AMD responds to 1080p gaming tests on Ryzen. Supports ECC RAM. Win 10 SMT bug

[zMeul]

13 minutes ago, Curious Pineapple said:

It's not possible to compile high performance applications that will run on any x86 CPU without optimisations anymore. Even Intel's architecture changes significantly between generations. Code optimized for a Core 2 processor will be far from optimal on a Core i, and the 2 generations will handle the same "generic" code very differently. You wouldn't do in software what can be done in 2 CPU cycles. A 2 cycle instruction for example on an i7 may replace a routine that takes 500,000 cycles to complete without the hardware support. A good toolchain should work this out, include both, and choose which code to run based on the CPU.

yeah, actually is

because you do not understand one basic principle - there is one arch: x86

the execution lies with the CPU's scheduler, prefetcher and the rest - if your CPU deviates from the base arch then you have problems that you need to solve

 

I will repeat this a n^th time, there is no code for Bulldozer - period

the same code that you claim is "for Bulldozer" can also run on Phenom if the dependencies are met .. imagine that

you can also run the same "Bulldozer code" on a Pentium2 CPU, again .. if the dependencies are met

 

immagine for a moment that what you're saying it's true - Windows 10 would have to have a separate kernel for each CPU generation

and guess what!? it doesn't!

[Mira Yurizaki]

6 minutes ago, Curious Pineapple said:

Code optimized for a Core 2 processor will be far from optimal on a Core i, and the 2 generations will handle the same "generic" code very differently.

From what the developer has access to, I find this very hard to believe. Especially when http://www.agner.org/optimize/instruction_tables.pdf seems to suggest otherwise.

[Mira Yurizaki]

1 minute ago, zMeul said:

yeah, actually is

because you do not understand one basic principle - there is one arch: x86

the execution lies with the CPU's scheduler, prefetcher and the rest - if your CPU deviates from the base arch then you have problems that you need to solve

 

I will repeat this a n^th time, there is no code for Bulldozer - period

the same code that you claim is "for Bulldozer" can also run on Phenom if the dependencies are met .. imagine that

you can also run the same "Bulldozer code" on a Pentium2 CPU, again .. if the dependencies are met

If the software uses AVX instructions, it's not going to run on anything before Bulldozer or Sandy Bridge.

[zMeul]

Just now, M.Yurizaki said:

If the software uses AVX instructions, it's not going to run on anything before Bulldozer or Sandy Bridge.

as I said, and I'm putting a bing emphasis on it if the dependencies are met

[Mira Yurizaki]

1 minute ago, zMeul said:

as I said, and I'm putting a bing emphasis on it if the dependencies are met

And they're saying if the software uses instructions not available to an architecture, it's not going to run on that architecture.

 

So you're basically arguing the one the same coin, just on the other side. For the sake of arguing.

[zMeul]

Just now, M.Yurizaki said:

And they're saying if the software uses instructions not available to an architecture, it's not going to run on that architecture.

 

So you're basically arguing the one the same coin, just on the other side. For the sake of arguing.

you can't create code for sets that don't exists - what is this, time travel?

[Curious Pineapple]

12 minutes ago, M.Yurizaki said:

From what the developer has access to, I find this very hard to believe. Especially when http://www.agner.org/optimize/instruction_tables.pdf seems to suggest otherwise.

It depends what the application is. The newer Core processors have hardware features and instructions that the Core 2 doesn't (AVX being one you mentioned). If an application could make use of them instead of performing the task in software, it's going to be one hell of a lot faster.

 

One application that proves my point is cpuminer. There are 5 different executables: generic, SSE2, AMD, Haswell/Broadwell and, what a shock, Sandy/Ivybridge. There are 2 different versions for different generations of Intel chips, 2 generic (one requiring SSE2), and an AMD. They won't run on the wrong CPU type either.

 

Correction: They will run, but like a bag of crap

[zMeul]

5 minutes ago, Curious Pineapple said:

One application that proves my point is cpuminer. There are 5 different executables: generic, SSE2, AMD, Haswell/Broadwell and, what a shock, Sandy/Ivybridge. There are 2 different versions for different generations of Intel chips, 2 generic (one requiring SSE2), and an AMD. They won't run on the wrong CPU type either.

and that points exactly to what I said in a previous post

that is code compiled with specific hardware path and executes better on the series of CPUs - whereas generic compilers, like the ones found in VS, for example don't do that

 

what are you implying is that the game developers should've known years in advance to create specific exes for Zen - because reasons!?

[Curious Pineapple]

Just now, zMeul said:

and that points exactly to what I said in a previous post

that is code compiled with specific hardware path and executes better on the series of CPUs - whereas generic compilers, like the ones found in VS, for example don't do that

So in other words, assuming the code for the software used for testing WAS compiled in a purely generic fashion, there could very well be hardware features in Ryzen that would potentially push it's performance even higher.

