Discussion: AMD Ryzen

[ComputerSecurityGuy]

But for the server market and to a lesser extent to prosumer market more cores and more multithreaded performance are important.

Eight core Zen never was and never will be a gaming chip. That is what quad core Zen is for (and six core if we get it.)

 

[elbert]

Info has leaked suggesting AMD will launch their X370 extreme chipset February 2017. This may be the 12 and 16 core zen+ core plateform.

The X370 chipset is for Zen, not for Zen+. Moreover, there are zero indications/leaks that Zen+ will come in 12 and 16 core configurations.

Here is a few leaks on the 16 core but there is a whole lot more if you want to search..
http://wccftech.com/amd-allegedly-preparing-apu-sixteen-zen-cores-greenland-hbm-graphics/
http://wccftech.com/amd-naples-32-core-zen/
http://betanews.com/2016/02/13/amd-zen-processors-will-have-32-cores-and-symmetrical-multi-threading/
http://www.digitaltrends.com/computing/rumors-suggest-amd-is-eyeing-a-comeback-with-its-new-16-core-chip/
Now we know AMD has 24 and 32 core server chips and only have proof they can splice 2 physical silicon chips together. Thus 12 and 16 core physical silicon chips must exist.

Digitaltrends say's " a leaked slide acquired by Fudzilla"

Betanews only leaks to a article that says we will have 32 core Zen we will for servers

Also it doesn't matter if Amd made a 100 core Zen CPU if it lacks behind Ivy-sandy in IPC it just doesn't matter. I doubt they will lack that however i expect sandy-ivy level IPC always did based off the design.

It really does as future CPU's are locked at windows 10. Thus threads will really start to matter as we've seen thread improvements all the way up to 8 cores on windows 10 games. Even the insignificant threads that rob the games of 3~5% will be placed on cores beyond 4 which does make for inprovments even in older titles.
Now does all those link mean AMD is or isn't making 12 and 16 core chips? We again know of the 24 and 32 core spliced chips which lends a great likly hood to these enthusiased CPU's.

Wouldn't bet on it, Also if one core gets stressed over 80 percent out of the 16 cores a dual core I3 with more single core performance will win. Again software works in a certain way at least for games and a lot of other programs need certain things to finish before the program can continue its task which causes stuttering and lag.

Nope as the single core performance has to be far more substantial in newer operating systems, dx12, and opengl.

 

[jdwii]

But for the server market and to a lesser extent to prosumer market more cores and more multithreaded performance are important.

Eight core Zen never was and never will be a gaming chip. That is what quad core Zen is for (and six core if we get it.)

Info has leaked suggesting AMD will launch their X370 extreme chipset February 2017. This may be the 12 and 16 core zen+ core plateform.

The X370 chipset is for Zen, not for Zen+. Moreover, there are zero indications/leaks that Zen+ will come in 12 and 16 core configurations.

Here is a few leaks on the 16 core but there is a whole lot more if you want to search..
http://wccftech.com/amd-allegedly-preparing-apu-sixteen-zen-cores-greenland-hbm-graphics/
http://wccftech.com/amd-naples-32-core-zen/
http://betanews.com/2016/02/13/amd-zen-processors-will-have-32-cores-and-symmetrical-multi-threading/
http://www.digitaltrends.com/computing/rumors-suggest-amd-is-eyeing-a-comeback-with-its-new-16-core-chip/
Now we know AMD has 24 and 32 core server chips and only have proof they can splice 2 physical silicon chips together. Thus 12 and 16 core physical silicon chips must exist.

Digitaltrends say's " a leaked slide acquired by Fudzilla"

Betanews only leaks to a article that says we will have 32 core Zen we will for servers

Also it doesn't matter if Amd made a 100 core Zen CPU if it lacks behind Ivy-sandy in IPC it just doesn't matter. I doubt they will lack that however i expect sandy-ivy level IPC always did based off the design.

It really does as future CPU's are locked at windows 10. Thus threads will really start to matter as we've seen thread improvements all the way up to 8 cores on windows 10 games. Even the insignificant threads that rob the games of 3~5% will be placed on cores beyond 4 which does make for inprovments even in older titles.
Now does all those link mean AMD is or isn't making 12 and 16 core chips? We again know of the 24 and 32 core spliced chips which lends a great likly hood to these enthusiased CPU's.

Wouldn't bet on it, Also if one core gets stressed over 80 percent out of the 16 cores a dual core I3 with more single core performance will win. Again software works in a certain way at least for games and a lot of other programs need certain things to finish before the program can continue its task which causes stuttering and lag.

Nope as the single core performance has to be far more substantial in newer operating systems, dx12, and opengl.

