AMD Officially Announces AMD Ryzen Threadripper CPUs For Consumers, Up to 16 Cores and 32 Threads

[Tomsen]

39 minutes ago, Dylanc1500 said:

The issues were with the first (Merced) but the issues with the wide instructions and large cache caused by the memory subsystem. Most of the issues were relieved with (McKinley). I do understand.

itanium didn't get a large following because in 03 AMD came out with the 64bit x86 extension which was  an easier upgrade due to native legacy compatibility.

That really wasn't the biggest issues at all. Itanium is an entirely different architecture than x86 is. IItanium implemented EPIC ( Explicitly Parallel Instruction Computing), which means the compiler/software is responsible to abstract parallel execution in their instruction stream (code). The cores were very wide too (6-12 issue wide), which showed itaniums major performance draw backs in low IPC workloads (java environment, transactional databases, etc).

 

Itanium didn't get a large following because it sucked outside a few special cases. Two of the biggest tech giants at the time was pushing it through, and the market wanted nothing of it. The only reason it was kept alive for so long was because of contracts.

 

1 minute ago, Drak3 said:

Note that even though the hardware might not be native 8/16bit, it is still part of modern day computing. For it to dissapear completely for modern day computing, it has to be totally abandoned. Same with 32bit, even if we move to hardware that is only 64bit, it'll still be compatible with 32bit, or it will fail.

I never said it had to disappear completely, just that the native hardware support isn't necessary.

[MadyTehWolfie]

3 minutes ago, The Benjamins said:

well the designed it to scale to multiple markets, that's why it is not the top dog in game/single threaded tasks. but now they can gain market share and work on improving clocks and IPC. it might not of been the best move to make the best gaming CPU but it was the best move to build the company back up.

That's why I want them to destroy Intel's i7 with only 4 cores and hyperthreading. To give only gaming people something really powerful for games.

[Tomsen]

1 minute ago, MadyTehWolfie said:

That's why I want them to destroy Intel's i7 with only 4 cores and hyperthreading. To give only gaming people something really powerful for games.

Or maybe just work in collaboration with game developers to have better support for more cores?

[The Benjamins]

1 minute ago, MadyTehWolfie said:

That's why I want them to destroy Intel's i7 with only 4 cores and hyperthreading. To give only gaming people something really powerful for games.

ya but they don't have the money to develop a CPU to a minority, they first have to fill their product stack, and gain market share before they can even consider doing more specialized design. the best you will get in 5 years is clock and IPC gains on zen 2 & 3.

 

if they make a good boost on zen 2 then you may see the 4-8 core parts be up to intels single threaded performance.

[MadyTehWolfie]

11 minutes ago, The Benjamins said:

ya but they don't have the money to develop a CPU to a minority, they first have to fill their product stack, and gain market share before they can even consider doing more specialized design. the best you will get in 5 years is clock and IPC gains on zen 2 & 3.

 

if they make a good boost on zen 2 then you may see the 4-8 core parts be up to intels single threaded performance.

Except Intel will just do the same thing with bigger pockets.

[Energycore]

6 hours ago, David89 said:

Oh - wait...at 5 GHz the 7700K does use pretty much the same amount of power than the FX...

Um, no? The FX-9590 is 220W. You can't pull that off a 7700K on air or water

[The Benjamins]

1 minute ago, MadyTehWolfie said:

Except Intel will just do the same thing with bigger pockets.

But they haven't. they haven't developed tech for modularizing their CPU/CPUGPU designs. they can't just slap pieces together, when AMD has. AMD has made a white paper of a APU design with 8 4 core clusters, 8 GPU clusters with 4 stacks of HBM per GPU cluster for servers with a target of 100 Tflops per chip. Neither Intel or Nvida has technologies that can do that. 

[Dylanc1500]

3 minutes ago, Tomsen said:

That really wasn't the biggest issues at all. Itanium is an entirely different architecture than x86 is. IItanium implemented EPIC ( Explicitly Parallel Instruction Computing), which means the compiler/software is responsible to abstract parallel execution in their instruction stream (code). The cores were very wide too (6-12 issue wide), which showed itaniums major performance draw backs in low IPC workloads (java environment, transactional databases, etc).

