News AMD Unveils 5 Third-Gen Ryzen CPUs, Including 12-Core Flagship

[PaulAlcorn]

AMD's Third-gen Ryzen processors span from six cores and 12-threads up to 12 cores and 24-threads and bring along a 15% improvement in IPC.

AMD Unveils 5 Third-Gen Ryzen CPUs, Including 12-Core Flagship : Read more

 

[NightHawkRMX]

I just finished watching this. I like some of what i see. I figured 16 core wouldnt come till later.
I wish that the pricing was cheaper, but performance seems on point. Clocks arent high, but thanks to tha ipc, performance seems good.
Dayng like 70mb of cache on the 3900x...

 

[NightHawkRMX]

Where did the 6 core ryzen 5 data come from?

 

[Walter Smith]

The real focus here, performance-wise, falls to the Core i9-9900K versus the Ryzen 9 23900X.
Should be Ryzen 9 3900X.

Also, in the price per thread chart, the 9900K is divided by 8 not 16....math a lil but fuzzy on that.

Great article tho!!!! Can't wait for some early benchmarks!!

waltsmith

 

[NightHawkRMX]

Im curious if any ryzen 3000 skew can hit 5ghz overclocked on air or water cooling.

 

[nitrium]

With those boost clocks, Intel will presumably still easily beat these in single thread performance, which is a bit of shame. Also those prices are higher than I would have liked. Also is there really a difference (other than a factory OC and/or binning) between the 3700X and 3800X? And why does it take 40W more to get a <10% all core OC?

 

[hannibal]

Im curious if any ryzen 3000 skew can hit 5ghz overclocked on air or water cooling.

Most likely 4.5 4.6 is near the maximum that 3000 can achieve untill the temperature and power usage goes sky high. Still a good upgrade! The sweet point seems to be 4.4GHz. Or Little bit slower. I would Expect that 5GHZ may be possible 2 to 3 years if this production node matures save way as previous.

 

[jpe1701]

With those boost clocks, Intel will presumably still easily beat these in single thread performance, which is a bit of shame. Also those prices are higher than I would have liked. Also is there really a difference (other than a factory OC and/or binning) between the 3700X and 3800X? And why does it take 40W more to get a <10% all core OC?

IDK if that's necessarily the case with the increase in the ipc. It will be interesting to see the reviews. The 3800x looked neck and neck with the 9900k in the demo but of course it was controlled demo.

 

[rhysiam]

That gaming head to head in Pubg was really impressive. That has always been a rough game for AMD. The 2700x loses to the 8700K by ~15% at 1080p. IF that is indicative of AMD's improvements in their worst games (given this was a hand picked, controlled demo), then Ryzen 3000 could be super competent gaming CPUs. No doubt there will still be games where it can't quite match the 9900K, but if the worst-case scenarios are 10-15% rather than the 20-30% we see at present, then the 9900K (or 9900KS) becomes even more of a niche product.

I still have a few areas of interest though: not one mention of XFR I believe? So are we really looking at a hard ~4.6Ghz limit for Zen2 cores, or will we see XFR on top of this? Is AMD still holding a few cards close to their chest while they finalise XFR with the mobo makers? A good number of leaks had peak clock speeds approaching 5Ghz, so is that adding XFR, or were those leaks just overly optimistic?

Also: What about the gaming performance on the dual chiplet Ryzen 9s? Lisa said they would be "no compromise" CPUs, but they never showed their gaming performance head to head with a mainstream Intel or a Ryzen 7. Splitting CPU cores across multiple dies will increase worst-case core to core latency. Have AMD really managed to negate any performance impact from this in gaming workloads to truly create a "no compromise" CPU? Or will we see certain workloads be the Achilles heel of Rzyen 9, just like we have for Threadripper?

I'm still really excited about the announcement today. Can't wait to see independent testing.

 

[rhysiam]

With those boost clocks, Intel will presumably still easily beat these in single thread performance, which is a bit of shame. Also those prices are higher than I would have liked. Also is there really a difference (other than a factory OC and/or binning) between the 3700X and 3800X? And why does it take 40W more to get a <10% all core OC?

It depends what you mean by "single thread performance". AMD's slides have the Ryzen 7 3800x beating the 9900K in single-thread Cinebench despite the boost frequency deficit: https://images.anandtech.com/doci/14407/COMPUTEX_KEYNOTE_DRAFT_FOR_PREBRIEF.26.05.19-page-024.jpg
Cinebench is a consistent and repeatable benchmark. Anyone can download and run it for free. If AMD are somehow fudging those numbers they'll get exposed pretty quickly. I'd be surprised if they're doing that.