We will see..... It's 2011 all over again with the fanbase at least

 

[gamerk316]

Just going to point out that Total War titles and AoS can be considered "best case" outcomes for CPU scaling. RTS tend to scale better because, unlike other types of games, you have significant amounts of AI processing taking place (compared to the relatively simplistic AI in a FPS, by contrast). In any case, both tend to maximize at three CPU cores (Tested TW personally on my 2600k), which is about what I'd expect.

Strangely enough, I gave some thought to this very same topic today. I was playing DOOM yesterday and thought about the number of unit types and how they were spawning throughout the game. Why they weren't constantly there and abundant like in the old DOOM (original), where you were zerged all the time, pretty much.

I started sketching an AI type of engine and initially it would actually scale incredibly well. The thing with AI is how complex you want to make it takes SO MUCH TIME. So, the more "decisions" you want to build into it, the more code lines they are intrinsically. It's not even about a broad approach to how you want to thread it, but how "smart" you want it to be.

To validate my train of thought, I started reading about the new Tesla cards from nVidia (sorry for the side-topic!) and yeah; depending on how you model it, then a single unit could pretty much take into account all of the world/scene aspects and make decisions on top of them. So, from the theoretical POV, AIs are meant to be well threaded by nature, but also scale in complexity per... let's say "intelligence unit".

It's a very interesting train of thought that took half of my day, hahaha.

Cheers!

Not to get too off-topic, but it really comes down to:

A: How many unique AIs you have to process

B: How much work each AI Does

C: How much communication between AI objects

In the DOOM case, you have all the AIs that are active moving toward the player, which makes the AI simple; you can have one thread do it and just process each active enemy, since you'll lose more performance by threading the AI then you'd gain due to it's simplicity. Even for a RTS, you can simplify the decision tree by using simplified decision making for a group of individual units.

In theory, AI processing should scale reasonably well, on the provision they don't need to communicate information to eachother (in which case, they'll scale like crap). But most games use VERY simplified AI decision trees to keep overall processing down, so only RTS and Strategy games really have AIs that do significant amount of work. Let's face it, FPS AI basically dumbs down to "See enemy, move to cover, occasionally peak out and shoot enemy".

 

[elbert]

Just going to point out that Total War titles and AoS can be considered "best case" outcomes for CPU scaling. RTS tend to scale better because, unlike other types of games, you have significant amounts of AI processing taking place (compared to the relatively simplistic AI in a FPS, by contrast). In any case, both tend to maximize at three CPU cores (Tested TW personally on my 2600k), which is about what I'd expect.

Strangely enough, I gave some thought to this very same topic today. I was playing DOOM yesterday and thought about the number of unit types and how they were spawning throughout the game. Why they weren't constantly there and abundant like in the old DOOM (original), where you were zerged all the time, pretty much.

I started sketching an AI type of engine and initially it would actually scale incredibly well. The thing with AI is how complex you want to make it takes SO MUCH TIME. So, the more "decisions" you want to build into it, the more code lines they are intrinsically. It's not even about a broad approach to how you want to thread it, but how "smart" you want it to be.

To validate my train of thought, I started reading about the new Tesla cards from nVidia (sorry for the side-topic!) and yeah; depending on how you model it, then a single unit could pretty much take into account all of the world/scene aspects and make decisions on top of them. So, from the theoretical POV, AIs are meant to be well threaded by nature, but also scale in complexity per... let's say "intelligence unit".

It's a very interesting train of thought that took half of my day, hahaha.

Cheers!

Not to get too off-topic, but it really comes down to:

A: How many unique AIs you have to process

B: How much work each AI Does

C: How much communication between AI objects

In the DOOM case, you have all the AIs that are active moving toward the player, which makes the AI simple; you can have one thread do it and just process each active enemy, since you'll lose more performance by threading the AI then you'd gain due to it's simplicity. Even for a RTS, you can simplify the decision tree by using simplified decision making for a group of individual units.

In theory, AI processing should scale reasonably well, on the provision they don't need to communicate information to eachother (in which case, they'll scale like crap). But most games use VERY simplified AI decision trees to keep overall processing down, so only RTS and Strategy games really have AIs that do significant amount of work. Let's face it, FPS AI basically dumbs down to "See enemy, move to cover, occasionally peak out and shoot enemy".

Its not totally the games but major game engines. Crytek for example made cryengine plans to upgrade to DX12, phyx, and Vulkan by Febrary. http://wccftech.com/cryengine-get-vulkan-physx-support-november-dx12-mgpu-support-february-2017/
Look at IDs game engine that includes doom to all go Valkan and the hundreds of Ids engine supported games.