 

Itanium didn't get a large following because it sucked outside a few special cases. Two of the biggest tech giants at the time was pushing it through, and the market wanted nothing of it. The only reason it was kept alive for so long was because of contracts.

I know its an entirely different architecture, that was why I said

Quote

Itanium didn't get a large following because in 03 AMD came out with the 64bit x86 extension which was an easier upgrade due to native legacy compatibility.

I have had to deal with them myself being a database developer.

 

The problem arrived in that it was difficult to implement them and was entirely different way to compute. It took away the complexity of the hardware and allowed for the focus of the computation itself on hardware. The idea would have worked well, and it did. You are able to complete multiple instructions with a single instruction word. It required a different approach and different way of computing, just like RISC. It was only pushed for those specific use cases doesn't mean it couldn't have been developed for, just that it required a different approach than what was commonly used.

 

It doesn't honestly really matter, either way it was given up on. It just had a possibility of becoming something much better if it had been better adopted. Backwards compatibility is too important to a lot of the industry. (It is why i can boot DOS on my system)

[MadyTehWolfie]

9 minutes ago, The Benjamins said:

But they haven't. they haven't developed tech for modularizing their CPU/CPUGPU designs. they can't just slap pieces together, when AMD has. AMD has made a white paper of a APU design with 8 4 core clusters, 8 GPU clusters with 4 stacks of HBM per GPU cluster for servers with a target of 100 Tflops per chip. Neither Intel or Nvida has technologies that can do that. 

Thought we were talking CPUs not APU's. Was saying Intel could just put strong single cores in a 6-8 core if that's what AMD does. For instance if they made their single core on par with the cores of the 7700k just add more of them. Probably misunderstood your post.

[SpaceGhostC2C]

25 minutes ago, Energycore said:

Um, no? The FX-9590 is 220W. You can't pull that off a 7700K on air or water

So, have you measured the power draw by a 7700K at 5GHz? 

(also, remember that 220W was a TDP rating, not a power consumption measurement. The 9370 had the same rating, yet it won't need much more than 180W at full load). 

[DocSwag]

1 hour ago, MadyTehWolfie said:

Never said it was but the fact remains games devs rarely code for more than 4 cores 4 threads. I want the to focus on making better cores instead of more cores, and maybe they'll be able to compete with the i7 range CPUs for gaming. Otherwise myself and others will just pass over ryzen for a I7. If their CPUs can't beat the top Intel 4 core I have no use for more cores. For gaming I can't recommend ryzen over the top of the line i7 right now.

Not true, as evidenced by stuff like Watch Dogs 2 and Ashes. And the fact that the 1600 is definitely around 10% ahead of the 1500x.

[Energycore]

2 minutes ago, SpaceGhostC2C said:

So, have you measured the power draw by a 7700K at 5GHz? 

(also, remember that 220W was a TDP rating, not a power consumption measurement. The 9370 had the same rating, yet it won't need much more than 180W at full load). 

Care to show me sauce?

 

Looks like 250W for a 9590 and around 200W for a 7700K at 5GHz.

Spoiler

http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/62166-amd-fx-9590-review-piledriver-5ghz-17.html

https://www.bit-tech.net/hardware/cpus/2016/12/05/core-i7-7700k-performance-overclock-preview/5

It's always good to provide multiple sources though.

[Taf the Ghost]

45 minutes ago, Tomsen said:

Or maybe just work in collaboration with game developers to have better support for more cores?

It's a tad more about the Driver implementations.  Moving to Vulkan/DX12 will all but eliminate the major differences.

[MadyTehWolfie]

13 minutes ago, DocSwag said:

Not true, as evidenced by stuff like Watch Dogs 2 and Ashes. And the fact that the 1600 is definitely around 10% ahead of the 1500x.

