AMD Ryzen 3000 Series Matisse CPUs Listed With Specs

[yourusernameyour]

Do you know how CPU binning works?

I am guessing not. Here is a rough guide.
They take the best CPU's with no defects and use them in the best parts with the highest frequencies and features. Threadripper chips are the best CPU's they can make.

Then the ones that have defects are used for lesser parts with lesser frequency and lesser features like cache, smt. That is also why all of the current Ryzen 3-7 CPU's run at either 95W or 65W, excluding 2700X at 105W.

 

[grape1829]

Awesome stuff, but I'm just waiting for Haswell Xeon E3s to be sold off en masse by the large corporations so I can upgrade my G3420 for $20. If AMD's strides get them to change their setups faster, all the more power to them! I'll buy the 3000 series Ryzens when the 10000 series come out and everyone is using DDR8s.

 

[valeman2012]

@valeman2012
go to guru3d and read up a little about real life power consumption (vs max tdp, which DOES NOT equal consumption), and how much its actually impacting power bills.
not even talking about the fact that most ppl (i know) that run a +6C/12T rig dont tend to use a gt720 for gaming, so just running a Nv instead of an amd card can easily make up for (any) higher use by the cpu.




so far all info is pointing towards much higher clocks, maybe going up a bit on cores as well,
and even if there will be more cores on AM4, they could implement a gaming mode thru bios to cut cores in half and increase base/boost clocks in the process.

and if intel had anything comparable, it would be out already.
they are at least 6-8 month late, by the time they finally get there 100the gen of a 4C/ with HT out (sarcasm mode: off),
amd is "waiting" with 3rd gen R.

at the end, it doesnt really matter how ryzen performs anyway. they all make money selling to companies, not the consumer.
even if intel completely f#$%@# up stuff (which in my view they did, as they have lots of money for R&D)
it will still have +50% of the market, an enough cash to recover the next few cycles.
and they still have the option to counter with price, and the blind consumer will still buy intel, to get those "2 fps faster than amd..."

no need kinda obviousl AMD Products are power hungry in many ways even their RX 500 series crap.
There many AMD Fanboys for some reason say AMD Ryzen CPU are better than Intel 8th Gen or later cpu....cause they are more core mainly and ahead of nm process...which is not true. 100%.

I already been discuss who the sub par product - http://www.tomshardware.com/answers/id-3839185/years-amd-cpu-beat-intel-gaming.html

 

[XMEN_2012]

After all those year being underperformer, AMD is not showing that they can punch above their heights and seriosly against Intel they're so small but an ant can screw Elephant(we all learned in school) we can see int reality now.

 

[XMEN_2012]

Will they detrhone Intel, I can't say but hell yeah all gamers will rejoice as golden times are ahead in battle of kingship We "The Gamers" will be true winner.. Cheers.

 

[bigpinkdragon286]

The issue isn't cores per die - its how the die interfaces with the rest of the system. The AM4 socket only supports dual-channel memory, and you'd physically need to change the pin layout of the socket to support more (which AMD isn't likely to do since it'd break compatibility).

Threadripper is more than just extra cores - its also a different socket (TR4) with more pins and support for quad-channel memory. Search online for dual- vs. quad-channel benchmarks on TR4 - the extra channels make a significant difference in speed.

With just support for 2 channels of RAM, I don't think you can send data to/from memory quickly enough to keep 16 cores busy most of the time, so you'd likely see little/no performance improvement over say, 12 cores at the same clock speed. Its not to say AMD wouldn't sell a 16-core product, just that it'd have little performance advantage over the 12 core, even for highly threaded work.

One thing this article leaves out is, the high end, 16 core chips are rumored to have 32 MB of L3 cache. This will go a long way in addressing memory bandwidth problems.

