UPDATED AGAIN: New Ryzen prices and clock speeds rumoured: lowest-end 8c/16t for $322, lowest-end 6c/12t for $223

[captain cactus]

Well, it seems that VideoCardz has an 8c16t at 4.0GHz. And apparently it does good.

 

 

[LAwLz]

9 hours ago, leadeater said:

Also AMD is pretty hot on pushing a lot of FP tasks to GPUs using OpenCL meaning if AMD can get some really helpful tools out to developers FP tasks that would be weak on Ryzen could be pushed to GPU.

I really hope this is not the case. If it is then I have no idea what AMD is thinking. Was everyone at the entire company dropped as babies?

YOU SHOULD NOT RELEASE A PRODUCT THAT RELIES ON DEVELOPERS CHANGING THEIR CODE TO BE GOOD.

 

This would be the exact same mistake they made with Bulldozer. Like, quite literally the same. Bulldozer had crippled FP performance (along with other issues) because AMD was betting on developers using OpenCL and other tools to push that type of workload to the GPUs. And what happened? Developers didn't bother and Bulldozer flopped (no pun intended).

 

Zen will be compared to Intel CPUs using programs that are available today, not programs someone with a time-machine brought back from the future.

If AMD is making their FP weak again in the hopes that developers will push that workload to GPUs, then quite frankly they deserve to go bankrupt.

They need to be competitive or better than Intel across the board.

[Mr.Meerkat]

4 hours ago, lots of unexplainable lag said:

Well, it seems that VideoCardz has an 8c16t at 4.0GHz. And apparently it does good.

0.9-1V? 65amps is quite high no? 

[zMeul]

21 minutes ago, LAwLz said:

I really hope this is not the case. If it is then I have no idea what AMD is thinking. Was everyone at the entire company dropped as babies?

YOU SHOULD NOT RELEASE A PRODUCT THAT RELIES ON DEVELOPERS CHANGING THEIR CODE TO BE GOOD.

 

This would be the exact same mistake they made with Bulldozer. Like, quite literally the same. Bulldozer had crippled FP performance (along with other issues) because AMD was betting on developers using OpenCL and other tools to push that type of workload to the GPUs. And what happened? Developers didn't bother and Bulldozer flopped (no pun intended).

 

Zen will be compared to Intel CPUs using programs that are available today, not programs someone with a time-machine brought back from the future.

If AMD is making their FP weak again in the hopes that developers will push that workload to GPUs, then quite frankly they deserve to go bankrupt.

They need to be competitive or better than Intel across the board.

this makes me wonder what exactly rendered those Blender tests from AMD's PR presentation months ago

if they did it on a Radeon graphics instead of the CPU ... oh boy 

 

I also heard Zen lacks some AVX sets?! if true, they are fuuucked

[Jahramika]

On 2/8/2017 at 8:38 AM, leadeater said:

Could be power limited via the CPU microcode, or something similar. Not read the source just listing a possibility to the mentioned power limiting of lower chips. 

It is not power limited all chips are unlocked. It will be based on the motherboard you buy X370 will let you increase power and overclock past the boost clock.

[leadeater]

4 hours ago, LAwLz said:

I really hope this is not the case. If it is then I have no idea what AMD is thinking. Was everyone at the entire company dropped as babies?

YOU SHOULD NOT RELEASE A PRODUCT THAT RELIES ON DEVELOPERS CHANGING THEIR CODE TO BE GOOD.

 

This would be the exact same mistake they made with Bulldozer. Like, quite literally the same. Bulldozer had crippled FP performance (along with other issues) because AMD was betting on developers using OpenCL and other tools to push that type of workload to the GPUs. And what happened? Developers didn't bother and Bulldozer flopped (no pun intended).

 

Zen will be compared to Intel CPUs using programs that are available today, not programs someone with a time-machine brought back from the future.

If AMD is making their FP weak again in the hopes that developers will push that workload to GPUs, then quite frankly they deserve to go bankrupt.

They need to be competitive or better than Intel across the board.

Well it's not like FP will be weak, it might be or it might not be. It may also only be slightly less than Intel, by how much is the key. Just keep in mind AMD has been looking at OpenCL & HSA for some time and it isn't a bad idea, it's actually a very good idea if done correctly which in lies the problem. It flopped for a number of reasons in the past, most of them explained in the linked Reddit: No DX12/Vulkan, bad Async computer on PCs, immature tools etc. Most of these are now solved and don't forget this is already being done on the console side by the very same developers.

 

And to be fair all hardware relies on developers properly making use of the instruction sets and APIs which historically in the gaming industry has been rather lack luster.

 

Also you don't design new hardware and architectures solely for current and past workloads, future planning is required and that is a balance that needs to be done with great care else you already know what will happen.

