Possible 12-Core 24-Thread Third-Gen Ryzen 'Matisse' CPU Pops up in UserBenchmark Database

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

This. Indeed, I got a 1600X then a 2600X (the latter particularly good). Like I said, I'm cautiously optimistic this time; it looks like AMD may very well have taken a lesson. I really hope so.
 

InvalidError

Titan
Moderator

While whatever Ryzen 3600 will ultimately be would certainly be faster, it wouldn't necessarily be $600+ faster (CPU+MoBo+RAM) to me. Twice as fast is the low watermark for me to be remotely interested in upgrading. Two cores, SMT and a 30% increase in IPC barely gets over that bar.
 
While whatever Ryzen 3600 will ultimately be would certainly be faster, it wouldn't necessarily be $600+ faster (CPU+MoBo+RAM) to me. Twice as fast is the low watermark for me to be remotely interested in upgrading. Two cores, SMT and a 30% increase in IPC barely gets over that bar.
"Twice as fast"! That's gonna cost you either time or money.
 

Giroro

Splendid
I don't think a 16C32T Ryzen processor is going to end up making sense in their product stack. In order to make the effort worth it, a processor with 2 good 8C16T dies on it would have to sell for, at a minimum, double what they can charge for a processor with 8C16T. Otherwise it would make more sense for them to sell those dies as 2 8C16T processors instead.
So lets say performance improvements are as good as everyone hopes, so they price their 8C16T processor at $300 to undercut the similarly-performing 8C8T core-i7s.
That would push pricing for a 16C32T processor into the $600 range, or roughly an entry level threadripper. AMD won't want to eat into their higher-margin HEDT market. like that
 

Krazie_Ivan

Honorable
Aug 22, 2012
102
0
10,680


eh, with that line of reasoning nobody should have been excited for Sandy Bridge after 6+ years of Intel far behind with P3/P4 flops (which i might add, cost more while being behind in every bench Intel didn't buy-off). but Sandy was a leap, and deserved credit, just as Ryzen has been so-far.

and really, the $120 i paid for my FX8320 on sale less than a year after launch was a bargain. it was slower than an i5 in many games, but not by half ...like the price... and it did pretty good in the apps that could use it all. the "Faildozer" label is quite a narrow view. it wasn't great, but wasn't priced delusionally as-though it was... another example where AMD/Intel diverge hard, historically.
 

nobspls

Reputable
Mar 14, 2018
902
12
5,415
That horrid 116 single core score is very disappointing. Might as well stick with the $100 R% 1600 and get the same gaming performance. Which means Intel still gets to get away with bloody murder on overpriced CPUs for top-end gaming builds.

BTW here is my R5 1600 getting similar single core score and much better quad core score.
https://www.userbenchmark.com/UserRun/12406130

So why can't AMD make real gains in IPC? I want to see single core cores hit 150 to be on par with the 8700k.
 

InvalidError

Titan
Moderator

If you genuinely need 16C32T for serious everyday Photoshop, Premiere, Handbrake, etc. use, chances are you also need quad-channel memory to mitigate memory bottlenecks as many of those will severely bottleneck on memory bandwidth. 16C32T on AM4 also doesn't give you any of the extra IO you'd get from ThreadRipper for full-speed multi-GPU GPGPU and whatever else.

Cost-wise, adding a second chiplet doesn't add anything to substrate cost since AMD will likely use the same one for all AM4 Zen 2 CPUs and possibly even APUs, the IO die will also be the same so no change there, assembly tooling is the same apart from placing a second chiplet on the package, testing and packaging tooling should be practically identical too. So I doubt adding the second chiplet will add more than 50% to total cost, very far from doubling manufacturing cost. Retail prices are a different story with manufacturers typically wanting larger profit margins on their premium parts.
 


Now, clock your CPU at 3.4 GHz with a 3.6 GHz Turbo and post those single core scores. Those similar scores from your CPU are from a mature design as it is a released processor, overclocked no less. Your single core scores at 3.9 GHz match this CPU's single core score at 3.6 GHz assuming the turbo is operating properly on the engineering sample, so if you want an apples to apples comparison, clock your CPU at exactly the same and see what happens... also please post the results because I REALLY want to see it. It will give us a much closer look at what sorts of gains AMD has made.
 

joeblowsmynose

Distinguished



Think of it this way: Intel's 16 core has an MSRP of $1700. An 8 core 9900k retails at ~$600. AMD showed that its Zen2 8 core can be faster than Intel's 8 core at most likely much lower clocks (I highly, highly, doubt AMD's CES eng sample had clocks equal to the Intel part). So we might reasonably assume that AMDs 16 core part will be faster than Intel's 16 core HEDT part.

