Ryzen 3000 series might have more than 8 cores

[C2HWarrior]

3 minutes ago, AngryBeaver said:

So you are taking rumors from a less than accurate source as gospel?

 

Did you know that the whole 7nm stuff is BS. Do you know why? Because there is no normalized way they are being measured. All the manufacturers measure them differently and that is one of the reasons intel wants there to be a set standard for determining the NM of a process.

 

Now as to the 15% IPC improvement, first off that is a pretty unrealistic number. Second they are much farther behind intel than a mere 5% which you are basically implying. They are anywhere from 15-18% behind intels IPC currently.

Hardware unboxed did a video comparing the 2600x to the 8700k clock for clock and in all workloads but gaming they were very close to each other in terms of performance. So I really don't think they are 15-20% behind in IPC it's more likely a single digit figure

[AngryBeaver]

3 minutes ago, C2HWarrior said:

Hardware unboxed did a video comparing the 2600x to the 8700k clock for clock and in all workloads but gaming they were very close to each other in terms of performance. So I really don't think they are 15-20% behind in IPC it's more likely a single digit figure

A lot of that was because of how well the infinity fabric is doing in multi-threaded tasks vs the ring bus. The ring is good for gaming, but is slowly losing ground when it comes to multi-threading. IPC improvements will show up most in things like gaming by the way. Where Single Thread performance is king.

[Brooksie359]

22 minutes ago, AngryBeaver said:

If AMD is making the move to more cores on the consumer platform that isn't a good sign. It means they are hitting a clock speed wall already. So they are just throwing more cores at the problem.

 

This is what we are seeing with intels consumer move from 4c to 6c now to 8c in the next release. They are starting to hit the frequency wall while also being able to only make minimal IPC gains... so in that scenario if you want to improve you add more cores. Now the question will be how long until the ring bus needs to be axed for a better solution on the consumer side.

 

Now back to my amd point. AMD is already struggling for single threat performance. So if they are hitting a frequency wall already at the 4.2ghz range, then they will be hard pressed to ever catch up to intel. They are also behind in IPC which will make that hurdle even taller.  So their solution appears to be back to where it was in the old days... Produce weaker cores, but add more of them to each chip. Gives them an edge in multi-threading, but hurts single-thread and by proxy gaming performance.

We have no idea if they are adding more cores and we don't know if they are hitting a wall in terms of frequency. My guess is that with 7nm it will improve clocks by a good amount to where the frequency difference won't be nearly as large as it is now. Also the IPC difference isn't that large and with a relatively new architecture it is much easier to find ipc improvement compared to a mature one so the gap could close relatively quickly. I would say we will just have to wait and see but I think there is a good possibility that Zen 2 will have major performance improvement compared to Zen and Zen+. 

[C2HWarrior]

1 minute ago, AngryBeaver said:

A lot of that was because of how well the infinity fabric is doing in multi-threaded tasks vs the ring bus. The ring is good for gaming, but is slowly losing ground when it comes to multi-threading. IPC improvements will show up most in things like gaming by the way. Where Single Thread performance is king.

Single thread cinebench was within 3%. I'll agree though that amds smt seems to be better than hyper-threading for multi-threaded workloads which would bolster the results in in favor of Amd for some of the results. Still I do believe Amd's main problem seems to be latency given the results shown in the video and the 3% IPC gain from zen to zen+ due to the reduction in latency. If amd could cut down on ccx latency in zen 2 then they may close the gap between to Intel's ring bus (they will never beat it though). faster ram would really help so fingers crossed that ddr5 arrives at least in time for zen 3.

[The Benjamins]

36 minutes ago, C2HWarrior said:

Hardware unboxed did a video comparing the 2600x to the 8700k clock for clock and in all workloads but gaming they were very close to each other in terms of performance. So I really don't think they are 15-20% behind in IPC it's more likely a single digit figure

 

32 minutes ago, AngryBeaver said:

A lot of that was because of how well the infinity fabric is doing in multi-threaded tasks vs the ring bus. The ring is good for gaming, but is slowly losing ground when it comes to multi-threading. IPC improvements will show up most in things like gaming by the way. Where Single Thread performance is king.