 

[jimmysmitty]

http://www.pcper.com/news/Processors/Leaked-Geekbench-Results-AMD-Zen-Performance-Extrapolated?destination=node/66172#comment-308881

Specifically:

AES numbers make sense, given AMD has dedicated resources on the CPU die. For all other tasks, you see performance 10-30% slower then IB at the same clocks.

If true (And that's a BIG if), Zen likely doesn't even match Sandy Bridge IPC levels.

EDIT

I want to make this very clear, if these results are accurate, Zen won't even beat Sandy Bridge head to head. This would be significantly below even my worst-case estimations for Zen's performance.

There's a side of me that says the results are so bad, they can't possibly be real. Then again, I said the same thing in regards to Bulldozer when the first real numbers started to leak.

The AM4 was released on Aug 5 but this test isn't on zen 8 core or AM4. Its based on AMD's 64 core server plateform and CPU's which aren't luanching for another 6 months. IE no ways close to mature.

Now even by the 80% given per core and 8 core zen would beat a 6 core Intel. Still amazing compared to the old AMD 8 core fx-8300's couldn't even beat an i5 quad.
Math: 8 * 0.80 = 6.4 > 6 * 1.00 = 6.0

I would not be impressed by it beating their old CPUs much like Core 2 was not impressive because it beat Netburst, it was impressive because it also beat Athlon 64 X2 in every area. That is what would make it impressive. AMD literally had nowhere to go but up from the FX series.

Ummm Core 2 didnt beat FX60 and FX62 in every area. It was in fact a pretty hard toes up. The splicing of 2 to make the Q6600 Kentsfield was the impressive chip that broke Athlon. Now granted AMD will have to come with a plus core and still its going to be the 24 and 32 core that has any shot of out and out win.

On a per clock level for consumers Core 2 Duo pounded even the FX-62:

http://www.anandtech.com/show/2045

The FX-62 is always behind the lower clocked E6700 and sometimes behind the even lower clocked E6600 while using less power under load. At the time of that review the FX-62 was $1031 but just short of Core 2 availability they dropped it to $827. At that time a $316 E6600 was a vastly better buy.

That is off topic but my point is that Zen beating FX is not amazing nor should it be praised. It should be expected. Zen has to beat FX. They cannot have another Phenom, Phenom was slower than Athlon 64 X2 per clock, nor can they have another FX series CPU.

If they want to be truly impressive to the computing community they have to offer a better CPU in performance and per dollar than Intel and in the server market beat them in not only perf/$ but also per watt as servers are designed around thermal limits of CPUs and right now Intel is killing AMD in the perf/watt for servers.

You link even has the FX-62 beating some of the core 2 duo's. Re-read my post as my point was an AMD shift from 90nm to 65nm would have turned or either put them both in a tight race. It was the Q6600 that totally blocked AMD out.
Now times have changed and consumers aren't getting the top end anything thus my point the only place of any win is the 24 and 32 cores. Any win the consumers will see isn't even in your sight line yet. Intel will probably release 12 and 14 core broadwell-e's if AMD brings their 12 and 16 core part to an enthusias plateform.

That's what I said but you missed what I said. If you look at it on a per clock level, Core 2 was superior. It also cost less, again look at what I put. The FX-62 was vastly more than the E6700 which is almost never beat and insanely more than the E6600 which it only traded a few blows with. And that was just at stock clocks.

The Q6600 was devastating but only because the AMD equivalent was the QuadFX setup which required a very expensive dual socket board and two very expensive CPUs. It was about $1K in just board and CPUs not including RAM and yet failed to outperform it. It even used vastly more power.

Again my point is that Zen beating FX means nothing and is not note worthy at all. It has to beat Intel and do it well in multiple areas in order to be impressive enough to be considered.

Even then who cares about the consumer market? The server market makes way more money which is what AMD needs more than anything. A strong win in the server market would be very good and might benefit us as well.

However I have not seen anything about a 12 core Zen CPU. I have head the 16 core might take a page from Core 2 Quad and use a MCM approach but I highly doubt AMD will push a 16 core on the enthusiast market when quad cores are still under utilized in the mainstream market.

 

[elbert]

http://www.pcper.com/news/Processors/Leaked-Geekbench-Results-AMD-Zen-Performance-Extrapolated?destination=node/66172#comment-308881

Specifically:

AES numbers make sense, given AMD has dedicated resources on the CPU die. For all other tasks, you see performance 10-30% slower then IB at the same clocks.

If true (And that's a BIG if), Zen likely doesn't even match Sandy Bridge IPC levels.

EDIT

I want to make this very clear, if these results are accurate, Zen won't even beat Sandy Bridge head to head. This would be significantly below even my worst-case estimations for Zen's performance.

There's a side of me that says the results are so bad, they can't possibly be real. Then again, I said the same thing in regards to Bulldozer when the first real numbers started to leak.

