Speculation: Ryzen 3000 series

[Vattila]

Kyle Bennet from HardOCP corroborates the information in AdoredTV's latest video:

"There is a whole lot of reality in that video. A lot. There is a little wrong, but not a lot."

https://hardforum.com/threads/adore...d-ryzen-and-radeon-3000-series-leaks.1973015/

I take that as mounting evidence that mainstream desktop will use chiplet design. If MCM packaging is feasible in the mainstream, from a cost and capacity perspective, it makes a whole lot of sense for AMD to reuse their 7nm 8-core chiplet.

Although I've earlier opined that 8-core is more than enough for mainstream, I now think it makes sense for AMD to extend that to 16 cores, to make sure they surpass Intel's offerings. In particular, I didn't think Intel would go beyond 8 cores on 14nm, seeing how i9-9900K pushed the limits on frequency and power. But we now have firm rumours about 10-core 14nm Comet Lake, probably planned for the next annual refresh, late 2019. Also, AMD needed (and still needs) to be prepared for Intel actually getting 10nm to work, which may give Intel opportunity for further core-count increase.

Another concern of mine was memory bandwidth. But a dual-channel memory controller per 16 cores is what 64-core "Rome" is designed around, so 16-core on AM4 may be just fine as a product in the mainstream client market.

I have argued that AMD is well positioned to fight the core war with the Threadripper platform, and I still think that. However, there are cost considerations. At the same price, AMD would prefer to sell a lower-cost mainstream chip, for higher margin.

So I'm persuaded that 16-core will be the new upper limit for mainstream.

I am also persuaded by the tri-chiplet approach. Previously it concerned me that AMD would use two expensive 7nm chiplets in a mainstream socket, but considering they can use a dummy die, as they do for Threadripper now, it makes sense, especially with the opportunity to combine salvaged dies with some cores disabled. It also makes sense to be able to replace a CPU chiplet by a GPU chiplet to implement an APU.

PS. Regarding clock speeds, I know the widespread opinion is that TSMC's 7nm HPC process is far behind Intel's 14nm on performance, but I think this is pessimistic and makes little sense (both AMD and TSMC have done 5 GHz before). I will be disappointed if we don't see boost speed at least reaching 5 GHz for the top bins.

 

[ericlp]

https://www.cnbc.com/2018/12/04/int...ngrade-it-takes-forever-to-turn-the-ship.html

With intel downgraded ... Looks like AMD is stepping up to add the final nail in casket. Intel still has some life left, but... if it doesn't turn around quickly to match AMD... They won't survive.

Rumores or not, I predict AMD being the new silicon king while Intel slides further into the gutter. Just me tho, but I hope they get their act together, as nothing is more boring than watching a one sided fight, even if my stock doubles from it.

 

[Veradun]

My guess is 8c/16t monolithic CPU on a revamped 1x 8c CCX layout on 7nm and a direct shrink to 12nm for the APU.

The chiplet approach, while incredibly smart and full of upsides, has a few downsides that can be very relevant in a desktop scenario.

Threadripper will just be derived from EPYC, binning the chiplets for speed and pairing four of them with a salvaged motherchip (you just need 1/4 of a full motherchip capabilities for TR) to help improving yelds on a big part done on a stable process tech.

The APU will be on 12nm while they work on the next design on 7nm, also waiting for more capacity on the foundry side.

 

[Capt Caveman]

https://wccftech.com/amd-ryzen-3000-specs-prices-leaked-upto-16-cores-5-1ghz-on-am4/

 

[CHADBOGA]

https://www.cnbc.com/2018/12/04/int...ngrade-it-takes-forever-to-turn-the-ship.html

With intel downgraded ... Looks like AMD is stepping up to add the final nail in casket. Intel still has some life left, but... if it doesn't turn around quickly to match AMD... They won't survive.

Rumores or not, I predict AMD being the new silicon king while Intel slides further into the gutter. Just me tho, but I hope they get their act together, as nothing is more boring than watching a one sided fight, even if my stock doubles from it.