 

https://www.techspot.com/article/1616-4ghz-ryzen-2nd-gen-vs-core-8th-gen/

 

just linking so everyone can see it with out looking for it.

[Lawliet93]

1 hour ago, AngryBeaver said:

So you are taking rumors from a less than accurate source as gospel?

 

Did you know that the whole 7nm stuff is BS. Do you know why? Because there is no normalized way they are being measured. All the manufacturers measure them differently and that is one of the reasons intel wants there to be a set standard for determining the NM of a process.

 

Now as to the 15% IPC improvement, first off that is a pretty unrealistic number. Second they are much farther behind intel than a mere 5% which you are basically implying. They are anywhere from 15-18% behind intels IPC currently.

 

Wtf are you talking about? Clock for clock, Amd is about 3% behind intel, there are only two reasons why intel is better let's say at gaming. They use higher clockspeeds and they are using a ring bus architecture which has lower latencies between cores when compared to AMDs infinity fabric and latencies between two ccxs. That's the only reason why Amd is behind, where have you found that type of information about the ipc difference? Do you even know what ipc means? 

[Lawliet93]

1 hour ago, MyName13 said:

But then Intel will have their 10nm

But when, that's the question ?

[Dylanc1500]

17 minutes ago, Lawliet93 said:

But when, that's the question ?

Soon™

[asus killer]

3 hours ago, The Benjamins said:

Where have you been, clocks are IPC have both hit walls, 7nm it expected to bring better clocks and zen2 is also expected to have a slight IPC improvment which will put it very close to intel. core/thread count is the area that yield the biggest improvements. The same amount of money it takes to gain 5% IPC would of gained 50% core improvements, why would anyone waste money on a poor investment.

have you seen the new macbook pro? the question is not do we don't need more cores, the question is do we need this many and what does it sacrifice. No one wants to continue with quad cores but do you need 10 cores, really? I mean you could but for that you have threadripper and xeon,... like you always could buy more then 4 cores.

6 cores with hypertreading is more than enough for average use, what we need is higher clocks. I don't want a xeon with 20 cores at 2.0Ghz

 

2 hours ago, Jito463 said:

The thing is, we'd still be stuck on 4-core systems as mainstream if it weren't for AMD pressuring Intel (in fact, we'd probably be stuck on dual-core or evens single-core systems if not for AMD).  While it may not be immediately beneficial to all people, the hardware has to come first before the software catches up.

 

Also - as has already been pointed out - AMD is working on both IPC and core count for Zen 2.  It's not like they're only able to do one or the other.

i never said 4 cores was enough, still 10 or more, what for? are you planning to build a server

[The Benjamins]

2 minutes ago, asus killer said:

have you seen the new macbook pro? the question is not do we don't need more cores, the question is do we need this many and what does it sacrifice. No one wants to continue with quad cores but do you need 10 cores, really? I mean you could but for that you have threadripper and xeon,... like you always could buy more then 4 cores.

6 cores with hypertreading is more than enough for average use, what we need is higher clocks. I don't want a xeon with 20 cores at 2.0Ghz

I am not current on the macbook issue, AFAIK it could of been apple not updating the cooler to handle the higher TDP, or it could be intel misleading apple with a TDP that is not fully accurate. but this issue has zero correlation to how many cores it has.

 

but you are saying that AMD should not improve because we don't need it today, so just stop the progression of tech.

 

hey when they sell a 10c for $300 that means we can get a 6c for around $150 or even less, I don't see how that's a bad thing.

 

with games using 4-6 cores now, and people that use more then 1 screen WILL have other applications open, which can take use of the other 2-6 cores. this will also allow for more casual users get access to the ability to edit 4k video on the cheap, or live stream 1080p60+, 3D model rendering.