A lesser known issue that the first two Ryzen releases are plagued with is the underwhelming front end is insufficient to keep the cores fully fed, so they end up underutilized quite regularly. Ryzen cores have more execution resources than the Skylake architecture, but unfortunately due to being starved by their front ends, kick out less IPC. If AMD fixes their front end, this should be enough to catch up most of the IPC that seems to be lacking.

If the engineers balance the chip right, and that's a big if, the dual-channel memory support shouldn't be all that big of an issue, just as it isn't the issue now. Can you even say how much bandwidth you need per core! Probably not, especially as this changes with each software suite you run on a CPU. Not all well threaded software is well written to keep the important bits in the relevant caches to avoid costly system calls, and not all poorly threaded software will end up needing to go outside the bounds of the CPU.

There is still benefit from the extra threads, having such wide execution resources goes a long way in letting the system run with much lower clocks, while still running massively inefficient software loads such as we're seeing with the current bloat that is Windows 10, or any number of user systems that have been filled with junk-ware.

People forget, are ignorant, or maybe they just prefer the shock and awe of hyperbolic statements, but just because a CPU is rated at 125 watts or higher, doesn't mean it's running at that wattage all of the time.

 

[elroy.coltof]

If this is true the TDP of the new Ryzen 7 3700X is the same as my current 2700X and I'll be able to use it in my current MB.

That's worth it not for the extra cores but for the proper support of 256 bit vector instructions alone. Even going to a Ryzen 5 3600X will give my code a major boost with double the number of parallel instructions and a higher clockspeed.

 

[caqde]

no need kinda obviousl AMD Products are power hungry in many ways even their RX 500 series crap.
There many AMD Fanboys for some reason say AMD Ryzen CPU are better than Intel 8th Gen or later cpu....cause they are more core mainly and ahead of nm process...which is not true. 100%.

I already been discuss who the sub par product - http://www.tomshardware.com/answers/id-3839185/years-amd-cpu-beat-intel-gaming.html

1) The 9900K uses a LOT more power than the Ryzen 7 2700X regardless of the 9900K being "95W" and the 2700X being "105W" chips. Looking at reviews the 105W TDP Ryzen 2700X stays around 105W in power consumption while the 95W TDP 9900K can push above 200W under a similar workload. So although AMD GPU's are power hungry the Ryzen processors are very power efficient chips in comparison and I hope and expect the 7nm Ryzen chips to be similar in this regard.

2) As for being better it depends on the usage. Ryzen in many cases can be better than an 8th gen chip. For gaming which seems to be your sole focus AMD is between 8-15% behind clock for clock in gaming (Ryzen 2600X vs i7 8700K at 4Ghz both 6core/12thread).

3) If AMD manages a ~15% IPC increase across the board in games AMD will be as fast or faster than Intel in most games with the Ryzen 3x00 series. This would likely mean that the Ryzen 7 3700X would be a 9900K killer across the board. But we will know for sure later this year. But Intel may be able to reclaim their lead around December with the Ice Lake platform but we will see.

Intel vs AMD IPC at 4Ghz (8th gen Intel vs 2nd and 1st gen AMD Ryzen)
https://www.techspot.com/article/1616-4ghz-ryzen-2nd-gen-vs-core-8th-gen/page3.html

 

[jimmysmitty]

Competition has forced significant advance by AMD in the last couple years. Intel has been able to cruise merrily along with status quo tic/toc updates. Anyone who thinks Intel won't respond to this is delusional.

The issue is that Intel's response has been with the 8th and 9th gen processors. 14nm is currently maxed out for them. 10nm will come out at the end of the year, but 1st gen 10nm is likely to perform worse than 14nm. It will likely take one or maybe even two 10nm updates to bring it to par with curent 14nm+++. Even worse, the first architecture to come out on 10nm will be Cannon Lake, which will probably be only a little better than Skylake.

In short, due to process and architecture constraints, Intel probably won't be able to seriously respond for at least 1 year, possibly 2.

Now, when Jim Keller finishes his new architecture and Intel starts producing those CPU's, AMD will probably be screwed.