Intel fell 4.75%, whilst the new king fell 10.92%.

What a strange succession process.

 

[Vattila]

By the way, here is the summary of rumoured SKUs from the AdoredTV video:

 

[PotatoWithEarsOnSide]

Tech stocks in general have been going down in the last couple of months.
Whilst AMD might become a pretender to the throne, whilst having a stronger claim by virtue of better performing products, I very much doubt that Intel gets toppled for the simple fact that it's scale gives it a particularly strong base to fight from.
I feel that Intel only really needs to sort itself out by the time DDR5 hits the market, which may be in 2-3 years yet. The 9900k, and the likely 10900k will keep them relevant for existing Intel owners until that time.

 

[epsilon84]

By the way, here is the summary of rumoured SKUs from the AdoredTV video:

I'm actually more interested in the low end of these rumours because I personally have no use for 12C or 16C for a desktop CPU.

If these leaks are true, budget gamers can finally rejoice! A $100 CPU that will run AAA games well, plus going by the boost clocks of the higher end SKUs 5GHz overclocks will be a reality. Sweet!

I do wonder how this will affect AMDs ASPs though. How many people actually need anything more than a 6C/12T CPU these days? Is AMD cannibalising their own TR range as well?

 

[SPBHM]

20 CU for an IGP with 128bit DDR4 sounds unlikely
also feeling skeptic on those clocks.

well, I hope it's true for sure.

 

[Gideon]

Yay, looks like my speculation of a 16-core AM4 CPU (before the October Adored Epyc chiplet video) looks more likely by the minute!

I'd really want this list to be true, but the combination of pricing and clocks together seem just a bit too optimistic. At the very least, some SKUs will be a bit more expensive.

Now we just have to wait a bitt to see the actual gaming performance (latency and whatnot). If the 3300X indeed has the gaming performance of a stock 8600K, then the desktop landscape will change beyond recognition, compared to a few years back

I would still reserve some caution however, until we have more leaks (not long anyhow, should know for certain in a month).

 

[lixlax]

I still find this leak suspicios. What are the odds that AMD has already decided on concrete model numbers, pricing and clockspeeds? The launch has to be much closer than we think if this is true.

 

[Gideon]

In the light of this new info, I really hope AMD did not skimp on the memory controller deveopment. The 16 core chip would leave a lot of performance on the table, if it were limited to, say 3466 MHz DDR compared to 4266 MHz.

 

[epsilon84]

I still find this leak suspicios. What are the odds that AMD has already decided on concrete model numbers, pricing and clockspeeds? The launch has to be much closer than we think if this is true.

Yeah, I would agree on that. I also think power numbers would be higher on a 16C 4.3GHz base clock chip.

Generally with a node shrink, you can either get more cores at the same clocks, or higher clocks with the same amount of cores. You can't double the core count and increase clocks by 20% and expect the same power envelope. That or AMDs 7nm process will be most successful node shrink in modern history in terms of performance/watt increases.

 

[Vattila]

Regarding this rumoured line-up, I think it is interesting to note that all the SKUs, except for the 8-core APUs and 16-core Ryzen 9, are allegedly based on salvaged CPU chiplets — even the 8-core Ryzen 5 3600/3600X will supposedly use two CPU chiplets with 4 functional cores each.

And for 16-core Ryzen 9 3800X, which has fully functional CPU chiplets, it is interesting that it has lower clocks than 12-core Ryzen 7 3700X (4.7 vs 5.0 GHz boost), which means that these are not the top binned chiplets.

Of course, the best chiplets go into EPYC and Threadripper — as well as the Ryzen 9 3850X anniversary edition (5.1 GHz).

Which makes sense.

 

[coercitiv]

The entire "leaked" lineup looks like a gift for AMD fans made by Santa himself.

From the $99 SKU that would run every game with ease to the high-end models that double the core count and casually transform boost clocks of today into the base clocks of tomorrow, it would seem the 7nm costs and power usage are just like Santa: they don't exist.

Get your pinch of salt from Saint Nick and wait for better leaks.