[normpearii]

The Macbook is a poor example.


The only reason it's having issues is because Apple didn't bother to think about cooling. The same CPU will be fine in laptops of adequate cooling.

[asus killer]

17 minutes ago, The Benjamins said:

I am not current on the macbook issue, AFAIK it could of been apple not updating the cooler to handle the higher TDP, or it could be intel misleading apple with a TDP that is not fully accurate. but this issue has zero correlation to how many cores it has.

 

but you are saying that AMD should not improve because we don't need it today, so just stop the progression of tech.

 

hey when they sell a 10c for $300 that means we can get a 6c for around $150 or even less, I don't see how that's a bad thing.

 

with games using 4-6 cores now, and people that use more then 1 screen WILL have other applications open, which can take use of the other 2-6 cores. this will also allow for more casual users get access to the ability to edit 4k video on the cheap, or live stream 1080p60+, 3D model rendering.

what you are telling me is that they should lower prices of Epyc and threadripper processors, the lower the price the better, how can i disagree. 

[The Benjamins]

47 minutes ago, asus killer said:

what you are telling me is that they should lower prices of Epyc and threadripper processors, the lower the price the better, how can i disagree. 

I never said they should lower the prices of the whole line up, I said when the line up increases core count over the all products the price per core will get lower. let me give you example.

 

2800x xxxxx - N/A    > 3800x 12c24t - $500

2700x 8c16t - $300 > 3700x 10c20t - $300

2600x 6c12t - $200 > 3600x 8c16t - $200

2500x 4c8t  - $120 > 3500x 6c12t -$120

 

Epycs top chip is now 32c, but will be 48c on zen2 in h12019 it will be at around the same price point.

I expect Threadripper 2 24c48t part to be around the price of the 1950x (launch MSRP)

 

do you not see how adding cores to the stack gives us the same core count for lower price? even on intels side from 7000s to 8000s we now get a 4 core part a lot cheaper then the gen before.

 

so why would you not want AMD to increase core count every 1-2 generations which in turn will decrease the cost of equivalent core count parts in turn.

 

I side with progression and making parts at the same performance level cheaper.

 

lastly Threadripper prices have been coming down SO much lately due to gen2 coming out in the expected month time frame, that's because the new ones HAVE MORE CORES.

[asus killer]

9 minutes ago, The Benjamins said:

I never said they should lower the prices of the whole line up, I said when the line up increases core count over the all products the price per core will get lower. let me give you example.

 

2800x xxxxx - N/A    > 3800x 12c24t - $500

2700x 8c16t - $300 > 3700x 10c20t - $300

2600x 6c12t - $200 > 3600x 8c16t - $200

2500x 4c8t  - $120 > 3500x 6c12t -$120

 

Epycs top chip is now 32c, but will be 48c on zen2 in h12019 it will be at around the same price point.

I expect Threadripper 2 24c48t part to be around the price of the 1950x (launch MSRP)

 

do you not see how adding cores to the stack gives us the same core count for lower price? even on intels side from 7000s to 8000s we now get a 4 core part a lot cheaper then the gen before.

 

so why would you not want AMD to increase core count every 1-2 generations which in turn will decrease the cost of equivalent core count parts in turn.

 

I side with progression and making parts at the same performance level cheaper.

 

lastly Threadripper prices have been coming down SO much lately due to gen2 coming out in the expected month time frame, that's because the new ones HAVE MORE CORES.

i do not disagree with you that the cheaper the better or that more cores don't hurt.

Still the average user (and the average user doesn't have 2 monitors, stream, have 20 chrome tabs open all at the same time) needs is more IPC and higher core clocks. An equilibrium. I still think more than 6 cores for average user is a waste, and then you want to play a game or use premiere and it only uses 4 to 6 cores at best 

 

https://www.pugetsystems.com/labs/articles/Adobe-Premiere-Pro-CC-Multi-Core-Performance-698/

 

[Jito463]

3 hours ago, mynameisjuan said:

Not necessarily, you can code a program to use as many cores as it can, if the cores are not there no harm. 12+ cores has been available for a while but devs dont seem to care. 