So if Intels first 10nm will most likely perform worse than their 14nm how would a new 7nm part somehow perform vastly better than their older 12nm part? I mean that's a massive boost to base and boost clock speeds all while doubling the core count. Thats not a typical scenario for most anyone on a brand new process all while doubling core counts.

That is what makes this "leak" hard to believe especially if they are trying to keep the same socket, I have to imagine throwing even more cores into the same socket with the same memory channels and same communication lanes will start to bottleneck something somewhere. Unless they plan to use MCM designs and its not a single die I am not sure how they plan to clock that many cores that high and keep power and heat in check all while not bottlenecking some part of the CPU somewhere.

The 12nm process was only a tweak on Gloflos 14nm process which was built for low power not high performance. This process has an issue scaling up in frequencies as you can see with Vega and Ryzen as they are both super efficient until the clocks get to a point. This leak could be a tad optimistic but make no mistake they should see large frequency gains on TSMC's 7nm.

I can see some frequency gains. I am not saying that it would be worse. I expect it to be better just as I expect Intels 10nm to be better than even their 14nm++.

It still seems a tad odd for them to double the core count and increase the frequency by quite a bit.

And again I still see there will be some point where there will be bottlenecking with that many cores on the same socket.

 

[Aspiring techie]

Competition has forced significant advance by AMD in the last couple years. Intel has been able to cruise merrily along with status quo tic/toc updates. Anyone who thinks Intel won't respond to this is delusional.

The issue is that Intel's response has been with the 8th and 9th gen processors. 14nm is currently maxed out for them. 10nm will come out at the end of the year, but 1st gen 10nm is likely to perform worse than 14nm. It will likely take one or maybe even two 10nm updates to bring it to par with curent 14nm+++. Even worse, the first architecture to come out on 10nm will be Cannon Lake, which will probably be only a little better than Skylake.

In short, due to process and architecture constraints, Intel probably won't be able to seriously respond for at least 1 year, possibly 2.

Now, when Jim Keller finishes his new architecture and Intel starts producing those CPU's, AMD will probably be screwed.

So if Intels first 10nm will most likely perform worse than their 14nm how would a new 7nm part somehow perform vastly better than their older 12nm part? I mean that's a massive boost to base and boost clock speeds all while doubling the core count. Thats not a typical scenario for most anyone on a brand new process all while doubling core counts.

That is what makes this "leak" hard to believe especially if they are trying to keep the same socket, I have to imagine throwing even more cores into the same socket with the same memory channels and same communication lanes will start to bottleneck something somewhere. Unless they plan to use MCM designs and its not a single die I am not sure how they plan to clock that many cores that high and keep power and heat in check all while not bottlenecking some part of the CPU somewhere.

The 12nm process was only a tweak on Gloflos 14nm process which was built for low power not high performance. This process has an issue scaling up in frequencies as you can see with Vega and Ryzen as they are both super efficient until the clocks get to a point. This leak could be a tad optimistic but make no mistake they should see large frequency gains on TSMC's 7nm.

I can see some frequency gains. I am not saying that it would be worse. I expect it to be better just as I expect Intels 10nm to be better than even their 14nm++.

It still seems a tad odd for them to double the core count and increase the frequency by quite a bit.

And again I still see there will be some point where there will be bottlenecking with that many cores on the same socket.

One thing that may have not been considered, at least with the potential frequency uplift, is that Zen 2 may have been designed to handle higher frequencies. I don't know how much of an effect the new architecture may have (since they aren't doing Bulldozer 2.0), but it seems plausible that 100 or 200 MHz of these potential frequency gains may come from a more optimized architecture.

 

[caqde]

I can see some frequency gains. I am not saying that it would be worse. I expect it to be better just as I expect Intels 10nm to be better than even their 14nm++.

It still seems a tad odd for them to double the core count and increase the frequency by quite a bit.

And again I still see there will be some point where there will be bottlenecking with that many cores on the same socket.

