, and I also don't think there will be too many x class chips in the initial release. YOu might see top end "3700x" or something, the 8 -core, and maybe the 6 core 3600 with an x version as well.

Initial release will probably be the 3600/x, 3700/x, and the lower end 4/8 and 4/4 chips (if they exist with the Ryzen name...they probably will, but I'm wondering if going forward with the new chiplet design, this gets an entirely new branding--are they making 4 core chiplets?) ...anyhoo. I think the likely (and future) 12/24 and 16/32 chips will have only one version on release.

 

[zinfamous]

I'm voting for cheap. These aren't monolithic chips.

Single die 8c => $179 pls-mns $50
Dual die 12c => $239 pls-mns $100
Dual-die 16c => $359 pls-mns $150

the error bars in your pricing are astronomical. considering exactly one data point each, that is bad, bro.

 

[Topweasel]

yup, and I also don't think there will be too many x class chips in the initial release. YOu might see top end "3700x" or something, the 8 -core, and maybe the 6 core 3600 with an x version as well.

Initial release will probably be the 3600/x, 3700/x, and the lower end 4/8 and 4/4 chips (if they exist with the Ryzen name...they probably will, but I'm wondering if going forward with the new chiplet design, this gets an entirely new branding--are they making 4 core chiplets?) ...anyhoo. I think the likely (and future) 12/24 and 16/32 chips will have only one version on release.

God no. I don't know of AMD's plans to push a 12c or 16c Ryzen 3k. But what I do know that outside launching a late cherry picked die product later AMD always launches most expensive first. They did on Ryzen 1k, TR, and TR2. They do high margin to low hoping that people will break down and get the more expensive sku before they offer the cheaper on.

 

[DrMrLordX]

the error bars in your pricing are astronomical.

Consider the source.

 

[IEC]

God no. I don't know of AMD's plans to push a 12c or 16c Ryzen 3k. But what I do know that outside launching a late cherry picked die product later AMD always launches most expensive first. They did on Ryzen 1k, TR, and TR2. They do high margin to low hoping that people will break down and get the more expensive sku before they offer the cheaper on.

Yep, I remember paying $499 for the 1800X. Which was a GREAT deal at the time it launched. Now? Not so much

 

[Mopetar]

God no. I don't know of AMD's plans to push a 12c or 16c Ryzen 3k. But what I do know that outside launching a late cherry picked die product later AMD always launches most expensive first. They did on Ryzen 1k, TR, and TR2. They do high margin to low hoping that people will break down and get the more expensive sku before they offer the cheaper on.

I wonder if the economics change as a result of the move to a chiplet strategy though. Depending on what prices they want to charge and the supply of IO dies, it's clear that an 8-core chiplet can be sold by itself or paired with another 8-core chiplet. It's not hard to arrive at some figures where AMD makes more money selling two CPUs with one chiplet apiece than selling 1 CPU that contains two chiplets.

Those same chiplets could meanwhile be used in Rome or ThreadRipper CPUs as well, so if there's a large amount of demand in that segment, AMD may not want to use any more chiplets for the consumer market than they have to. If the rumors about the next Xbox and PlayStation using these same chiplets are true, that's even more supply constraints.

You can also construct an argument that if the prices are reasonable, AMD can get a large number of people to jump on an 8-core Ryzen 3000 CPU and then get those same customers to upgrade to a 12 or 16-core part later on as long as price is good. We know that putting the extra chiplet on only costs AMD another ~$20 or so, which means that what they charge will be shaped more by what they think the market will be willing to pay than it will be by their own costs to produce the product.

Imagine being able to get that 8-core Ryzen that can beat a 9900K that was shown at CES for $250. If half a year later, AMD offers a 16-core Ryzen 3000 for $400, would you upgrade? Maybe it would need to be closer to $350 for you to pull the trigger on an upgrade, but there's clearly a price where most people would. Historical trends make a good argument and are a safe bet, but sometimes situations change. I could see it turning out either way, but I think there's a stronger argument than previously for AMD bucking the trend.

 

[maddie]

the error bars in your pricing are astronomical. considering exactly one data point each, that is bad, bro.

How else to be correct?

 

[moinmoin]

If the rumors about the next Xbox and PlayStation using these same chiplets are true, that's even more supply constraints.

If the rumors about the next Xbox and PlayStation using these same chiplets are true there is no way the desktop and server market is coming close to make the chiplets supply constrained, like ever. Current PS4 and Xbox One so far total over 130 million units, and neither Sony nor Microsoft will plan for less for their next systems. Said rumor could never be true if TSMC wouldn't be able to deliver on that amount, and the amount required for the desktop and server market is an absolute joke in comparison.

 

[Atari2600]

I'm going to stick my neck out and say I don't think AMD is ever going to release a 16 core chip on desktop until they can pair it with DDR5. Dual channel DDR4 isn't going to cut it for the kind of monster chip most are expecting this chip to be. unless the strategy is to release a bandwidth starved chip for this generation, and sell DDR5 as the upgrade incentive in 2020 with the release of Ryzen 3000.

This fallacy has been repeated so often that it is almost accepted as truth at this point.


For ***most*** workloads - a 16core chip will ***not*** be memory starved on 2 channels of DDR4.

Look at the 2990WX. 32 cores on 4 channels of DDR4 = 8 cores/DDR channel. Same as a prospective 16C AM4. It performs perfectly well on many workloads.

Indeed, most of the problems encountered with other workloads appear to be a Microsoft induced problem, nothing fundamentally wrong with 8 cores/DDR channel.