[leadeater]

4 hours ago, zMeul said:

this makes me wonder what exactly rendered those Blender tests from AMD's PR presentation months ago

if they did it on a Radeon graphics instead of the CPU ... oh boy 

 

I also heard Zen lacks some AVX sets?! if true, they are fuuucked

Quote

Intel does retain two slight advantages relative to Zen: first, the slight memory-op advantage above; second, Skylake can execute two simple 256-bit AVX operations of the same type per cycle, whereas Zen can only do one (dividing it into two 128-bit halves). So Intel might still be able to win, by a substantial margin, on some carefully-selected benchmarks. However, I think Zen will show better or at least equal performance on a broad range of real-world tasks, and probably at a much lower price.

https://www.reddit.com/r/Amd/comments/5mklpw/ryzen_vega_bristol_ridge_am4_a_summary_and/

 

Also AVX has to actually be used for this to matter, and in the exact way to give the advantage.

 

[leadeater]

4 hours ago, Jahramika said:

It is not power limited all chips are unlocked. It will be based on the motherboard you buy X370 will let you increase power and overclock past the boost clock.

Unlocked multiplier yes, but what explains such a huge difference in TDP. Motherboard has nothing to do with the rated TDP of a CPU.

[Sakkura]

5 hours ago, zMeul said:

I also heard Zen lacks some AVX sets?! if true, they are fuuucked

 

31 minutes ago, leadeater said:

https://www.reddit.com/r/Amd/comments/5mklpw/ryzen_vega_bristol_ridge_am4_a_summary_and/

 

Also AVX has to actually be used for this to matter, and in the exact way to give the advantage.

 

Yeah AMD might be fuuucked in like 2025 when software really starts to rely on these instructions.

 

Still, gotta try to keep my excitement in check. Less than a month until real numbers.

[ARikozuM]

On 2/8/2017 at 8:32 AM, TOMPPIX said:

1700: 8/16 @ 3.7Ghz (65W) or is that not how it works?

 

On 2/8/2017 at 8:33 AM, Ashaira said:

that is how much power the cpu uses at stock speeds. not how much it can use

TDP is heat dissipated, not the wattage used.

[vanished]

7 minutes ago, ARikozuM said:

 

TDP is heat dissipated, not the wattage used.

Well, those should be virtually identical in theory.

 

The real issue with assuming TDP = power consumed is that TDP isn't an exact measurement, it's a rough upper bound the manufacturer decided on and often the same number is slapped on a whole group of products that actually differ quite greatly.

[leadeater]

14 minutes ago, Sakkura said:

Still, gotta try to keep my excitement in check. Less than a month until real numbers.

Just keep expecting it to perform slightly less than Intel then you can't be disappointed, however you can be surprised and made happy 

[leadeater]

2 minutes ago, Ryan_Vickers said:

Well, those should be virtually identical in theory.

 

The real issue with assuming TDP = power consumed is that TDP isn't an exact measurement, it's a rough upper bound the manufacturer decided on and often the same number is slapped on a whole group of products that actually differ quite greatly.

 

Quote

The thermal design power (TDP), sometimes called thermal design point, is the maximum amount of heat generated by a computer chip or component (often the CPU or GPU) that the cooling system in a computer is designed to dissipate in typical operation. Rather than specifying CPU's real power dissipation, TDP serves as the nominal value for designing CPU cooling systems.^[1]

The TDP is typically not the largest amount of heat the CPU could ever generate (peak power), such as by running a power virus, but rather the maximum amount of heat that it would generate when running "real applications." This ensures the computer will be able to handle essentially all applications without exceeding its thermal envelope, or requiring a cooling system for the maximum theoretical power (which would cost more but in favor of extra headroom for processing power).^[2]

Some sources state that the peak power for a microprocessor is usually 1.5 times the TDP rating.^[3] However, the TDP is a conventional figure while its measurement methodology has been the subject of controversy. In particular, until around 2006 AMD used to report the maximum power draw of its processors as TDP, but Intel changed this practice with the introduction of its Conroe family of processors.^[4]

A similar but more recent controversy has involved the power TDP measurements of some Ivy Bridge Y-series processors, with which Intel has introduced a new metric called scenario design power (SDP).^[5][6

https://en.wikipedia.org/wiki/Thermal_design_power

 

Also that's the first I've ever heard of SDP, dammit Intel stop making shit up. 

[vanished]

1 minute ago, leadeater said:

 

https://en.wikipedia.org/wiki/Thermal_design_power

 

Also that's the first I've ever heard of SDP, dammit Intel stop making shit up. 

Thanks for the details  Yeah, that's the idea ^^

[ARikozuM]

5 minutes ago, leadeater said:

snip

Your inner nerd is showing.

[leadeater]

Just now, ARikozuM said:

Your inner nerd is showing.

When is it not? lol.

[ARikozuM]

26 minutes ago, Sakkura said:

Yeah AMD might be fuuucked in like 2025 when software really starts to rely on these instructions.

We'll probably have quantum whateverthef#ck (technical term) by then.

[vanished]

5 minutes ago, ARikozuM said:

We'll probably have quantum whateverthef#ck (technical term) by then.

Not for home computing  

 

But it will be interesting to see where things have gone... I suspect we will finally be off silicon.