 

Hell, even with quad cores most programs still use 1-2 cores. Keep blaming the hardware but no one other than video editors have been pushing cores. 

3 hours ago, mynameisjuan said:

Dude many programs still use single core. Devs that dont give a **** about their program will not recode it, good devs will. I dont get your argument. You want full performance of your program. 

 

Games are pushing hardware, software that is not is just laziness. 

To be blunt, most software developers are lazy (and that's not criticism, just a fact).  It's not even necessarily a bad thing, either.  If there is only a small subset of your customers that will have more than a dual-core or quad-core system, then why go through the extra time and effort of coding for multi-processor setups if it will work fine running on one or two cores?

3 hours ago, AngryBeaver said:

Now as to the 15% IPC improvement, first off that is a pretty unrealistic number. Second they are much farther behind intel than a mere 5% which you are basically implying. They are anywhere from 15-18% behind intels IPC currently.

2 hours ago, C2HWarrior said:

Hardware unboxed did a video comparing the 2600x to the 8700k clock for clock and in all workloads but gaming they were very close to each other in terms of performance. So I really don't think they are 15-20% behind in IPC it's more likely a single digit figure

IPC isn't a fixed number.  It varies depending on the workload.  For example, in your average testing AMD and Intel may be within a few percentage points in IPC; whereas if you run a AVX2 workload, you will find that Intel's IPC is far greater than AMD.  There's no set "this is the IPC, without deviation", it all depends on what kind of work you're doing.

1 hour ago, asus killer said:

i never said 4 cores was enough, still 10 or more, what for? are you planning to build a server

My mom absolutely loves to play these stupid flash games, and she'll leave them running and minimized all the time while she does other stuff on her computer, no matter how many times I've tried to convince her otherwise.  Because of that, it constantly gets bogged down (she has a dual-core system).  If she had a 6 or 8-core setup, she'd never even notice the slowdown.  Just because your average person doesn't normally run software that requires multiple cores, doesn't mean they can't benefit from it.

[The Benjamins]

2 minutes ago, asus killer said:

i do not disagree with you that the cheaper the better or that more cores don't hurt.

Still the average user (and the average user doesn't have 2 monitors, stream, have 20 chrome tabs open all at the same time) needs is more IPC and higher core clocks. An equilibrium. I still think more than 6 cores for average user is a waste, and then you want to play a game or use premiere and it only uses 4 to 6 cores at best 

 

https://www.pugetsystems.com/labs/articles/Adobe-Premiere-Pro-CC-Multi-Core-Performance-698/

 

Fair enough.

 

But those users wouldn't or shouldn't be buying to top mainstream parts. They should be on the R3, R5 on the AMD side.

[Taf the Ghost]

2 hours ago, Lawliet93 said:

 

Wtf are you talking about? Clock for clock, Amd is about 3% behind intel, there are only two reasons why intel is better let's say at gaming. They use higher clockspeeds and they are using a ring bus architecture which has lower latencies between cores when compared to AMDs infinity fabric and latencies between two ccxs. That's the only reason why Amd is behind, where have you found that type of information about the ipc difference? Do you even know what ipc means? 

He's referencing professional space benchmarking with Zen cores vs Skylake-X cores. In AVX-accelerated workloads, the IPC difference is actually around 25%, per The Stilts testing. But unless you're doing Fluid Dynamic Modeling, you're never going to find that. AMD actually has a slight IPC advantage in straight INT performance, lol.

 

There's a reason we're expecting full AVX2 units with Zen2 designs. Better cache & those AVX2 units, on the die shrink, will put a world of hurt into Intel's market share.