Watch dogs was optimized for more cores and the other is a synthetic benchmark. Don't really care what AMD CPU beats what AMD CPU because I'm talking about amd making a 4 core that beats Intel 4 cores

[Taf the Ghost]

The gaming difference is less about the 4c8t of the 7700k than the really high Turbo Boost of a single core. This is, actually, most of the reason for the bigger differences in 1080p gaming on a Nvidia 1080 or 1080 Ti.  (Most of the reviewers singing the praises of Ryzen were on a 1070. There seems to be a Saturation level that Ryzen can't maintain under the current Nvidia full Pascal stack. I'd be curious of some 980 Ti testing as well.)  One of the Russian, I believe, reviewers had a Ryzen test in a game that showed average thread usage. One pegged over 95% and the rest down around 20%.  

 

As it currently stands, unless you can pair a 7700k with a >$500 USD GPU, the gaming difference doesn't exist between the entire Ryzen 6c or 8c stack and the 7600 or 7700 versions.

 

The 7350k and the 2600k are still really good gaming chips because of that Thread Usage issue.  Neither of those should be competitive in any game, but they very definitely are right now.

[SpaceGhostC2C]

19 minutes ago, Energycore said:

Care to show me sauce?

OK, I guess:

Spoiler

 

My sources are my 9370 and my Zalman PCM-1. It's no longer paired with the 9370, and the 9370 got OC'ed a bit since then. But I have it now with an 8370E OC'ed to 4.4, and it's a similar story.

Nothing surprising, though, since TDP rating was the same for 9370 and 9590, while there was no way power consumption was going to be the same.

 

For TDP not being a measure of power consumption I need no source, I guess  

 

Quote

 

Looks like 250W for a 9590 and around 200W for a 7700K at 5GHz.

  Hide contents

http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/62166-amd-fx-9590-review-piledriver-5ghz-17.html

https://www.bit-tech.net/hardware/cpus/2016/12/05/core-i7-7700k-performance-overclock-preview/5

 

It's unfortunate that "professional reviewers" can't be arsed to use proper tools, and we are left with Kill-a-Watt measures instead (oddly enough, the first link goes into great detail on how they added a line conditioner for "accurate measurement"... at the wall ).

Nevertheless, I'm willing to believe a rough 50W difference between the two OC'ed chips, and 200W seems the right ballpark to me.

[DocSwag]

44 minutes ago, MadyTehWolfie said:

Watch dogs was optimized for more cores and the other is a synthetic benchmark. Don't really care what AMD CPU beats what AMD CPU because I'm talking about amd making a 4 core that beats Intel 4 cores

4c 4t CPUs are dead. Just dead. There's a reason i7s beat i5s.

 

Threads are the future. I was just pointing out some games that can make use of a lot of cores well. Others can also make use of a lot of cores, just not as well.

[Tomsen]

34 minutes ago, Dylanc1500 said:

I know its an entirely different architecture, that was why I said

I have had to deal with them myself being a database developer.

 

The problem arrived in that it was difficult to implement them and was entirely different way to compute. It took away the complexity of the hardware and allowed for the focus of the computation itself on hardware. The idea would have worked well, and it did. You are able to complete multiple instructions with a single instruction word. It required a different approach and different way of computing, just like RISC. It was only pushed for those specific use cases doesn't mean it couldn't have been developed for, just that it required a different approach than what was commonly used.

 

It doesn't honestly really matter, either way it was given up on. It just had a possibility of becoming something much better if it had been better adopted. Backwards compatibility is too important to a lot of the industry. (It is why i can boot DOS on my system)

The idea clearly didn't work out well. The whole issue, as you also pointed out, is that it isn't easy to optimize for. For most workloads it will naturally fall behind x86/ARM, that is because itanium relies heavily upon the compiler to properly align the instruction works in such a fashion that it doesn't conflict with the applications actual intend. In my opinion, it is smarter to have the processor itself to abstract such things with out-of-order execution engine, than having to explicitly be optimized by the software running.

 

In many ways it is similar to bulldozer (and other failed microarchitectures), it excels it special cases (software that complements the hardware), but have major performance draw backs in general cases. It simply isn't as reliable in its performance throughput for a wide range of applications.