PS: every wish ever made on this forum is on that list, from chiplets to 8c monolithic APUs, low cost beastly CPUs, 5Ghz+ stock clocks (hence 5Ghz overclocks), Ryzen and Threadripper fusion, early 2019 availability (implied by having naming, pricing, clocks locked in).

 

[PuchForLife]

If the lineup from adoredtv is real it would be interesting to see whether it will be using 2 6core dies or a 4-8 configuration

With the last they could push the low latency 4 core CCX as the new norm for multithreading so that 4 becomes the norm for interdependent threads in multithreaded programming so all interdelendent threads always fit within a CCX

 

[DrMrLordX]

20 CU for an IGP with 128bit DDR4 sounds unlikely
also feeling skeptic on those clocks.

well, I hope it's true for sure.

I too think the clocks are unrealistic.

That being said, a "G" designated chip might not be an APU with a monolithic die ala Raven Ridge, or even a monolithic APU chiplet + I/O die. It might be something closer to Kabylake-G with some on-package HBM2 to serve as VRAM. In that scenario, 20 CUs is not at all unrealistic (Kabylake-G has as many as 24 CUs). It also would not technically violate AMD's roadmaps that show 2020 as the launch year for 7nm APUs.

I still find this leak suspicios. What are the odds that AMD has already decided on concrete model numbers, pricing and clockspeeds? The launch has to be much closer than we think if this is true.

If the launch is going to be April 2019, then you bet they've already picked out model numbers and pricing! And I still say Q2 2019 makes the most sense as the time to launch Zen2 for the desktop. The clocks still seem a bit off though. They are too high.

In the light of this new info, I really hope AMD did not skimp on the memory controller deveopment. The 16 core chip would leave a lot of performance on the table, if it were limited to, say 3466 MHz DDR compared to 4266 MHz.

If you have been watching DRAM development, both Samsung and Hynix are toying wtih DDR4 specs far above and beyond what we're used to. Hynix has already announced LPDDR4X-4266 as a new upcoming standard; it is not outside the realm of possibility that AMD has been testing samples of DDR4-4266 or similar on newer 1y processes. I am already sitting on some (admittedly 1x) DDR4-4400 in anticipation of Matisse at least supporting DDR4-4000. Hopefully 1x b-die will still work nicely with the new memory controller.

 

[Vattila]

The entire "leaked" lineup looks like a gift for AMD fans made by Santa himself. […] Get your pinch of salt from Saint Nick and wait for better leaks.

Don't be a scrooge.

Most of the base clocks in the line-up are nothing remarkable. The only two SKUs that stand out are the 12-core 3700X, with a 4.2 GHz base clock, and the 16-core 3850X, with a 4.3 GHz base clock. The latter is a special edition SKU with cherry-picked chiplets, presumably. It also has higher TDP (135 W). Meanwhile, 3700X would probably use the best performing chiplets from salvaged bins (6-core). Its 4.2 GHz base is just +14% compared to 2700X at 3.7 GHz. With 50% increase in core count and 50% reduction in power per core, afforded by the 7nm process, that is not too fanciful within the same TDP (105 W).

With the advances in boost technology, core count is no longer a limitation for single-core boost clocks. The line-up has only two SKUs with boost clocks that touch 5 GHz, which by the way is no more than a +16% uplift compared to 2700X, and a frequency milestone which Intel achieves with its ancient Skylake/CFL on a warmed-over 14nm process.

In short, clocks are about where I would expect for a product that was planned to compete against 10nm Ice Lake.

That said, don't get hung up on the specifications — they may be preliminary, even merely estimates. The interesting thing is the mounting evidence that AMD will use chiplets in the AM4 socket, and that this means we will likely get up to 16 cores in the mainstream.

I agree we should wait for more information before concluding that AMD has achieved competitive single-thread performance.

it would be interesting to see whether it will be using 2 6core dies or a 4-8 configuration

The video claims 6+6 for 12-core, 4+4 for 8-core (except APU) and 6+dummy for 6-core (except APU), i.e. all salvaged dies.