[The Benjamins]

8 minutes ago, valdyrgramr said:

Well, you can get a 1950x for around 700-750 now, wasn't MSRP like 1000?  I'm getting a 1950x myself for Maya and other workstation tasks for now, and when TR2 drops down in price I'm going TR2 since my ASRock Taichi will support it via a bios update.  Not sure if they confirmed the price yet, but one of the TR2s "might" be released on August 13th from what I was told, but not sure if that is actually confirmed yet.

ya its great.

the 1950x launched at $1000, but was almost always cheaper.

 

TR2s are expected to be announced end of this month and release mid August from what I heard.

 

I seen the 1920x as low as $450, and the 1900x is at $300 on amazon.

[Brooksie359]

20 minutes ago, valdyrgramr said:

Well, you can get a 1950x for around 700-750 now, wasn't MSRP like 1000?  I'm getting a 1950x myself for Maya and other workstation tasks for now, and when TR2 drops down in price I'm going TR2 since my ASRock Taichi will support it via a bios update.  Not sure if they confirmed the price yet, but one of the TR2s "might" be released on August 13th from what I was told, but not sure if that is actually confirmed yet.

Are you going to get the 32 core tr2 if prices drop down after awhile? I am actually curious who is going to buy it as it seems like a niche. I mean part of me wouldn't mind having a 32 core cpu just because for the hell of it.

[Brooksie359]

6 minutes ago, valdyrgramr said:

Was considering it because right now I'm going from a 4790k to a 1950x because of the increase on the multi-core performance is kinda huge.  The 4790k beats it slightly in terms of single-thread.  Originally I was going to go i9, but the pricing for i9 is ridiculous to get the base one it's like 800 something, to get equal core/thread count I'd need to spend 1200, and if I wanted more cores/threads for rendering purposes and an 11 percent increase in overall performance I'd need to spend 1800 USD.  

http://cpu.userbenchmark.com/Compare/Intel-Core-i7-4790K-vs-AMD-Ryzen-TR-1950X/2384vs3932  If it's a small increase in performance "maybe" for the fuck of it if the price is good, but if it's nearly a 30 percent increase in performance or more, then fuck ya I'd jump on that when the price drops.  Was thinking of doing 128gbs of ram just because I can, but for now I'm doing 32 got a 16gb 3200mhz ripjaw atm so I can afford the 1950x, picking up another in October, a 5tb x300, and an ekwb nickel block for my watercooling setup too.  Depending on what I'm doing in Maya I could easily use 32gbs from what I've been told, and if that's true I might go as far as 64 which is a tad more reasonable than 128.  Seriously can't think of any situation where I'd be using more than 32-64 gigs.  Maybe if I open a tab in Chrome while compiling C++ and rendering?

The maybe with the new chrome security update and 100+ tabs. 

[C2HWarrior]

1 hour ago, Jito463 said:

To be blunt, most software developers are lazy (and that's not criticism, just a fact).  It's not even necessarily a bad thing, either.  If there is only a small subset of your customers that will have more than a dual-core or quad-core system, then why go through the extra time and effort of coding for multi-processor setups if it will work fine running on one or two cores?

IPC isn't a fixed number.  It varies depending on the workload.  For example, in your average testing AMD and Intel may be within a few percentage points in IPC; whereas if you run a AVX2 workload, you will find that Intel's IPC is far greater than AMD.  There's no set "this is the IPC, without deviation", it all depends on what kind of work you're doing.

My mom absolutely loves to play these stupid flash games, and she'll leave them running and minimized all the time while she does other stuff on her computer, no matter how many times I've tried to convince her otherwise.  Because of that, it constantly gets bogged down (she has a dual-core system).  If she had a 6 or 8-core setup, she'd never even notice the slowdown.  Just because your average person doesn't normally run software that requires multiple cores, doesn't mean they can't benefit from it.