Two issues here to deal with the clock speed one and the memory "bottleneck".

1) The clockspeed increase is due to AMD moving to 7nm but also because they are using two 8 core dies instead of a single 16 core die. Given this Ryzen 7nm is smaller than Ryzen 12/14nm (comparitively speaking for the given process). This allows for better binning of the chips as the defect rate is much smaller on a smaller chip as defects are less likely to hit them especially on the new process like 7nm.

2) Memory is not quite as big of an issue as one might think. Unless you are dealing with something memory intensive like extracting a file. Memory is rarely a bottleneck on an average consumer machine. Although it would be nice to see at what point it actually is an issue. But from what can be found online with the memory benchmarks available for single to dual channel memory the difference is mostly negligible, but given how memory is sold it isn't worth not getting a dual channel kit for the ~1-5% speed boost and sporadic ~30% boost.

 

[jimmysmitty]

I can see some frequency gains. I am not saying that it would be worse. I expect it to be better just as I expect Intels 10nm to be better than even their 14nm++.

It still seems a tad odd for them to double the core count and increase the frequency by quite a bit.

And again I still see there will be some point where there will be bottlenecking with that many cores on the same socket.

Two issues here to deal with the clock speed one and the memory "bottleneck".

1) The clockspeed increase is due to AMD moving to 7nm but also because they are using two 8 core dies instead of a single 16 core die. Given this Ryzen 7nm is smaller than Ryzen 12/14nm (comparitively speaking for the given process). This allows for better binning of the chips as the defect rate is much smaller on a smaller chip as defects are less likely to hit them especially on the new process like 7nm.

2) Memory is not quite as big of an issue as one might think. Unless you are dealing with something memory intensive like extracting a file. Memory is rarely a bottleneck on an average consumer machine. Although it would be nice to see at what point it actually is an issue. But from what can be found online with the memory benchmarks available for single to dual channel memory the difference is mostly negligible, but given how memory is sold it isn't worth not getting a dual channel kit for the ~1-5% speed boost and sporadic ~30% boost.

1. Frequency gains from a process node are normally a few 100MHz, not jumping that much. And the design using MCM is not going to stop heat and power draw. It will help but the clock speeds still seem quite a bit. As for binning, new nodes tend to be vastly less efficient to start than older more mature nodes. It takes time to get to the profitable point but I doubt the 7nm will be as good as the more mature 12/14nm at the start.

2. The only people who will benefit from 16 cores is professionals. Gamers will not benefit. Its why Intel has the HEDT platform. People have been talking about multiple cores, DX12 and Vulkan were supposed to make that a reality. It still hasn't changed anything and games still benefit from fewer high clocked cores vs many lower clocked cores. With that said, the people who this 16 core would benefit will see bottlenecks. Beyond memory there is the interconnect between the chipset and socket to communicate out to the rest of the system. Since this should be drop in to any AM4 board, the older ones that are not designed around this will also have bottlenecking issues with feeding information to more cores.

I am not saying it will be bad. Still it is just a leak and not an official release so until then this may not even be true.

 

[salgado18]

But what about CCX divisions? I've read somewhere that the 12 and 16 core models will have separate CCX, while the 8 and 6 ones won't.

 

[JamesSneed]

Competition has forced significant advance by AMD in the last couple years. Intel has been able to cruise merrily along with status quo tic/toc updates. Anyone who thinks Intel won't respond to this is delusional.

The issue is that Intel's response has been with the 8th and 9th gen processors. 14nm is currently maxed out for them. 10nm will come out at the end of the year, but 1st gen 10nm is likely to perform worse than 14nm. It will likely take one or maybe even two 10nm updates to bring it to par with curent 14nm+++. Even worse, the first architecture to come out on 10nm will be Cannon Lake, which will probably be only a little better than Skylake.

In short, due to process and architecture constraints, Intel probably won't be able to seriously respond for at least 1 year, possibly 2.