[Organized]

2 minutes ago, zMeul said:

whereas generic compilers, like the ones found in VS, for example don't do that

Can you please post a reliable source for that? That VS compilers are not optimized more for the Intel side because of it's much larger market share and higher influence among the last years?

[zMeul]

Just now, Curious Pineapple said:

So in other words, assuming the code for the software used for testing WAS compiled in a purely generic fashion, there could very well be hardware features in Ryzen that would potentially push it's performance even higher.

and whose job is that? the game developer's, AMD's

[Curious Pineapple]

I think he is like a CD player trying to read a scratched disc. Just keeps playing the same noise over and over again, and no matter how much you press skip track, some times you just need to eject the disc and start again.

 

Quite frankly I have free compilers for microcontrollers that can determine if I am trying to do an operation in code that the CPU supports in hardware, and they will remove my code and get the hardware to do it instead. The Arduino does this kind of thing at damn runtime!

[zMeul]

15 minutes ago, Organized said:

Can you please post a reliable source for that? That VS compilers are not optimized more for the Intel side because of it's much larger market share and higher influence among the last years?

can you post a source that tells otherwise, because you are full of shit at this moment 

 

let's see what a bloke sais on Intel's dev forums:

oh no! Intel's compiler produces results that executes 3 times slower than the MS one

good god!

[Dabombinable]

So if I bought an 1800X and it ran Skyrim or BF4 worse at 1080p than my 4790K, the reason would be Intel bias not lackluster single threaded performance according to AMD?

[Soonercoop]

Nobody marketed this cpu as better in games than an i7 7700k. They marketed it as a 6900k competitor and the only game comparisons were for streaming applications. This cpu is worse for gaming than an i7 7700k just like a 6900k is. The hype train on this forum somehow convinced people that this lineup was a gaming platform. I really hope consumers are not as ignorant as some of the comments I've read here. If you want to game, buy a cpu with 4 cores.

[Valentyn]

3 minutes ago, Soonercoop21 said:

Nobody marketed this cpu as better in games than an i7 7700k. They marketed it as a 6900k competitor and the only game comparisons were for streaming applications. This cpu is worse for gaming than an i7 7700k just like a 6900k is. The hype train on this forum somehow convinced people that this lineup was a gaming platform. I really hope consumers are not as ignorant as some of the comments I've read here.

We started the speculation: by just trying to be close to Haswell's IPC, AMD will have a solid product.

The first leaks shows ryzen has higher IPC than haswell.

AMD Presentation shows that has IPC close to Broadwell.

LLT is like : "this cpu doesn't beat skylake on IPC and is garbage for games"

399$ for 8 haswell/broadwell cores is extremely good, but AMD bungled their launch. They apparently don't have all the micro code finished, or have Windows Drivers out yet. The same applies to motherboard manufactures not done with BIOSs yet either.

AMD should have waited another month before launching, they came out early, and fumbled on the consumer gaming side sadly; but for workstation and enterprise they got a massive win.
 

[LAwLz]

3 minutes ago, Valentyn said:

399$ for 8 haswell/broadwell cores is extremely good, but AMD bungled their launch. They apparently don't have all the micro code finished, or have Windows Drivers out yet. The same applies to motherboard manufactures not done with BIOSs yet either.

That's a very big assumption, that it will improve with future updates. This might be as good as it gets.

[Valentyn]

5 minutes ago, LAwLz said:

That's a very big assumption, that it will improve with future updates. This might be as good as it gets.

Maybe, but Joker and GamerNexus are live at the moment and discussing the difference between their motherboards and what they managed to get out Ryzen.

Joker had great results, and made sure to keep checking for motherboard updates; use them and recheck. We don't know if other reviewers did the same.

Here's their live stream, so will be worth check when it's over and up as a full video.

I'd still recommend waiting a month or two before buying anyway. I did that with my x99 system and was spared the horrible bugs they had.
Although I still dealt with RAM issues for another two months before I could run my ram at XMP speeds, as Asus were slow at updating the BIOS for it all.
 

 

His 1700 @3.9Ghz vs 7700K @5Ghz

 

 

[Soonercoop]

7 minutes ago, Valentyn said:

We started the speculation: by just trying to be close to Haswell's IPC, AMD will have a solid product.

-snip-
 

I understand they fumbled the launch that regard, but a considerable amount of people actually thought that this 8c/16t processor was targeted at gamers as a competitor to the 7700k. These people are a perfect example of what people imagine up when they hype things up like this.

[Valentyn]

1 minute ago, Soonercoop21 said:

I understand they fumbled the launch that regard, but a considerable amount of people actually thought that this 8c/16t processor was targeted at gamers as a competitor to the 7700k. These people are a perfect example of what people imagine up when they hype things up like this.