 

14 minutes ago, Taf the Ghost said:

It's a tad more about the Driver implementations.  Moving to Vulkan/DX12 will all but eliminate the major differences.

DX12 and Vulkan doesn't just magically solved game developers game code. It makes it easier for certain parts (much better multi-threaded rendering engine, etc), but you can still have "DX11 efficiency" with DX12 implementation. We saw that with some of the early DX12 titles after all.

 

Both AMD and Nvidia (nvidia more so) invest much resources into development houses to optimize titles.

[MadyTehWolfie]

46 minutes ago, DocSwag said:

4c 4t CPUs are dead. Just dead. There's a reason i7s beat i5s.

 

Threads are the future. I was just pointing out some games that can make use of a lot of cores well. Others can also make use of a lot of cores, just not as well.

4c 4t is what the 7700k has. 4 physical cores and 4 logical ones making 8 in total. Don't think the 7700k has 4 core then two threads per core. Maybe wrong though.

[Drak3]

2 hours ago, MadyTehWolfie said:

That's why I want them to destroy Intel's i7 with only 4 cores and hyperthreading. To give only gaming people something really powerful for games.

Again, the 1700, which is a weaker chip for single core performance, due to both IPC and clock deficits (a high OC for any Ryzen chip is 4.1GHz, whereas the 7700K starts at 4.2 and the average for OC's under water is 5GHz), yet they're basically delivering the same experience, as per Level1Tech's double blind test, so long as one invests in RAM that can hit good clocks as well.

AMD pushing out something really powerful now, wouldn't do us much. Not until games are actually demanding more than chips like the 4790K, 6700K, 7700K.

1 minute ago, MadyTehWolfie said:

4c 4t is what the 7700k has. 4 physical cores and 4 logical ones making 8 in total. Don't think the 7700k has 4 core then two threads per core. Maybe wrong though.

The 7700K is 4c/8t. 4 physical cores total, capable of doing 8 "threads" total, two "threads" per core. Of course, "threads" in this context meaning logical core.

[DocSwag]

34 minutes ago, MadyTehWolfie said:

4c 4t is what the 7700k has. 4 physical cores and 4 logical ones making 8 in total. Don't think the 7700k has 4 core then two threads per core. Maybe wrong though.

When you say 4t, that refers to total threads. 7700k has 8 threads, so you say 4c 8t. i5s are 4c 4t. 

 

7700k does have two threads per core, that's what hyperthreading is... I have no idea what you mean 4 physical and 4 logical making 8 in total, that would make 4. Physical and logical are different measurements.

[alamox]

my guess on prices for the X variants, and regular non OC cpus a 100$ less

10cores starts at 799$

12 cores at 999$

14 cores  at 1199$

and 16 cores at 1499$

 

[Jito463]

 

1 hour ago, Drak3 said:

1 hour ago, MadyTehWolfie said:

4c 4t is what the 7700k has. 4 physical cores and 4 logical ones making 8 in total. Don't think the 7700k has 4 core then two threads per core. Maybe wrong though.

The 7700K is 4c/8t. 4 physical cores total, capable of doing 8 "threads" total, two "threads" per core. Of course, "threads" in this context meaning logical core.

@MadyTehWolfie

I wondered why you wanted a 4c/4t CPU from AMD.  I was literally scratching my head in confusion, heh.

[Drak3]

Just now, Jito463 said:

 

@MadyTehWolfie

I wondered why you wanted a 4c/4t CPU from AMD.  I was literally scratching my head in confusion, heh.

Budget oriented options are still welcome, and the R3's are going to be quadcores, assuming there's more than one model. If not, the R3 is going to be a quadcore.

[Jito463]

Just now, Drak3 said:

Budget oriented options are still welcome, and the R3's are going to be quadcores, assuming there's more than one model. If not, the R3 is going to be a quadcore.

I know, but the way it was posted was intended as a gaming CPU alternative to Intel, not a budget CPU.