Now, when Jim Keller finishes his new architecture and Intel starts producing those CPU's, AMD will probably be screwed.

So if Intels first 10nm will most likely perform worse than their 14nm how would a new 7nm part somehow perform vastly better than their older 12nm part? I mean that's a massive boost to base and boost clock speeds all while doubling the core count. Thats not a typical scenario for most anyone on a brand new process all while doubling core counts.

That is what makes this "leak" hard to believe especially if they are trying to keep the same socket, I have to imagine throwing even more cores into the same socket with the same memory channels and same communication lanes will start to bottleneck something somewhere. Unless they plan to use MCM designs and its not a single die I am not sure how they plan to clock that many cores that high and keep power and heat in check all while not bottlenecking some part of the CPU somewhere.

The 12nm process was only a tweak on Gloflos 14nm process which was built for low power not high performance. This process has an issue scaling up in frequencies as you can see with Vega and Ryzen as they are both super efficient until the clocks get to a point. This leak could be a tad optimistic but make no mistake they should see large frequency gains on TSMC's 7nm.

I can see some frequency gains. I am not saying that it would be worse. I expect it to be better just as I expect Intels 10nm to be better than even their 14nm++.

It still seems a tad odd for them to double the core count and increase the frequency by quite a bit.

And again I still see there will be some point where there will be bottlenecking with that many cores on the same socket.

The current Ryzen CPU's are very efficient unless you OC them. They likely are getting 40% -50% power improvement from the new process partly because its a full node shrink and partly because Gloflo's 14nm process is not that great. If you look at how efficient the current chips are when not OC'ed they have headroom to scale frequency that is if the 14nm process did not hold them back ie have to add tons of power to increase frequency.

These numbers may be optimistic but they are not totally unrealistic.

 

[justin.m.beauvais]

Considering that AMD commissioned the 1000 and 2000 series chips on Global Foundries low power 14 and 12 nm processes, and are moving to a full power 7nm process, I don't think 5 GHz is a stretch at all. The hard limit on overclocking was completely based on the process, not the architecture. Since 7nm is supposed to use less power than even the 12nm low power manufacturing process I don't see a problem with these specs at all.

Also, given the chipplet design of the new Epyc CPUs, I don't see why AMD couldn't take some defective 8 core CCX units and throw them on a CPU and have a very cheap 6 core chip. Heck, we might even see a 4 core/8 thread Athlon 300GE at some point. None of this seems technically impossible, or even improbable.

The real question isn't if it is technically possible... the real question is if AMD really will do it. There isn't an obvious technical reason for them to not do it. It all comes down to how hard of a spanking they want to give Intel... and honestly, after all this time, I can't see them doing this any other way. I believe this leak. Technically it is sound enough to make sense, and if AMD can deliver it, they will.

As for people getting grumpy about AMD making such a leap in performance, you're not very old are you? This sort of thing happened REALLY often back in the old days.

 

[fry178]

@valemen2012
as as soon as you take price into account,
intel is not better.

 

[jimmysmitty]

Considering that AMD commissioned the 1000 and 2000 series chips on Global Foundries low power 14 and 12 nm processes, and are moving to a full power 7nm process, I don't think 5 GHz is a stretch at all. The hard limit on overclocking was completely based on the process, not the architecture. Since 7nm is supposed to use less power than even the 12nm low power manufacturing process I don't see a problem with these specs at all.

Also, given the chipplet design of the new Epyc CPUs, I don't see why AMD couldn't take some defective 8 core CCX units and throw them on a CPU and have a very cheap 6 core chip. Heck, we might even see a 4 core/8 thread Athlon 300GE at some point. None of this seems technically impossible, or even improbable.

The real question isn't if it is technically possible... the real question is if AMD really will do it. There isn't an obvious technical reason for them to not do it. It all comes down to how hard of a spanking they want to give Intel... and honestly, after all this time, I can't see them doing this any other way. I believe this leak. Technically it is sound enough to make sense, and if AMD can deliver it, they will.