 

That's true. People getting some silly notions. There are games that'll use all the cores; like Watchdogs 2 and a few others; but the majority won't; especially older titles.

 

Will be worth waiting to see how Ryzen's 6 and 4 core processors do when they're out in two months.

[LAwLz]

1 minute ago, Valentyn said:

Maybe, but Joker and GamerNexus are live at the moment and discussing the difference between their motherboards and what they managed to get out Ryzen.

Joker had great results, and made sure to keep checking for motherboard updates; use them and recheck. We don't know if other reviewers did the same.

Here's their live stream, so will be worth check when it's over and up as a full video.

 

His 1700 @3.9Ghz vs 7700K @5Ghz

Joker's results are outliers. They should be ignored.

If all but one review shows consistent results, then you ignore the inconsistent one.

 

Gamers Nexus also said in the stream that they used the latest BIOS versions, and even contacted motherboard manufacturers directly to talk about their BIOS versions.

[zMeul]

16 minutes ago, LAwLz said:

Joker's results are outliers. They should be ignored.

If all but one review shows consistent results, then you ignore the inconsistent one.

 

Gamers Nexus also said in the stream that they used the latest BIOS versions, and even contacted motherboard manufacturers directly to talk about their BIOS versions.

there's one thing odd about Joker's testing and setup

he said he has a GigaByte mobo while Nexus has an ASUS one

 

the thing they talk about, the microcode updates .. well ... GigaByte's latest BIOS updates are dated 20th last month while ASUS has BIOS updates as new as yesterday (2 days for me as of 45minutes ago)

and yet Joker's results are higher

[anotherriddle]

Here is also a comment from Level1Techs on ECC Memory

seems like it works, but I am sure they will do more tests.

And I can tell you, it can be really hard to be sure ECC really works, you Kind of have to take the word of the Software.

[Valentyn]

12 minutes ago, zMeul said:

there's one thing odd about Joker's testing and setup

he said he has a GigaByte mobo while Nexus has an ASUS one

 

the thing they talk about, the microcode updates .. well ... GigaByte's latest BIOS updates are dated 20th last month while ASUS has BIOS updates as new as yesterday (2 days for me as of 45minutes ago)

and yet Joker's results are higher

 

From Techspot, seems they had to dump their Asus mobo due to issues. They ended up using the Asrock x370 Taichi

 

http://www.techspot.com/review/1345-amd-ryzen-7-1800x-1700x/page7.html
 

Quote

Scorpus

 

I will put this here for people wondering about the gaming results:
We had a lot of trouble benchmarking games with Ryzen. It seems we weren't the only ones, and many other reviewers have reported strangely low performance here. Our initial Asus board was plagued with bugs, and we saw some gains simply by switching to a Gigabyte or Asrock board. This really isn't the sort of behavior you'd expect, and AMD even acknowledged there were some issues with some Asus boards.
While we are pretty confident in our application test results, there could be some unresolved early issues with Ryzen and AM4 boards that is leading to strangely low gaming performance. We're not 100% sure what is going on there; Steve and I spent a while discussing what could be up, and we ended up confused more than anything else.
So if you're a gamer that's looking at our gaming results and thinking "that's disappointing", there could be an unresolved story here.
Of course one possible conclusion is simply that Ryzen isn't that amazing in games, but we're just not fully sure that is truly the case if all hardware was working correctly

EDIT: Don't get your hopes up about a potential fix. The results we achieved could be it, and you should make any buying decisions accordingly at this stage. The best thing may be to wait a few weeks just to make sure

 

[leadeater]

5 hours ago, Hunter259 said:

Games will never use 16 threads. Period. Vulkan and DX12 are there to move more number crunching onto the GPU.

There are games that will, Real Time Strategy and Turn Based Strategy some of which are already very highly thread capable.

 

Vulkan and DX12 isn't for pushing compute to GPUs, that is GPGPU/OpenCL/CUDA/Direct Compute. Vulkan and DX12 APIs are purposed around giving lower hardware level access to GPUs and removing the heavy API penalty on draw calls and scheduling, this allows the CPU to send more data to GPUs along with using less CPU resources to do it.

 

Moving computation tasks from the CPU to GPU in games is actually rather rare, if you do this that is taking away rendering resources. Another reason this hasn't been common place is the lack of Async Compute in APIs and hardware along with the above mentioned issues with older APIs. With Vulkan and DX12 and true hardware support for Async Compute that does allow for the possibility to actually start doing it but you are still taking up GPU rendering resources to do it.

 

AMD's idea quite some time ago was to try and do this with HSA and the GPU in their APUs, that hasn't panned out so far yet for the very same API reasons and very small target audience issue.