I wouldn't really include avx here because it's not a realistic workload for 99% of consumers. That's why amd didn't bother to focus on it. Perhaps they will use some of the extra space for better avx performance in zen2

[Mira Yurizaki]

3 hours ago, C2HWarrior said:

I wouldn't really include avx here because it's not a realistic workload for 99% of consumers. That's why amd didn't bother to focus on it. Perhaps they will use some of the extra space for better avx performance in zen2

There's a lot of applications consumers use that could benefit from what AVX offers, like:

• Content creation apps
• Multimedia playback apps
• File compression
• Excel (I mean, if you do a lot of things on the side)
• The RPCS3 emulator uses AVX

To put in another way a random user on the internet said: it's much easier to vectorize algorithms than it is to multithread them.

[leadeater]

15 minutes ago, M.Yurizaki said:

There's a lot of applications consumers use that could benefit from what AVX offers, like:

• Content creation apps
• Multimedia playback apps
• File compression
• Excel (I mean, if you do a lot of things on the side)
• The RPCS3 emulator uses AVX

To put in another way a random user on the internet said: it's much easier to vectorize algorithms than it is to multithread them.

Plus Zen is not weak in AVX1, it's actually really good at that. It's just not good at AVX2. Issue is Intel doesn't have this behavior so there is no reason to not use AVX2 when it may not be required, on Zen it could be better to not use it in certain cases but you'd have to optimize in that way.

[Stefan Payne]

8 hours ago, Brooksie359 said:

Also the IPC difference isn't that large and with a relatively new architecture it is much easier to find ipc improvement compared to a mature one so the gap could close relatively quickly. I would say we will just have to wait and see but I think there is a good possibility that Zen 2 will have major performance improvement compared to Zen and Zen+. 

Yes, especially since we end users already know where the problems of Zen lie. So to improve the performance without increasing the clock shouldn't be hard.

Just look at how big the difference between Zen and Zen+ is!

And that doesn't even have "real changes"...

 

One of the possible modifications could be the introduction of uncore Multiplicators.

 

And the next thing they need to work on is memory latency. 

Just something to think about, here an FX9590:

http://www.guru3d.com/news-story/amd-fx-9590-benchmarks-surface.html

 

And now Zen:

https://www.techpowerup.com/243150/amd-2nd-generation-ryzen-pinnacle-ridge-available-in-russia

 

And you also see one of the Problems Bulldozer has...

 

Now here an Intel Chip:

https://www.techpowerup.com/forums/threads/share-your-aida-64-cache-and-memory-benchmark-here.186338/page-26

 

So what do we see now, when comparing:

a) Bulldozer has lower memory Latency than Zen

b) Zen has really shitty Memory Latency

c) L2 and L3 Latency of Zen is equal to the other side

d) L2 and L3 Latency of FX is abysmal

 

I couln't find Aida 64 benches of an older Athlon 64, but Science Mark should do it too:

http://hexus.net/tech/reviews/cooling/1255-amd-athlon-64-3000-venice-overclocking/?page=6

So although the memory latency is that bad, Zen performs pretty well and that obvious problem should be easier to work with. Because you know where you have to look to increase the performance and know what you have to do.

[Stefan Payne]

23 minutes ago, leadeater said:

Plus Zen is not weak in AVX1, it's actually really good at that. It's just not good at AVX2. Issue is Intel doesn't have this behavior so there is no reason to not use AVX2 when it may not be required, on Zen it could be better to not use it in certain cases but you'd have to optimize in that way.

There is one thing you are missing:

https://forums.anandtech.com/threads/lower-avx-clocks.2476725/

 

the decision of not going to 256bit AVX decreased power consumption and is one of the reasons that AMD is as efficient as they are now. If they would increase it to 256bit AVX, it would increase power consumption and the Intel fans would come out with "get Intel, more Power Efficient"

And its not really much better on Desktop CPUs either...

 

And why not go full GPGPU with that instead?! 

Even Wikipedia mentions GPGPU under "Vector Processor"...

https://en.wikipedia.org/wiki/Vector_processor#GPGPU

 

 

Especially since you have to widen all data transfer paths inside the CPU for 256bit AVX wich you only need for AVX2. Doesn't seem to make much sense, does it?