The AM4 was released on Aug 5 but this test isn't on zen 8 core or AM4. Its based on AMD's 64 core server plateform and CPU's which aren't luanching for another 6 months. IE no ways close to mature.

Now even by the 80% given per core and 8 core zen would beat a 6 core Intel. Still amazing compared to the old AMD 8 core fx-8300's couldn't even beat an i5 quad.
Math: 8 * 0.80 = 6.4 > 6 * 1.00 = 6.0

I would not be impressed by it beating their old CPUs much like Core 2 was not impressive because it beat Netburst, it was impressive because it also beat Athlon 64 X2 in every area. That is what would make it impressive. AMD literally had nowhere to go but up from the FX series.

Ummm Core 2 didnt beat FX60 and FX62 in every area. It was in fact a pretty hard toes up. The splicing of 2 to make the Q6600 Kentsfield was the impressive chip that broke Athlon. Now granted AMD will have to come with a plus core and still its going to be the 24 and 32 core that has any shot of out and out win.

On a per clock level for consumers Core 2 Duo pounded even the FX-62:

http://www.anandtech.com/show/2045

The FX-62 is always behind the lower clocked E6700 and sometimes behind the even lower clocked E6600 while using less power under load. At the time of that review the FX-62 was $1031 but just short of Core 2 availability they dropped it to $827. At that time a $316 E6600 was a vastly better buy.

That is off topic but my point is that Zen beating FX is not amazing nor should it be praised. It should be expected. Zen has to beat FX. They cannot have another Phenom, Phenom was slower than Athlon 64 X2 per clock, nor can they have another FX series CPU.

If they want to be truly impressive to the computing community they have to offer a better CPU in performance and per dollar than Intel and in the server market beat them in not only perf/$ but also per watt as servers are designed around thermal limits of CPUs and right now Intel is killing AMD in the perf/watt for servers.

You link even has the FX-62 beating some of the core 2 duo's. Re-read my post as my point was an AMD shift from 90nm to 65nm would have turned or either put them both in a tight race. It was the Q6600 that totally blocked AMD out.
Now times have changed and consumers aren't getting the top end anything thus my point the only place of any win is the 24 and 32 cores. Any win the consumers will see isn't even in your sight line yet. Intel will probably release 12 and 14 core broadwell-e's if AMD brings their 12 and 16 core part to an enthusias plateform.

That's what I said but you missed what I said. If you look at it on a per clock level, Core 2 was superior. It also cost less, again look at what I put. The FX-62 was vastly more than the E6700 which is almost never beat and insanely more than the E6600 which it only traded a few blows with. And that was just at stock clocks.

The Q6600 was devastating but only because the AMD equivalent was the QuadFX setup which required a very expensive dual socket board and two very expensive CPUs. It was about $1K in just board and CPUs not including RAM and yet failed to outperform it. It even used vastly more power.

Again my point is that Zen beating FX means nothing and is not note worthy at all. It has to beat Intel and do it well in multiple areas in order to be impressive enough to be considered.

Even then who cares about the consumer market? The server market makes way more money which is what AMD needs more than anything. A strong win in the server market would be very good and might benefit us as well.

However I have not seen anything about a 12 core Zen CPU. I have head the 16 core might take a page from Core 2 Quad and use a MCM approach but I highly doubt AMD will push a 16 core on the enthusiast market when quad cores are still under utilized in the mainstream market.

The thing is a nm reduction would have put the core 2 duo's back against a wall. It was the Q6600 that really done the damage. Even netburst was mixing it up with the low end X2-64's so its not much of leap. What your point to tho is what your missing. Think of the E6700 as todays 8~10 core broadwell_ep. If Zen outperforms the boardwell 6 core its a massive win for AMD.
QuadFX was way late to show after the fx-64 65nm to get power under control. If AMD has its power to performance under control it may win on the 32 core but thats the only win on the level of kentsfield.

 

[elbert]

The major problem with Zen is the never ending delays. In June AMD CEO demo'ed Zen and let slip an August. Later leaks moved it to Oct and then Late 4th quarter with only OEM's getting supply. Now leaks say February which means without question against Kaby Lake which will without quest put the 4 core Zen at a disadvantage against even the i5's.
http://fudzilla.com/news/processors/41598-amd-zen-desktop-might-slip-to-q1-17

 

[juanrga]

Info has leaked suggesting AMD will launch their X370 extreme chipset February 2017. This may be the 12 and 16 core zen+ core plateform.

The X370 chipset is for Zen, not for Zen+. Moreover, there are zero indications/leaks that Zen+ will come in 12 and 16 core configurations.