[zMeul]

2 hours ago, leadeater said:

Also AVX has to actually be used for this to matter, and in the exact way to give the advantage.

and what makes you think AVX is not used these days? except gaming where the on;y two games I know use AVX are the F1 CodeMasters game and GTA5 (I could not find official confirmation)

 

but there is a shit ton of software that uses and heavily relies on AVX, HandBrake for example, FFmpeg, everything computing

carefully selected benchmarks? more like deflecting from the issue

AMD wants to breach the enterprise sector? not with this shit they won't

[LAwLz]

3 hours ago, leadeater said:

Well it's not like FP will be weak, it might be or it might not be. It may also only be slightly less than Intel, by how much is the key. Just keep in mind AMD has been looking at OpenCL & HSA for some time and it isn't a bad idea, it's actually a very good idea if done correctly which in lies the problem. It flopped for a number of reasons in the past, most of them explained in the linked Reddit: No DX12/Vulkan, bad Async computer on PCs, immature tools etc. Most of these are now solved and don't forget this is already being done on the console side by the very same developers.

 

And to be fair all hardware relies on developers properly making use of the instruction sets and APIs which historically in the gaming industry has been rather lack luster.

 

Also you don't design new hardware and architectures solely for current and past workloads, future planning is required and that is a balance that needs to be done with great care else you already know what will happen.

 

2 hours ago, Sakkura said:

 

Yeah AMD might be fuuucked in like 2025 when software really starts to rely on these instructions.

 

Still, gotta try to keep my excitement in check. Less than a month until real numbers.

Kinda funny how you go "Ohhh but developers will surely update their code to use OpenCL" but when it comes to AVX all of a sudden it's "lol that won't happen until 2025. Developers are so slow". 

[leadeater]

14 minutes ago, zMeul said:

and what makes you think AVX is not used these days? except gaming where the on;y two games I know use AVX are the F1 CodeMasters game and GTA5 (I could not find official confirmation)

 

but there is a shit ton of software that uses and heavily relies on AVX, HandBrake for example, FFmpeg, everything computing

carefully selected benchmarks? more like deflecting from the issue

AMD wants to breach the enterprise sector? not with this shit they won't

Well yes I agree with this, not like what I quoted is my words.

 

But the caveat still remains, an Intel CPU is only better if two AVX operations of the same type are trying to be run else the difference isn't going to be significant. Intel also has AVX-512 on Xeon X5-E26xx v5 (Skylake-EP) &  Xeon Phi products so if you want to do heavy AVX stuff using these would be a better choice again.

 

Being better or worse at a single instruction set isn't a make or break either, for example if I'm spec'ing out a VM host do I care more about 32c/64t vs 28c/56t per processor or potentially better AVX performance on hardware that is going to be running a heavy mix of workloads. The answer is the CPU with 32c/64t, so long as they aren't garbage cores. Another nice thing about the Naples platform is 128 PCIe lanes. If your building a system that is going to have as many computer accelerators cards as possible plus 40Gb NICs plus FibreChannel or Infiniband this is going to help a great deal, or you could be building an NVMe storage controller where again more PCIe lanes is better than CPU performance.

 

To add to this Naples isn't the same as these desktops SKUs, I wouldn't go judging how well AMD's new server processors are based on these rumors and architecture diagrams.

[leadeater]

12 minutes ago, LAwLz said:

Kinda funny how you go "Ohhh but developers will surely update their code to use OpenCL" but when it comes to AVX all of a sudden it's "lol that won't happen until 2025. Developers are so slow". 

Well actually I agreed with what you said and in my first post about this I did say AMD would have to hand feed it to developers.

 

Plus I also said this

17 hours ago, leadeater said:

Looking at this old discussion, while we may have addressed most of the issues raised it might still not happen yet, OpenCL usage in games. The problem I see is actually Nvidia and poor async compute and HSA. Even with help from AMD the effort to do it vs the target audience isn't great.

https://www.reddit.com/r/gamedev/comments/2aabpc/why_dont_more_games_use_gpgpu_opencl/

 

There's a difference between proposing an idea or stating what you think a company might do versus what you personally think will happen.

 

Edit:

I am guilty of not always saying the second part, personal opinion.

Edited February 13, 2017 by leadeater

[zMeul]

36 minutes ago, leadeater said:

To add to this Naples isn't the same as these desktops SKUs, I wouldn't go judging how well AMD's new server processors are based on these rumors and architecture diagrams.

if Naples is Zen then the same issues will be present there

if it's not Zen, then that's another thing entirely

 

but the info so far points to Naples being based on the same Zen arch as Ryzen (god, that naming scheme .. )

[leadeater]

1 minute ago, zMeul said:

if Naples is Zen then the same issues will be present there

if it's not Zen, then there's another thing entirely

I believe it's the same Zen cores, doesn't seem likely they will be modified in any meaningful way. Don't think AMD has had the time or money for that. Only the bits around the cores are different, 4 or 8 channel memory controller (still up in the air) and more PCIe lanes.

[Space Reptile]