Cinebench is also, of course, one of AMD's strongest benchmarks relative to the competition which is why they use it all the time. So we can expect that other lightly threaded workloads won't look nearly as rosy for AMD. Still, with IPC and clockspeed improvements I don't expect there will be much in it with the exception of workloads that are impacted by latency issues inherent to the multi-die design.

RE 65W to 105W Ryzen for the Ryzen 7s, I noticed that too. The interesting thing here is that the 12 Core Ryzen 9 keeps the 105W TDP despite 4 extra cores and double the cache, all while dropping only 100mhz on the base clock compared to the 3800X. I wonder if AMD are trying to avoid too many different TDP brackets? I know they have 95W Ryzen 5s, but perhaps for the Ryzen 7 & 9 classes they want to stick to 65W, 105W and a future 125-135W rating for the 16 core premium parts. Just a thought. In any case, TDP doesn't necessarily relate to power draw, so I'd be very surprised if the R7 3800X and R9 3900X draw the same power under peak load.

 

[cryoburner]

AMD fired off the first salvo of its third-gen Ryzen lineup today at Computex 2019 as it revealed five new CPU models that range from Ryzen 5 with six cores and 12-threads for $399 up to a Ryzen 9 3900X with 12 cores and 24 threads for $499.

You guys might want to fix that first line in the article there. The 6-core part is obviously not the same same price as their top 8-core part. >_>

 

[Ajvy]

In the high segment, the prices are fine, they are really competitive with the Intel. In the mid-high range, the prices are the same as Intel's and I don't care about the number of cores, the clock speed is not on par with Intel.

 

[R_1]

typos please delete this when corrected.
"falls to the Core i9-9900K versus the Ryzen 9 23900X " Ryzen 9 3900X
"The disparity in pricing is apparent here, but the -9920X does " i9-9920X or 2920X? omitted the i9 again 4 lines down
"You'll also notice that the Ryzen 5 2600 has a 35W lower TDP than its predecessor. " the 1600, 2600 and 3600 are 65W chips. chart shows the 2600 having a 95W when the 2600X is the 95W chip.

 

[King Dranzer]

Well crazy launch from AMD. R7 3700X(OC) on par with Intel i9-9900K(OC) for much lower price is what we needed. R7 3700X launch price $330 but after a month or two probably around $280 will be even crazier deal. R7 3800X for around $400 at launch is also not bad but comparatively R7 3700X is a lot better deal.

R9 3900X with its crazy OC capabilities don't know why anyone will spend on Intel i9-9900K sucessor for minimal IPC improvement after its launch.

 

[NightHawkRMX]

So far the high-end ryzen skews sound great and like AMD will take the crown there, but I'm still waiting for E3 to see ryzen 3 specs since the CPUs listed cost more than I want to spend.
I'm betting 4c/8t, 3.2 base, 4.1 GHz turbo, 45 watts for Ryzen 3 3200. I hope since AMD already dominates the budget Intel offerings, AMD doesn't neglect to update performance a bit on their low end skews.

I'm watching ryzen 5 1600 prices. You can get one brand new for $107.99 on ebay. That's 6c/12t 3.6 turbo with an included Wraith Spire for just $107.99. With a decent cooler, 4ghz isn't that hard.

 

[hftvhftv]

It depends what you mean by "single thread performance". AMD's slides have the Ryzen 7 3800x beating the 9900K in single-thread Cinebench despite the boost frequency deficit: https://images.anandtech.com/doci/14407/COMPUTEX_KEYNOTE_DRAFT_FOR_PREBRIEF.26.05.19-page-024.jpg
Cinebench is a consistent and repeatable benchmark. Anyone can download and run it for free. If AMD are somehow fudging those numbers they'll get exposed pretty quickly. I'd be surprised if they're doing that.

Cinebench is also, of course, one of AMD's strongest benchmarks relative to the competition which is why they use it all the time. So we can expect that other lightly threaded workloads won't look nearly as rosy for AMD. Still, with IPC and clockspeed improvements I don't expect there will be much in it with the exception of workloads that are impacted by latency issues inherent to the multi-die design.