Here is a few leaks on the 16 core but there is a whole lot more if you want to search..
http://wccftech.com/amd-allegedly-preparing-apu-sixteen-zen-cores-greenland-hbm-graphics/
http://wccftech.com/amd-naples-32-core-zen/
http://betanews.com/2016/02/13/amd-zen-processors-will-have-32-cores-and-symmetrical-multi-threading/
http://www.digitaltrends.com/computing/rumors-suggest-amd-is-eyeing-a-comeback-with-its-new-16-core-chip/
Now we know AMD has 24 and 32 core server chips and only have proof they can splice 2 physical silicon chips together. Thus 12 and 16 core physical silicon chips must exist.

No. The 16-core APU is made of four dies; two CPU dies plus two GPU dies. And the 32-core CPU is made of four CPU dies. AMD only has 8-core dies for Zen. Your second link even makes this explicit:

Thanks to AMD’s lego-like build philosophy even Naples will actually be fabricated in an 8-core base design, with 16 and 32 core chips being produced in an MCM (Multi Chip Module) format.

This quad-die configuration has been know for a while. In fact I predicted it about two years ago

http://www.tomshardware.co.uk/forum/id-2585271/amd-cpus-socs-rumors-speculations-temp-thread/page-9.html#15745175

and reposted it in this thread first page

http://www.tomshardware.co.uk/forum/id-2986517/discussion-amd-zen.html#18062733

I even used this image for illustration of the concept

Theory with no working proof nor does your first link work. Plus do you relize what your saying really doesn't block a possible 12 and 16 core. Really just makes it a possible but one should look at the AM4 socket size. Why would AMD just do what you suggest when not near the size limit of a single die?

The proof is on your own links, which disprove your claims and state clearly that there is only 8-core Zen dies.

I am not discussing what is technically possible or not. Current technology allows to build a 32-core die, but AMD is not building any. AMD is only fabricating 8-core dies for economic reasons and then using multi-die packages for the higher core count chips:

- AM4 socket (Summit Ridge): one die; up to 8 core CPU
- SP4 socket (Naples): four dies; up to 8x4 core CPU
- SP3 socket (Snowy Owl): two dies; up to 8x2 core CPU

The Consumer/Prosumer socket is AM4. The other sockets are for server Opteron.

Also I don't think that the AM4 socket is so big like you claim. It has about same size than FM2+ or AM3+ sockets.

 

[juanrga]

It really does as future CPU's are locked at windows 10.

The CPUs will be "locked' to W10" are the consumer 4-core and 8-core Zen. The 16-core APUs for HPC and the 32-core CPUs for servers will evidently use linux.

Thus threads will really start to matter as we've seen thread improvements all the way up to 8 cores on windows 10 games.

Last reviews I see DX12 games scaling up to 4 cores

for most gamers a high clocked 4C CPU will be better than a low clocked 8C CPU.

I see scaling all the way to 6 cores even in your example proof. A 3% gain is a 3% gain.

The first benchmark puts it clearly in text form: "# of cores before scaling stops: 4 cores". And the second benchmark shows an 3.5GHz 6MB 4C/4T CPU giving virtually the same performance than the 3.5GHz 15MB 6C/12T CPU. A 3% gain is not only statistically insignificant but unnoticeable during gaming. I think they would find that an 4GHz 8MB 4C/8T CPU would be faster if they had tested one.

Tho there is other tests that show different results.

To get the context, let us read the review from where you got this image:

Ashes of the Singularity is probably a best-case scenario for DirectX 12. It’s been in development for a long time, with much thought put into supporting all the goodness of the CPU cores in today’s PCs.

But not all games are going to be Ashes of the Singularity, even if they carry the DirectX 12 label.

And then they provide us this other game benchmark proving their point

This image is from a talk given by Kronos group at GDC 2014. It is about Vulkan and it is a simulation using a simulated infinitely fast GPU running Ashes of the Singularity. I provided you a real benchmark of this game showing a different picture than in simulations of unreal hardware (no one has an infinitely fast GPU). Also durng the talk they said that not all games will behave as in that simulation. They only talked about "Certain types of games". In any case, here you have another testing of the Ashes of the Singularity game confirming my points

I think the same: even in the best case scenario a high clocked quad-core performs the same than a low clocked octo-core.

 

[elbert]

Info has leaked suggesting AMD will launch their X370 extreme chipset February 2017. This may be the 12 and 16 core zen+ core plateform.

The X370 chipset is for Zen, not for Zen+. Moreover, there are zero indications/leaks that Zen+ will come in 12 and 16 core configurations.