As for people getting grumpy about AMD making such a leap in performance, you're not very old are you? This sort of thing happened REALLY often back in the old days.

In the old days we were still in our infancy in many ways such as process tech, IPC etc. The reason we don't have massive leaps anymore is because we have hit a wall in many ways. Even with this 7nm process they will quickly hit a wall to what that can give them then its on to try and get to 5nm which will be even harder than 7nm was.

Lets put it this way. Intel, the company who invests billions per year into R&D, had issues with 10nm. Be it a very ambitious 10nm that would be more dense than TSMCs 7nm. GloFlo quit on 7nm. 5nm is going to be a long way out.

 

[monsta]

Russian website, say no more

 

[yourusernameyour]

Check out Level1tech 2290WX performance solution video. Particularly after 21:30 where he mentions no issues with 16 cores on dual channel. https://www.youtube.com/watch?v=M2LOMTpCtLA

 

[justin.m.beauvais]

In the old days we were still in our infancy in many ways such as process tech, IPC etc. The reason we don't have massive leaps anymore is because we have hit a wall in many ways. Even with this 7nm process they will quickly hit a wall to what that can give them then its on to try and get to 5nm which will be even harder than 7nm was.

Lets put it this way. Intel, the company who invests billions per year into R&D, had issues with 10nm. Be it a very ambitious 10nm that would be more dense than TSMCs 7nm. GloFlo quit on 7nm. 5nm is going to be a long way out.

Oh I agree wholeheartedly. Nothing you said is wrong. I just think that in the olden days we had it a LOT rougher. Tech was moving so fast that you could order the latest and greatest and before it got to your home it wasn't the latest and greatest anymore. So, while I can relate to the unhappiness, it doesn't hold a candle next to the speed of advancement in the 90's. I don't even have those feelings of "BUT I JUST BOUGHT THIS!" anymore. All the new PC stuff is plenty capable.

Although, I do feel a little bad for everyone that ended up with a 7700K... while it is still a good CPU, a $130 chip is soon going to have more cores and threads, and probably overclock pretty well too. If AMD can deliver on this leak and IPC improvements, that $130 CPU could be faster. What a time to be alive!

 

[dorsai]

12 core 24 thread 5ghz goodness...I'm in love

 

[Soaptrail]

If this leak is true i feel bad for anyone spending boatloads of money on a HEDT. Wouldn't this also kill AMD's HEDT threadripper market?

 

[yourusernameyour]

If this leak is true i feel bad for anyone spending boatloads of money on a HEDT. Wouldn't this also kill AMD's HEDT threadripper market?

No. Desktop, hedt, and server are all clearly defined separate markets at different price points. Hedt will be fine with more ram, more ram channels, more pcie devices, probably more cores again.

 

[alextheblue]

If this leak is true i feel bad for anyone spending boatloads of money on a HEDT. Wouldn't this also kill AMD's HEDT threadripper market?

No. Desktop, hedt, and server are all clearly defined separate markets at different price points. Hedt will be fine with more ram, more ram channels, more pcie devices, probably more cores again.

Indeed. HEDT will actually get a massive boost over existing designs in the near future with Zen 2. Look at their current Threadripper setup. Tons of raw memory bandwidth and cores, but not very well balanced when it comes to memory access. When they move HEDT to a Rome-derived chiplet + I/O design, memory access will be a lot more consistent across cores. Given that Rome will work in existing boards, I'd bet Zen 2 Threadripper won't require a socket change.

Not that I'll own an HEDT system, but it should be very interesting to see in action.

 

[bigdragon]

Looks like I'll be retiring my i7-4930K this year and stepping up to one of these new Ryzen 3000 CPUs. Really impressive specs coming from AMD! I love how they're pushing the market forward. Just a couple years ago we were still stuck in Intel's quad-core stranglehold. I'm glad to see that changing.