Intel's slower 16 core for $1700 or AMDs faster 16 core for $600 -- ~%70 less money for more performance. Which 16 core processor would you buy?

Note I made a lot of assumptions there, but to think that 16 core CPU @ $600 wouldn't make sense is ridiculous.

Now that they are on 7nm - Threadripper will also be on 7nm - that means that if history repeats, Zen2 Threadrippers could well have 24-64 cores. So a measly 16 core won't necessarily be eating into the Threadripper market at all.

I know its extremely hard to believe that iterative processor improvements can come so fast, but don't compare the current AMD to Intel's near decade of milking its customers with four core parts and 5% improvements every generation. The landscape has changed, and only for the better it seems.


 

nobspls

Reputable
Mar 14, 2018
902
12
5,415


I'm not going to waste time clocking to 3.4Ghz. Been there done that. But for satisfying your curiosity, this was the R5 1600 at stock:
https://www.userbenchmark.com/UserRun/7846883

It is already doing single core at 107. Any thing you can extraoplate is will put in at around the 110 mark. But it doesn't matter. AMD hasn't made enough progress on the IPC or single core front. They sandbagged their clock frequency numbers so they can sell it and not have their butts sued off. But the reality is that Ryzen of any flavor should be at 4.0 Ghz mark at the minimum. And the gain you can get for OC to 4.0 Ghz is very incremental at best. Get to 125 if your are lucky, but still a long ways from 150, and now there is no hope even for the next Ryzen release.

More cores is for the data centers. We need more IPC and better single core performance for games! Moar cores is not the solution!

Hopefully this means that the 2700x will get cheap in 6 months time so it can save people some money. After all you can get a 1700x for $150 or less now see:
https://www.microcenter.com/product/485473/ryzen-7-1700x-34-ghz-8-core-am4-boxed-processor

What did the 1700x go for at the nominal MSRP, like $400 or some massively overpriced BS, just like the 2700x right now is ridiculously overpriced. But none of this helps with building a top-end gaming machine.
 

joeblowsmynose

Distinguished


"So anything that can be extrapolated will put it around the 110 mark" - so what you are saying is that if 3.4 isn't final frequency for this part, and let's 3.8 or 4.0 is, that would also extrapolate to a 110 point score?

Since the launch seems to be more than 5 months out yet and much tuning can be done at that time, no matter what tuning they do from now until then, it can only score a 110?

Let's assume the sample found in their database, was running 2400 or 2666 (2666 would be most likely), and, reasonably, a person plugs some 3200mghz ram in there it would still only get 110?

I think your "statement of fact" there may not very logical or well thought out ...

Also note that userbench lumps all CPU score together for averages and makes no distinction between OCd and non OCd parts - and even considering this, if you look carefully, the average single point score an a 1600 is a 103 -- a fair bit lower than the 107 one you found.

But let me give you the benefit of the doubt here - let's assume AMD won't tune clocks further, let's assume they won't tune the architecture further, let's assume no more optimizations for 5 months, and let's assume one is somehow forced to run it with slow ram. Completely impossible, but let's go with your view for now.

What do you need this massive improvement in single core scores for? Are you going to tell me that when you build your gaming rig you purposely bottleneck the CPU - rendering CPU game performance as the most important factor? Or are you intelligent enough to know that in real world scenarios bottlenecking the CPU is a brain dead maneuver in real life, and the only disparity of any major consequence is had by difference in GPU? You do know that, right?

Youre not going to tell me that when you read the CPU gaming benchmarks in reviews that you don't know that the results you see are what happens only when you bottleneck the CPU and in real world scenarios (running a resolution and quality settings accordingly to the GPU power) there is almost zero difference between CPUs in gaming performance?

Do you know that a pentium with a 1050ti gets almost the exact gaming performance as an i7-8700k with a 1050ti? You do realize this and haven't fallen for the "bottleneck CPU indications are real life gaming scores" trick have you?

I'll assume that your intelligent enough to know all this. Do you use your CPU to play superpi all day something?

If you want faster game performance, you buy an AMD cpu and the money you saved to buy you a better GPU. This is how improving gaming performance is, and has always been done.

Sorry you bought into the whole bottlenecked CPU gaming results as reflecting real world scenarios ... I blame irresponsible product reviewers for this perception mess ...
 