Here is a few leaks on the 16 core but there is a whole lot more if you want to search..
http://wccftech.com/amd-allegedly-preparing-apu-sixteen-zen-cores-greenland-hbm-graphics/
http://wccftech.com/amd-naples-32-core-zen/
http://betanews.com/2016/02/13/amd-zen-processors-will-have-32-cores-and-symmetrical-multi-threading/
http://www.digitaltrends.com/computing/rumors-suggest-amd-is-eyeing-a-comeback-with-its-new-16-core-chip/
Now we know AMD has 24 and 32 core server chips and only have proof they can splice 2 physical silicon chips together. Thus 12 and 16 core physical silicon chips must exist.

No. The 16-core APU is made of four dies; two CPU dies plus two GPU dies. And the 32-core CPU is made of four CPU dies. AMD only has 8-core dies for Zen. Your second link even makes this explicit:

Thanks to AMD’s lego-like build philosophy even Naples will actually be fabricated in an 8-core base design, with 16 and 32 core chips being produced in an MCM (Multi Chip Module) format.

This quad-die configuration has been know for a while. In fact I predicted it about two years ago

http://www.tomshardware.co.uk/forum/id-2585271/amd-cpus-socs-rumors-speculations-temp-thread/page-9.html#15745175

and reposted it in this thread first page

http://www.tomshardware.co.uk/forum/id-2986517/discussion-amd-zen.html#18062733

I even used this image for illustration of the concept

Theory with no working proof nor does your first link work. Plus do you relize what your saying really doesn't block a possible 12 and 16 core. Really just makes it a possible but one should look at the AM4 socket size. Why would AMD just do what you suggest when not near the size limit of a single die?

The proof is on your own links, which disprove your claims and state clearly that there is only 8-core Zen dies.

I am not discussing what is technically possible or not. Current technology allows to build a 32-core die, but AMD is not building any. AMD is only fabricating 8-core dies for economic reasons and then using multi-die packages for the higher core count chips:

- AM4 socket (Summit Ridge): one die; up to 8 core CPU
- SP4 socket (Naples): four dies; up to 8x4 core CPU
- SP3 socket (Snowy Owl): two dies; up to 8x2 core CPU

The Consumer/Prosumer socket is AM4. The other sockets are for server Opteron.

Also I don't think that the AM4 socket is so big like you claim. It has about same size than FM2+ or AM3+ sockets.

The only correct info in that link is the SP4 being both 8 and 16 core compatible. It make the mistake of zeppline being 8 core instead of newer links correcting it to the 4 core's. There are so many links that clearly point out SP4 will work with 4, 8, 12, and 16 core processors with 4 and 8 being drop in compatiable. Also many links say the SP4 16 core isnt compatible with Naples. There is an enthusiast 12 and 16 core Zen+ core plateform with leaks suggesting launch in 2H of 2017.

 

[elbert]

It really does as future CPU's are locked at windows 10.

The CPUs will be "locked' to W10" are the consumer 4-core and 8-core Zen. The 16-core APUs for HPC and the 32-core CPUs for servers will evidently use linux.

Thus threads will really start to matter as we've seen thread improvements all the way up to 8 cores on windows 10 games.

Last reviews I see DX12 games scaling up to 4 cores

for most gamers a high clocked 4C CPU will be better than a low clocked 8C CPU.

I see scaling all the way to 6 cores even in your example proof. A 3% gain is a 3% gain.

The first benchmark puts it clearly in text form: "# of cores before scaling stops: 4 cores". And the second benchmark shows an 3.5GHz 6MB 4C/4T CPU giving virtually the same performance than the 3.5GHz 15MB 6C/12T CPU. A 3% gain is not only statistically insignificant but unnoticeable during gaming. I think they would find that an 4GHz 8MB 4C/8T CPU would be faster if they had tested one.

Tho there is other tests that show different results.

To get the context, let us read the review from where you got this image:

Ashes of the Singularity is probably a best-case scenario for DirectX 12. It’s been in development for a long time, with much thought put into supporting all the goodness of the CPU cores in today’s PCs.

But not all games are going to be Ashes of the Singularity, even if they carry the DirectX 12 label.

And then they provide us this other game benchmark proving their point

This image is from a talk given by Kronos group at GDC 2014. It is about Vulkan and it is a simulation using a simulated infinitely fast GPU running Ashes of the Singularity. I provided you a real benchmark of this game showing a different picture than in simulations of unreal hardware (no one has an infinitely fast GPU). Also durng the talk they said that not all games will behave as in that simulation. They only talked about "Certain types of games". In any case, here you have another testing of the Ashes of the Singularity game confirming my points

I think the same: even in the best case scenario a high clocked quad-core performs the same than a low clocked octo-core.

You dont understnad what all is at play. Not only is it a lower clocked octal nearly at a tie but all so 2 gen back octal. The 8 core 5960X is at a huge disadvantage.both speed and core logic. Put the 5960X up against the 4770k.

 

[juanrga]

But for the server market and to a lesser extent to prosumer market more cores and more multithreaded performance are important.