RE 65W to 105W Ryzen for the Ryzen 7s, I noticed that too. The interesting thing here is that the 12 Core Ryzen 9 keeps the 105W TDP despite 4 extra cores and double the cache, all while dropping only 100mhz on the base clock compared to the 3800X. I wonder if AMD are trying to avoid too many different TDP brackets? I know they have 95W Ryzen 5s, but perhaps for the Ryzen 7 & 9 classes they want to stick to 65W, 105W and a future 125-135W rating for the 16 core premium parts. Just a thought. In any case, TDP doesn't necessarily relate to power draw, so I'd be very surprised if the R7 3800X and R9 3900X draw the same power under peak load.

AMD rates their TDP at full boost clocks, while Intel rates their TDP for base clock speeds. A 105W AMD CPU is going to run cooler than a 95W Intel CPU that has Turbo Boost. Also, AMD has a 7nm die while Intel has 14nm, so they are able to keep the power consumption and heat lower in theory.

 

[NightHawkRMX]

Their design using TSMC's 7nm process must be pretty efficient.
Having the Ryzen 7 3700X 8C/16T 32mb with a 4.4 turbo only consuming 65w is impressive. I would like to see what stock coolers are going to be thrown on these CPUs.
I bet the upcoming 16 core will not have any stock cooler. MSI says they built their high-end X570 boards are designed to deliver 300 watts to the CPU. Gamers Nexus says when overclocked they are expecting like 300w power consumption from the 16 core.

 

[salgado18]

With those boost clocks, Intel will presumably still easily beat these in single thread performance, which is a bit of shame. Also those prices are higher than I would have liked. Also is there really a difference (other than a factory OC and/or binning) between the 3700X and 3800X? And why does it take 40W more to get a <10% all core OC?

The 3800X has more thermal headroom to stay on higher clocks. If the 3800X could do all-core 4.0, for example, the 3700X would keep at 3.6 to stay within it's TDP.

 

[Arzhur]

When comparing the Intel 9900k and AMD 3900x the article says that there is a 60% difference. However, that is using the Intel 9700k as a reference. Should the values not be normalized to the 9900k values (divide by 135%), giving 9700k = 74.1%, 9900k = 100.0% and 3900x = 144.4%? This give a 44.4% increase, not 60%.

Naming their 12 core part the 3900x, doesn't give much room for the name of a possible 16 core part. They could go with the Threadripper convention of changing the third digit (e.g. a 3950x), but that would loose the emphasis of increased core count that the rest of their Ryzen parts have.

 

[King Dranzer]

Well this time in motherboard sector ASUS will have dominance over other brands as they are claiming all the boards throughout the lineup to support 16C/32T CPU and its power draw and even more important claim of implementing additional tweak to OC memory even higher with ease throughout its lineup. For Ryzen as high we can get the memory to it will perform that good and will show noticeable performance improvement.

Till now no other board partner were bold enough to claim something like that for their lower end offerings.

 

[NightHawkRMX]

MSIs high-end X570 boards are rated for 300w power delivery to the cpu.
Expecting overclocked 16 cores.

 

[King Dranzer]

MSIs high-end X570 boards are rated for 300w power delivery to the cpu.
Expecting overclocked 16 cores.

ASUS low end boards are prepared to handle that.

 

[digitalgriffin]

Since there's no thread talking about Radeon (But it is semi related)

I was surprised to see they did away entirely with GCN for consumers. Yet AMD kept it for commercial workspaces (ie: Cloud).

My predictions were a heavily modified GCN and other tricks along with architectural changes including more ROPs and making the front end rear end compute units more "modular" in a prep for APU's and Arcturus.

I have to give AMD credit on keeping the lid tight on this one. However if it's $500 for 2070 performance they are dead in the water for that market segment. So my judgement is still out till Next Horizon.

As to the 12 Core Ryzen part, I'm a little disappointed. 4.6Ghz boost? I was expecting 4.8GHz at the very least. So many people were saying 5GHz was a reality.

Is AMD holding it in the hopper in case Intel does release something faster? If 5Ghz chiplettes were 5% of the 6 core chiplette parts, then there is no way they could meet the demand even at $750. So they are binning those and setting them aside for things like Rome? AMD has a lot of design wins, I'm sure supply of "golden" 7nm chips will soon be an issue for them. Higher clocked chips also run at extremely low voltages. Could they be waiting for Zen 2+ before boosting?

 

[Jimbojan]

Did not Intel also release it 10nm parts, we need to compare 7nm to 10nm parts to be fair(er).

 

[NightHawkRMX]

Not for desktop gaming cpus.
They are on 14nm.