Giroro

Splendid


You're illustrating my point, though. Why would AMD want to make and sell a Ryzen processor that undercuts their HEDT market and lowers profit margins like that? It makes sense from a customer's perspective, but so would AMD charging $500 or $300 of free, I would totally be in favor of that. Therese a difference between being competitive to grab market share, and just giving money away. My thesis is that AMD won't make a 16C/32T, because it would be less profitable and that's not how premium products work.
But, AMD needs to sell off their faulty cores somehow, and I'm sure they would much rather continue charging you $650 for 4 faulty 4 core dies than $600 for two flawless dies that will ultimately by far more profitable if sold separately, or for that matter put into the high-end Threadripper or 64-core epyc. Especially when, on top of that, Threadripper 3 is going to come later so the market needs to continue to support threadripper 2 for most of this year. AMD would be making a mistake if their top end Ryzen 3000 outperformed low-end Threadripper at heavily threaded tasks (at least at launch).
I'm saying this based on the assumption that AMD goes back to the "one die for everything" philosophy of Zen1.
I think AMD increasing the core counts in Threadripper3 are nearly zero, even though the next threadripper will almost certainly be built on the Epyc substrate again and will have the space to support 64 cores. Server processors are by far the most profitable. Why would somebody buy a ~$3500 Epyc over a ~$1800 Threadripper if they were essentially the exact same processor?
What is more likely is Threadripper uses 4 dies with ranging from 16-32 cores and epyc uses all 8, with the line ranging from 32-64
 

joeblowsmynose

Distinguished


Zen+ Threadripper ALREADY offers 32 cores ... same maximum as Epyc. So like I said ... if history repeats itself, the new Threadrippers will likely match in core counts with Epyc -- same like it is right now with zen+. That means threadripper can easily have 24 to 64 cores - which is a long distance away from 16 cores ... did you even read my post before responding?

Let's assume that You're right (but I don't know why you would be) and 32 cores is still the threadripper max core count and it goes only from 16 to 32 ...

Let's look at Intels lineup -- Hmmm freshly launched 9900k mainstream socket - 8 cores, the also freshly launched 9800x, also 8 cores ...

Let's look at last gen Ryzen ... 1800x - 8 cores, Threadripper 1900x, also 8 cores ...

Let's consider that Threadripper has 64 PICe lanes and Ryzen doesn't. Let's consider that Threadripper has quad channel memory and Ryzen doesn't. There is massive amounts of differentiation between the platforms.

I find you argument a bit silly. There's far more to a PC platform than just cores.
 
Threadripper with zen2 will negate most of the current issues with the two dies with direct access and two without. My guess is the full Eypic IO die with latency and bandwidth somewhere in the middle of the current direct memory access and remote access. Hopefully the Infinity Fabric is upgraded to pci-e 4 cause that will help with bandwith.
 

nobspls

Reputable
Mar 14, 2018
902
12
5,415


So many words for not a whole lot value. But this tells me you don't understand one lick what a gaming build should be and what dedicated gamer wants.

If anyone is spending top dollars for RTX2080 (ti or not) they would have be utterly stupid to gimp it with a Ryzen CPU. The money you save there isn't worth crap. The extra cores that help with blender and handbrake is not worth crap either. Top FPS even if it is just 5% more or 10 more FPS, that is what is worth going on in with the dollars for. Ideally you want 4k at 144hz G-synced no screen tearing, but reality is just doing 1440p at 144hz would challenging. On the other hand, AMD sadly disappoints, and haven't been doing jack for gamers since their socket 939 days.

If the goal is to have a budget build, then Ryzen is great, and it better be cheap, like a $100 R5 1600 cheap and it would go perfect with a GTX 1060 class GPU. The little bumps they can get with Ryzen will make them good enough for the RTX2060, but that is about it.
 

Krazie_Ivan

Honorable
Aug 22, 2012
102
0
10,680


the higher the Res, the less effect the performance gap between CPUs matters. the GPU becomes the bottleneck, and even stock 2600 vs 9900k OK'd to 5Ghz will show little gap at 4k+. try an 8k Res and watch an FX chip run right along with them. if you want 265hz on a 1080p monitor, then your points have some validity.
 

nobspls

Reputable
Mar 14, 2018
902
12
5,415


That gap might be small, but that gap makes a big difference, why is that the 2700x can't even crack the top 100 here on 3DMark:
https://www.3dmark.com/hall-of-fame-2/port+royal+3dmark+score+performance+preset/version+1.0

There are more than a few 8700K and 9900K on there. When people put that much money into their GPU, and then just to gimp them and fall short is just a real shame, even for a little gap. It makes no sense. Trying to save a few bucks and fall short and get mediocre results is just missing the goal, like missing the game winning field goal.