 

[Atari2600]

If it sounds too good to be true, it probably isn't.

If we got a ~10% IPC improvement and a ~10% clock rate improvement, I'd think that'd be the height of reasonable expectations.

So ~4.5 GHz boost clock on something that is very similar (IPC wise) to the Skylake architecture on most workloads with various outliers favouring one or the other.

 

[sonoma1993]

If this lineup is true, I'll probably upgrade my current Ryzen 7 1700 setup that I have in my server to the Ryzen 7 3700.

 

[coercitiv]

Most of the base clocks in the line-up are nothing remarkable.

Many base clocks and boost clocks make little sense, including the APU boost clocks which should not be any different than their counterparts. iGPU power consumption affects base clocks, not boost.

Let's look at the 3700X though, and infer some power figures for the new Epyc based on Zen2, see how it all fits:

• based on this leak, two Zen chiplets can do 4Ghz within 130W
• that means 1 chiplet = 65W @ 4Ghz
• that leads to 1 chiplet = 12W @ 2Ghz, assuming a ~40% drop in voltage
• hence this leak tells us 64C Epyc does 2Ghz @ 100W (or 3Ghz @ 210W if we assume a ~25% drop in voltage)

So AMD manages to double the core count, make wider cores, increase frequency and still get lower power than on 14nm. Personally I think the wolf ate grandma and if it takes a scrooge to save the kids, I'll be the Deadpool of this grim winter tale.

 

[Despoiler]

I still find this leak suspicios. What are the odds that AMD has already decided on concrete model numbers, pricing and clockspeeds? The launch has to be much closer than we think if this is true.

The odds are pretty high if these are launching in the next 6 months or so. I have 5 years of technical product development under my belt and the product developers and marketing people will have begun creating the marketing campaign at around that point. Technical details are actually one of the first things to get nailed down. Don't take the leaks as absolute, but they will be ballpark correct from a product dev standpoint. You need the technical details to go forward with marketing. Then it's a lot of iterative work as you approach the targeted launch. Marketing and then internal training, for sales and influencer type people, are always the last steps of development. Also keep in mind the recent x570 mobo leak was slated for a Computex release. That's May/June. Things are looking like what I would expect from a product dev perspective.

 

[Zucker2k]

The odds are pretty high if these are launching in the next 6 months or so. I have 5 years of technical product development under my belt and the product developers and marketing people will have begun creating the marketing campaign at around that point. Technical details are actually one of the first things to get nailed down. Don't take the leaks as absolute, but they will be ballpark correct from a product dev standpoint. You need the technical details to go forward with marketing. Then it's a lot of iterative work as you approach the targeted launch. Marketing and then internal training, for sales and influencer type people, are always the last steps of development. Also keep in mind the recent x570 mobo leak was slated for a Computex release. That's May/June. Things are looking like what I would expect from a product dev perspective.

So what do you think a 16-Core 5.1GHz Ryzen 3000 is going to consume, ballpark?

 

[Despoiler]

So what do you think a 16-Core 5.1GHz Ryzen 3000 is going to consume, ballpark?

You are stating a leaked SKU as an absolute when all of the details can be ballpark. 5.1 can be the ballpark which could end up being closer or less than the other lesser core SKUs. 135w could be higher. Both could simultaneously be different values than what was leaked. Though the TDP of 135w is also possible given that TDP has a component of use case in it. These aren't Threadripper. TR is designed for professionals that would actually be using all of the cores more of the time thus the higher TDP. Ryzen is gamer focused SKU. Outside of Ashes of the Singularity I don't think there is a single game that would use that many cores. If the cores aren't used then they aren't generating heat. I see the 16c part as being marketed to gamers that stream as a side hustle or living. The ones that would be likely to build a dedicated streaming box. They wouldn't have build a dedicated box with a SKU like the top tier leak.

All that being said, the obvious answer is that Ryzen 3000 will consume the tears of Intel fanboys.

 

[Flayed]

So what do you think a 16-Core 5.1GHz Ryzen 3000 is going to consume, ballpark?

My guess is 250W.