Speculation: Ryzen 3000 series

[moinmoin]

The simple question to ask; has anyone else achieved 5 GHz at a smaller budget than AMD?

Imo this is kind of the wrong question. High frequency CPUs are essentially a niche within the high end desktop PC niche. The desktop PC market is dominated by x86 which has two players, Intel and AMD. For other CPU manufacturers there is essentially no need at all to chase high frequencies (crass niche exceptions like IBM notwithstanding), power efficiency is far more important.

Intel has its own foundries whereas AMD was stuck to GloFo, with the first CPUs using non-GloFo components not yet launched. GloFo was claiming their 7nm process to be prepared for regular usage at 5 GHz before they shelved it altogether. TSMC obviously has been involved with AMD even before GloFo shelved its process, so they likely had to match the same process requirements that GloFo claimed being able to fulfill, including the prospect of reaching 5GHz eventually.

Do I think 5 GHz is still a long shot? Yes. Do I think 5 GHz is impossible? No, for the above reasons. It's in TSMC's own interest that its 7nm HP process fares well at high frequencies.

 

[jpiniero]

It's not like Intel didn't go through the same thing with 14 nm... turbo and typical OC clocks were lower initially. And Apple did miss on their frequency target on the A12.

One thing is for sure, I can't see the single die products having a higher turbo clock by much. If 4.3 is the best they are going to do on a dual die product, the single die isn't going to have a 5 turbo.

The clocks are a wait and see thing I guess.

 

[Arkaign]

Well, all of that is true. It's also possible that the IPC improvement effort means they're looking for more efficiency without necessarily chasing extreme clock speeds. If the process gets really leaky or otherwise tricky beyond certain frequencies, or simply needs more refinement to reach those heights, then making some adjustments to hit higher performance per clock is another way to skin that cat. Of course it's not quite as exciting, or easy to market, but we've been down that road a few times already.

We know they have gone through this once with 7nm Vega, not really representative of a CPU shrink/evolution, but definitely at least some real world experience with some of the characteristics with TSMC 7. By now things should be a bit further along of course.

 

[DownTheSky]

There's a rumor a really high clocked 12c is in the wild.

https://www.reddit.com/r/Amd/comments/bmwxtw

What I don't know and nobody seems to bring up the question is, since the new ryzen is IO die + chiplets, how will the IO die be clocked? I assume all the talk is about chiplets clocks.

 

[Arkaign]

First, Adored has been relatively atrocious with AMD news. Second, all indications are that latency is going to be a step back with Zen2.

This doesn't mean that it will be a bad product, but gaming gains may be very limited while overall encoding/MT/compute go notably upwards.

 

[AtenRa]

First, Adored has been relatively atrocious with AMD news. Second, all indications are that latency is going to be a step back with Zen2.

This doesn't mean that it will be a bad product, but gaming gains may be very limited while overall encoding/MT/compute go notably upwards.

Even with a single 8C 16T chiplet ??

 

[DownTheSky]

First, Adored has been relatively atrocious with AMD news. Second, all indications are that latency is going to be a step back with Zen2.

This doesn't mean that it will be a bad product, but gaming gains may be very limited while overall encoding/MT/compute go notably upwards.

First, I hate to be the only one to defend the guy, but there's too much hate on him lately. He's just shared with us what he got. And what he got were the target specs of ryzen 2. That's the only thing he could have 6 months in advance. Now if those will be met or not is another story. Guess we'll see in 2 weeks. Same with navi.

Second, your "all indications" are benchmarks done on ES with horrible timings slow ram.

 

[Arkaign]

Even with a single 8C 16T chiplet ??

This, I hope, is where Ryzen 2 will really shine for gaming performance. It could be slightly awkward if Zen 16C was noticably worse for these tasks, but it could be a situation mitigated by disabling a chiplet in specific use cases perhaps through a software setting or switch mode to reboot/etc.

 

[Arkaign]

First, I hate to be the only one to defend the guy, but there's too much hate on him lately. He's just shared with us what he got. And what he got were the target specs of ryzen 2. That's the only thing he could have 6 months in advance. Now if those will be met or not is another story. Guess we'll see in 2 weeks. Same with navi.

Second, your "all indications" are benchmarks done on ES with horrible timings slow ram.

I'm not trying to crap on him, but his track record is far more than Zen.

He predicted or 'leaked' things for CES that simply didn't happen, while totally missing others entirely. There's a great summary on Reddit, but he's nearly 0% for AMD news so far.

Meanwhile, the guy with the 3.3/4.2 news has a pretty stellar record, and seems to be someone at least pretty close to OEM engineer levels of access historically.

Nothing is for sure until it hits the hands of AT/etc, so we should hope for the best regardless. I just don't want to fall into overhype. Seen that road more than a few times.

 

[thigobr]

I won't be disappointed if we don't get 5GHz (even for turbo)... But I think at least 10% from node change AMD needs to achieve. That already put things around 4.7GHz which is not bad.

And regarding the latency I agree we might see a small regression in some cases... Unless the current memory controller path is really not optimized there's no way moving the MC to another die could improve things. But that doesn't mean Zen 2 will be worse running gaming workloads. There are a lot of tricks to improve performance as data prefetch, bigger buffers, better inter-core communication latency, etc...

There are still a lot of unknowns (do we even know the CCX organization??) so hard to predict much by now.

 

[itsmydamnation]

Remember people Zen2 has 256bit execution units, So if Zen doesn't have a negative AVX offset then that base clock is providing ~4 times the throughput compared to 14nm parts, that will have a big impact on power consumption right across the package. That would suggest that under non avx2 high ILP workloads average clocks should be much higher relative to base clocks compared to the 14nm Zen parts.