 

nobspls

Reputable
Mar 14, 2018
902
12
5,415


You can't save a few buck when the new Ryzen's are always massively overpriced. They need to be like the $100 R5 1600. That is real value. See:
https://www.microcenter.com/product/478826/ryzen-5-1600-32ghz-6-core-am4-boxed-processor-with-wraith-spire-cooler

The 2700x doesn't even come close providing 3x the performance of the 1600 but they sure price it as if it did. You can expect the new third-gen to be overpriced the same way. This is the AMD's mode of operation, which means you will lose a lot of value in 6 to 9 months time. Just like how you can get a 1700x for $150 now (vs when they starred at $450):
https://www.microcenter.com/product/485473/ryzen-7-1700x-34-ghz-8-core-am4-boxed-processor

Which is to say given the 3rd gen is only going be very mediocre improvements on the 2nd gen, you will get the most value by buying the previous gen if you choose to go budget for Ryzen.

I want AMD to be back like they were in the socket 939 days when the beat Intel on the gaming benches and cost less than Intel, and that is why the 3rd gen Ryzen is so disappointing.
 

hannibal

Distinguished
Heh... in every product you get better deal by buying previous generation! Look smart phones... well maybe not by buying Nvidia GPUs at this moment, but in everything else.
That is because shops sells those out at less Profit.
But if you want to have those improvements you have to buy the full price of the new product. And for some time amd has been much better deal than Intel when building a Gaming rig, because the price difference... but ofcourse if you can get used Intel k9990 for cheap like 150$... go for it...
 

InvalidError

Titan
Moderator

You can't generalize based exclusively on Microcenter's unusual extreme in-store-only discounts.
https://pcpartpicker.com/product/jLF48d/amd-ryzen-5-2600-34ghz-6-core-processor-yd2600bbafbox
https://pcpartpicker.com/product/mV98TW/amd-ryzen-5-1600-32ghz-6-core-processor-yd1600bbaebox

For anyone who doesn't live close enough to walk into a Microcenter without putting $20+ in the gas tank, the Ryzen 1600 and 2600 can be had for $160 vs $165 respectively, practically the same price.
 


You do realize that IPC means Instructions Per Clock (or Cycle depending on who you talk to) right? You DO realize that the single core score and IPC are not the same thing right? You do realize that the only way to do a fair IPC measurement is at the same clock speed, right? You do realize that Intel chips currently turbo WAY higher than AMD ones do, right? You do realize that the single core score only uses a single core and tests single core speed at turbo clocks, right? I mean realizing all of that and looking at Intel's turbo speeds it must be plainly obvious.

Right now Intel can clock nearly a full GHz over Ryzen (more if overclocked). That is how they get all their much higher single core speed in UserBenchmark. Those single core tests are running those numbers at 3.9 GHz on your CPU and at 5 GHz on the 9900K or 4.7 GHz on the 8700K. Remember what we realized earlier? That the only fair way to measure IPC is to measure them at the same clock speed? Remember where we realized that the SC measurement is not a measure of IPC? THIS is why. When you clock these CPUs identically, AMD is much closer that it seems. Only a few points behind... and Zen 2 should be a few points better. Now if only AMD could get those clock speeds up.

Guess what! Great news everyone. AMD can hit those clocks with Zen 2. How do I know this? Because they went core for core against the 9900K in their CES demo... which would have been impossible at even Ryzen 7 2700x's overclocked speeds. Which means that they HAD to have improved clock speeds. On top of that the node shrinks and new optimizations *almost always result in improved IPC. Well, them having to improve clocks and also the fact that they moved from GloFo's low power node to a full power 7nm at TSMC... clock speeds were going to improve. Current AMD clock speeds are not a result of a bad design... they are a result of the node they are created on. AMD was still playing catch up at the time and decided they needed to lower power consumption to be more competitive so they fabricated their CPUs on a low power node to make them more appealing to enterprise server folks... and that is where the hard clock limit came from. But no more! Zen 2 is Zen Unleashed.

So, scoff all you will at current Ryzen CPUs and how "horrible" they are for gaming (when realistically they are usually only slightly behind and on screen performance is often indistinguishable). Zen 2 should be exactly what AMD needs to finally match or exceed Intel's performance in UserBenchmark... or you know what? Maybe a lot of benchmarks, oh and some games.

*Bulldozer is why I have to say almost... AMD is never getting away from that are they?
 

joeblowsmynose

Distinguished


Above is a quick summary why no can, nor should, take your views seriously.

If that's your opinion, you are welcome to have it, but how you come to form opinions is obviously now in question.
 

nobspls

Reputable
Mar 14, 2018
902
12
5,415


Good job trying to lie. It is obvious from the posts here that this was written:


Making up false statements is all you can do when you have no valid points, so you resort for false personal attacks.

And when I asked if anyone can a better value here:
http://www.tomshardware.com/answers/id-3871506/beat-1600-100-price-performance-gaming.html

There were no alternatives provided. So explain to how the 2700x is NOT massively overpriced compared to a R5 1600? The dollar per fps, per watt etc, just show how badly overpriced the 2700x is .