Eight core Zen never was and never will be a gaming chip. That is what quad core Zen is for (and six core if we get it.)

If that was true, IBM wouldn't be selling >4GHz 12-core server chips but BlueGene manycores at 850Mhz, and Intel would be selling the 72-core Xeon Phi for servers instead the 22-core Broadwell Xeons. And AMD would need to develop Zen, but just pack 64 or 128 Jaguar cores at 14nm.

Quad core Zen is not a gaming chip either. It is just failed dies that didn't pass server qualification and are reused for consumer/prosumer market. The quad-core Zen Pro CPUs in the roadmap that I posted days ago will not be used for gaming.

 

[juanrga]

The major problem with Zen is the never ending delays. In June AMD CEO demo'ed Zen and let slip an August. Later leaks moved it to Oct and then Late 4th quarter with only OEM's getting supply. Now leaks say February which means without question against Kaby Lake which will without quest put the 4 core Zen at a disadvantage against even the i5's.
http://fudzilla.com/news/processors/41598-amd-zen-desktop-might-slip-to-q1-17

What delays? I have been saying during last two years that Zen was a 2017 product. The delay only exists for people that believed that Zen was a 2016 product.

 

[juanrga]

There is an enthusiast 12 and 16 core Zen+ core plateform with leaks suggesting launch in 2H of 2017.

That is simply false, as proven before.

 

[elbert]

The major problem with Zen is the never ending delays. In June AMD CEO demo'ed Zen and let slip an August. Later leaks moved it to Oct and then Late 4th quarter with only OEM's getting supply. Now leaks say February which means without question against Kaby Lake which will without quest put the 4 core Zen at a disadvantage against even the i5's.
http://fudzilla.com/news/processors/41598-amd-zen-desktop-might-slip-to-q1-17

What delays? I have been saying during last two years that Zen was a 2017 product. The delay only exists for people that believed that Zen was a 2016 product.

Good to know your crystal ball is clear. Still doesn't help the fact Zen is taking to long and will end up against Kaby lake quads and skylake-E.

 

[juanrga]

In any case, here you have another testing of the Ashes of the Singularity game confirming my points

I think the same: even in the best case scenario a high clocked quad-core performs the same than a low clocked octo-core.

You dont understnad what all is at play. Not only is it a lower clocked octal nearly at a tie but all so 2 gen back octal. The 8 core 5960X is at a huge disadvantage.both speed and core logic. Put the 5960X up against the 4770k.

Your point is irrelevant. Because we don't have to compare Haswell octo-core to quad-core Haswell just to please you, but we compare chips are available on the market. Skylake is available, therefore...

Also the 4770k is not the faster quad-core Haswell. The faster is the 4790k which has 1GHz more than the octo-core Haswell. Precisely that was my point: higher-clocked quad-cores will be the favorite for most gamers.

 

[juanrga]

The major problem with Zen is the never ending delays. In June AMD CEO demo'ed Zen and let slip an August. Later leaks moved it to Oct and then Late 4th quarter with only OEM's getting supply. Now leaks say February which means without question against Kaby Lake which will without quest put the 4 core Zen at a disadvantage against even the i5's.
http://fudzilla.com/news/processors/41598-amd-zen-desktop-might-slip-to-q1-17

What delays? I have been saying during last two years that Zen was a 2017 product. The delay only exists for people that believed that Zen was a 2016 product.

Good to know your crystal ball is clear. Still doesn't help the fact Zen is taking to long and will end up against Kaby lake quads and skylake-E.

What crystal ball? I provided a link to a linkedin profile of an AMD engineer that worked on Zen and stated it was a 2017 product. I also gave a link to the interview made to the Zen team, where it was stated that Zen would first appear in 2017 products.

Main target for AMD is server and Zen is before all a server design. AMD is not attacking first the HEDT market because they think that HEDT market is worth, but because AMD will have lots of 8-core dies not qualified for servers, and those dies cannot be reused for laptops, tablets, or phones.

 

[elbert]

In any case, here you have another testing of the Ashes of the Singularity game confirming my points

I think the same: even in the best case scenario a high clocked quad-core performs the same than a low clocked octo-core.

You dont understnad what all is at play. Not only is it a lower clocked octal nearly at a tie but all so 2 gen back octal. The 8 core 5960X is at a huge disadvantage.both speed and core logic. Put the 5960X up against the 4770k.

Your point is irrelevant. Because we don't have to compare Haswell octo-core to quad-core Haswell just to please you, but we compare chips are available on the market. Skylake is available, therefore...

Also the 4770k is not the faster quad-core Haswell. The faster is the 4790k which has 1GHz more than the octo-core Haswell. Precisely that was my point: higher-clocked quad-cores will be the favorite for most gamers.