Remember AMD also launched 3 fully functional dies with Zen with a range of 600mhz base and 400mhz boost. So i see it plausible that this ES chip is the equivalent of a 1700 with 1 or two higher bins and for there to be a 12core with a clock just below/at the same clock as the highest 16 core bin ( just like 1600x).

I think the first gen Ryzen is a better comparison then second gen because we will have higher variation within the 7nm process which can drive the need for more bins. Also remember the original Ryzen ES's that were found in the wild were low 3ghz for base+boost.

Do i think they will hit 5 probably not, we know what the finfet wall looks like, do i think that ES is max, nope, add ~600mhz to base and boost ( maybe a little less for base) and i think we are in the ballpark.

 

[Hans Gruber]

Ryzen 7 3700X 12 / 24 cores 4.2 / 5.0 GHz 105W $329

Can anybody confirm this speculation to be accurate or plausible? These numbers are from tomshardware.com

 

[jpiniero]

Ryzen 7 3700X 12 / 24 cores 4.2 / 5.0 GHz 105W $329

Can anybody confirm this speculation to be accurate or plausible? These numbers are from tomshardware.com

I believe Toms took that from the list that Adored made up, so not accurate for sure.

 

[Shivansps]

I believe Toms took that from the list that Adored made up, so not accurate for sure.

This, Petty much everyone did that.

 

[amd6502]

Ryzen 7 3700X 12 / 24 cores 4.2 / 5.0 GHz 105W $329

Can anybody confirm this speculation to be accurate or plausible? These numbers are from tomshardware.com

It is bogus. 12c would clearly be Ryzen 9

 

[Hans Gruber]

It is bogus. 12c would clearly be Ryzen 9

Ryzen 9 3850X 16 / 32 cores 4.3 / 5.1GHz 135W $499

This would be Ryzen 9

 

[Topweasel]

Not if they are down marketing 8c and using 16c skus for Ryzen 9. The idea that 12c would be Ryzen 9 rests mostly on the theory that there isn't a 16c chip.

 

[Hans Gruber]

Not if they are down marketing 8c and using 16c skus for Ryzen 9. The idea that 12c would be Ryzen 9 rests mostly on the theory that there isn't a 16c chip.

I am not disagreeing with you. If the speculation is true. Ryzen 9 with 16 cores would be going directly after the 9900K and the Ryzen 7 the i7 series with 12 cores and the Ryzen 5 series with 8 cores going after the i5 series. For innovation to exist there must be competition. Intel hasn't had to innovate because AMD had nothing for a decade. Now they have Ryzen.

It would be nice if all this speculation is accurate. At the same time, there would be no need for OCing because these numbers appear to be pushing the Zen2 to the limits already.

 

[amd6502]

I'm just going to stick by common sense here. Also, I don't see how anyone can take adored tv "leaks" as serious. Even just looking at the timing. When that _____ hit the www, even AMD would not know the final binning strategy, nor the frequencies that are achievable and practical.

From a marketing point of view it also makes no sense at all. 7nm is a new node (and not cheap), and it would be very simplistic to see the transition of products as a step function, from 12nm to 7nm happening in just one year. (It will take two to three IMHO.) 14/12nm is the new 28nm folks; but as you say, it's fortunate they also have a leading edge node to branch out their products on. This allows for the best of both worlds, hitting with a very wide range of products all the way to the very high end.

 

[Thunder 57]

Ryzen 7 3700X 12 / 24 cores 4.2 / 5.0 GHz 105W $329

Can anybody confirm this speculation to be accurate or plausible? These numbers are from tomshardware.com

If those specs are true I can't imagine AMD letting them go for just $329.

 

[Mockingbird]

Not if they are down marketing 8c and using 16c skus for Ryzen 9. The idea that 12c would be Ryzen 9 rests mostly on the theory that there isn't a 16c chip.

Not at all.

Core i9 has anywhere from 8 to 18 cores.

There's no reason at all why there can't be 12-core Ryzen 9 and 16-core Ryzen 9.

 

[Panino Manino]

if you watch the video kapulek posted, they cover the reason: raytracing on cpu (or rather more likely path tracing). with 3x or 4x threads per core smt the cpu has tons of free cycles, path tracing is better done on cpu than gpu and requires memory to cast all the extra rays (with the large caches on the i/o or the stacked ram it may be more efficient than dedicating ray tracing units on gpu silicon)
if you let the navi lite gpu handle the raster stuff and the cpu side handle the pathtraced lighting there may not be a need for such a powerful gpu in the next xbox.

Wouldn't this cost a fortune?

I don't believe in this, the same way I don't believe all these optimistic predictions for Zen 2 SKUs.
Give me the works prediction and I'll put my money on that.

 

[Guru]

It's confirmed, at least one Zen 2 processor will work at 4.7GHz. We'll have to wait and see if they've reached 5GHz as well, but either way I'd take 4.7GHz for a turbo clock, its a solid speed. Intel's offerings are way worse in terms of thermals, power consumption and price, so I'd take a much faster 12core or 16 core processor in everything else other than games and equal in games over Intel's overpriced cpu's.

 

[Hitman928]

Where has it been confirmed?

 

[Veradun]

If 4.3 is the best they are going to do on a dual die product, the single die isn't going to have a 5 turbo.

The clocks are a wait and see thing I guess.

They currently sell zeppelin@12nm dice that clock 4.4ghz (turbo)

I would be super-surprised if they have lower clocks on 7nm chiplets