By that reasoning you could just compare the 6700k vs FX8100 to declare more than 4 cores degrades scalable. Try an apples to apples compare next time.

 

[juanrga]

Spoiler

In any case, here you have another testing of the Ashes of the Singularity game confirming my points

I think the same: even in the best case scenario a high clocked quad-core performs the same than a low clocked octo-core.

You dont understnad what all is at play. Not only is it a lower clocked octal nearly at a tie but all so 2 gen back octal. The 8 core 5960X is at a huge disadvantage.both speed and core logic. Put the 5960X up against the 4770k.

Your point is irrelevant. Because we don't have to compare Haswell octo-core to quad-core Haswell just to please you, but we compare chips are available on the market. Skylake is available, therefore...

Also the 4770k is not the faster quad-core Haswell. The faster is the 4790k which has 1GHz more than the octo-core Haswell. Precisely that was my point: higher-clocked quad-cores will be the favorite for most gamers.

By that reasoning you could just compare the 6700k vs FX8100 to declare more than 4 cores degrades scalable. Try an apples to apples compare next time.

I wouldn't declare what you say, because I know that performance is a function of IPC, frequency, and number of cores.

There are three problems with that core/thread performance graph you insist on posting again and again. The first problem is that it ignores IPC and frequency and treats both as constants.

The second is that it is not representative of all DX12 games; indeed, as stated in the same review from where you got the graph (I have bolded the relevant parts)

Ashes of the Singularity is probably a best-case scenario for DirectX 12. It’s been in development for a long time, with much thought put into supporting all the goodness of the CPU cores in today’s PCs.

But not all games are going to be Ashes of the Singularity, even if they carry the DirectX 12 label.

The third problem is that no one will purchase an octo-core chip and will disable four cores to play with benchmarks. People will see game benchmarks comparing real octo-cores to real quad-cores, will see that higher clocked quad-cores are so fast or even faster than low clocked octo-cores and will purchase the quads.

 

[-Fran-]

Juan, you absolutely missed the point of the "infinitely fast GPU" in the Vulkan demo.

Cheers!

 

[elbert]

Spoiler

In any case, here you have another testing of the Ashes of the Singularity game confirming my points

I think the same: even in the best case scenario a high clocked quad-core performs the same than a low clocked octo-core.

You dont understnad what all is at play. Not only is it a lower clocked octal nearly at a tie but all so 2 gen back octal. The 8 core 5960X is at a huge disadvantage.both speed and core logic. Put the 5960X up against the 4770k.

Your point is irrelevant. Because we don't have to compare Haswell octo-core to quad-core Haswell just to please you, but we compare chips are available on the market. Skylake is available, therefore...

Also the 4770k is not the faster quad-core Haswell. The faster is the 4790k which has 1GHz more than the octo-core Haswell. Precisely that was my point: higher-clocked quad-cores will be the favorite for most gamers.

By that reasoning you could just compare the 6700k vs FX8100 to declare more than 4 cores degrades scalable. Try an apples to apples compare next time.

I wouldn't declare what you say, because I know that performance is a function of IPC, frequency, and number of cores.

There are three problems with that core/thread performance graph you insist on posting again and again. The first problem is that it ignores IPC and frequency and treats both as constants.

The second is that it is not representative of all DX12 games; indeed, as stated in the same review from where you got the graph (I have bolded the relevant parts)

Ashes of the Singularity is probably a best-case scenario for DirectX 12. It’s been in development for a long time, with much thought put into supporting all the goodness of the CPU cores in today’s PCs.

But not all games are going to be Ashes of the Singularity, even if they carry the DirectX 12 label.

The third problem is that no one will purchase an octo-core chip and will disable four cores to play with benchmarks. People will see game benchmarks comparing real octo-cores to real quad-cores, will see that higher clocked quad-cores are so fast or even faster than low clocked octo-cores and will purchase the quads.

Yes real relevent 8 core 6900 broadwell vs 6700k. If you cant then seek out the next reasonable apples to apples compare 5960X vs 4790K.

 

[juanrga]

Juan, you absolutely missed the point of the "infinitely fast GPU" in the Vulkan demo.

I am rather sure I didn't.

 

[-Fran-]

Juan, you absolutely missed the point of the "infinitely fast GPU" in the Vulkan demo.

I am rather sure I didn't.

If you raise the restriction of a component being a bottleneck, you can see how far the other components can go. In this case, they where showing how wide the CPU can get using the full capabilities of the API.

That in turn means "yeah, you can go well beyond 4 cores". I do not know the upper limit though.

Cheers!

 

[elbert]

We've got a bit off subject. Lets get back on Zen. Many slides have it launching 2016. Yet latest leaks are saying Feb 2017.