AMD speaks on W10 scheduler and Ryzen

[Sakkura]

2 minutes ago, zMeul said:

except it's ~14% 

Ryzen on DDR4 does less than Haswell IPC .. or you did forgot that

Holy shit 14%, hold the press.

[DarkBlade2117]

30 minutes ago, AnonymousGuy said:

Can I just bottle up the tears of AMD fanboys?  They taste delicious and I want to be able to enjoy them year round.

looking at your signature you have more money than you could care for. I mean.. you seem like the person who cares purely about raw performance of saving money so good on ya

[DocSwag]

So basically if you're a pure gamer don't buy it. This still doesn't change my stance on the matter.

 

If you're only gaming, get the 7700k. If you're doing a mix of content creation and gaming, get the 1700. If you like giving AMD free money, get the 1800x or 1700x, and if you're filthy rich get the 6900k.

[Guest Darth Revan]

15 minutes ago, Sakkura said:

And Intel has only improved on Haswell a little. So AMD is only a little behind in gaming performance, because that's what the benchmarks show.

9 minutes ago, MageTank said:

Exactly, and within 4 years, Intel managed to improve IPC by a whopping 5-8% when comparing Haswell to Kaby.

That's because Intel was(and still) so far ahead that they didn't feel the need to push further beyond.

 

[Space Reptile]

i honestly find the point below that also worth mentioning

 

Quote

 

Temperature Reporting

The primary temperature reporting sensor of the AMD Ryzen™ processor is a sensor called “T Control,” or tCTL for short. The tCTL sensor is derived from the junction (Tj) temperature—the interface point between the die and heatspreader—but it may be offset on certain CPU models so that all models on the AM4 Platform have the same maximum tCTL value. This approach ensures that all AMD Ryzen™ processors have a consistent fan policy.

 

Specifically, the AMD Ryzen™ 7 1700X and 1800X carry a +20°C offset between the tCTL° (reported) temperature and the actual Tj° temperature. In the short term, users of the AMD Ryzen™ 1700X and 1800X can simply subtract 20°C to determine the true junction temperature of their processor. No arithmetic is required for the Ryzen 7 1700. Long term, we expect temperature monitoring software to better understand our tCTL offsets to report the junction temperature automatically.

 

The table below serves as an example of how the tCTL sensor can be interpreted in a hypothetical scenario where a Ryzen processor is operating at 38°C.

 

 

Product Name	True Junction Temp (Example)	tCTL Offset for Fan Policy	Temp Reported by tCTL
AMD Ryzen™ 7 1800X	38°C	20°C	58°C
AMD Ryzen™ 7 1700X	38°C	20°C	58°C
AMD Ryzen™ 7 1700	38°C	0°C	38°C

 

 

[zMeul]

2 minutes ago, DocSwag said:

So basically if you're a pure gamer don't buy it. This still doesn't change my stance on the matter.

 

If you're only gaming, get the 7700k. If you're doing a mix of content creation and gaming, get the 1700. If you like giving AMD free money, get the 1800x or 1700x, and if you're filthy rich get the 6900k.

not that simple

Puget Systems has done some real world test workloads (not games) and concluded, in two cases, the 7700K is the better buy

 

• https://www.pugetsystems.com/labs/articles/SOLIDWORKS-2017-AMD-Ryzen-7-1700X-1800X-Performance-908/
• https://www.pugetsystems.com/labs/articles/Premiere-Pro-CC-2017-AMD-Ryzen-7-1700X-1800X-Performance-909/

[Frank6247]

Pcper did a video regarding this topic, which shes some light into the situation:

 

 

[DocSwag]

6 minutes ago, zMeul said:

not that simple

Puget Systems has done some real world test workloads (not games) and concluded, in two cases, the 7700K is the better buy

 

• https://www.pugetsystems.com/labs/articles/SOLIDWORKS-2017-AMD-Ryzen-7-1700X-1800X-Performance-908/
• https://www.pugetsystems.com/labs/articles/Premiere-Pro-CC-2017-AMD-Ryzen-7-1700X-1800X-Performance-909/

Hmmm interesting. Seems like these workloads are more single core heavy than many rendering workloads. That's still pretty specific though, so generally for content creators I'd still recommend Ryzen.

[MageTank]

24 minutes ago, zMeul said:

except it's ~14% 

Ryzen on DDR4 does less than Haswell IPC .. or you did forgot that

Where is that 14% number? We see Broadwell giving a 3% improvement, and Skylake/Kaby giving another 2-3% on top of that.

[Valentyn]

13 minutes ago, zMeul said:

not that simple

Puget Systems has done some real world test workloads (not games) and concluded, in two cases, the 7700K is the better buy

 

• https://www.pugetsystems.com/labs/articles/SOLIDWORKS-2017-AMD-Ryzen-7-1700X-1800X-Performance-908/
• https://www.pugetsystems.com/labs/articles/Premiere-Pro-CC-2017-AMD-Ryzen-7-1700X-1800X-Performance-909/

Good ole Adobe, still can't optimise their software to save their lives.

You know they're shit when Apple can get FCPX on a MacBook Pro to beat a vastly more powerful custom Windows Workstation that's running Premiere Pro.

One day they'll get their finger out, and the world shall rejoice! 

[Drak3]

14 minutes ago, Darth Revan said:

That's because Intel was(and still) so far ahead that they didn't feel the need to push further beyond.

 

No, it's because Intel can't do the impossible. The best we're going to see are minor clockspeed boosts, from refining of their manufacturing, and maybe a 1% IPC improvement in cherry picked scenarios, so long as we're still on silicon.

 

2 minutes ago, MageTank said:

Where is that 14% number? We see Broadwell giving a 3% improvement, and Skylake/Kaby giving another 2-3% on top of that.

His bullshit 14% is his claim on how far behind Ryzen is, when in reality, under operating systems that are actually using Ryzen correctly, it falls in between Haswell and Broadwell for all but specific revisions of AVX.

 

And the 3% number for Broadwell is quite generous. It's more in the neighborhod of 1% for an across the board figure.

[WkdPaul]

Just now, Drak3 said:

No, it's because Intel can't do the impossible. The best we're going to see are minor clockspeed boosts, from refining of their manufacturing, and maybe a 1% IPC improvement in cherry picked scenarios, so long as we're still on silicon.

 

His bullshit 14% is his claim on how far behind Ryzen is, when in reality, under operating systems that are actually using Ryzen correctly, it falls in between Haswell and Broadwell for all but specific revisions of AVX.

 

And the 3% number for Broadwell is quite generous. It's more in the neighborhod of 1% for an across the board figure.

Actually, Haswell to Broadwell IPC improvments are just above 3%, not 1% ... and less than 3% from Broadwell to Skylake, so around 6% from Haswell to Skylake.

 

Source;

http://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/9

 

[79wjd]

30 minutes ago, gabrielcarvfer said:

Still think that MS can mitigate the cache latency between CCXs by adjusting their scheduler.

 

  Hide contents

 

 

https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/page-19

How come ryzens latency within CCX'S is so much lower? 

20 minutes ago, MageTank said:

Where is that 14% number? We see Broadwell giving a 3% improvement, and Skylake/Kaby giving another 2-3% on top of that.

Check out stilts benchmarks from zmeuls IPC thread. It seems to indicate a roughly 14% IPC advantage for kaby lake. Although, there is also a lot of variance in test runs (in general), so it kind of makes me question the way we generally test IPC -- you really need to average out a lot of runs to minimize variances in testing, and I'm not sure how many tests reviewers end of averaging.

[MageTank]

9 minutes ago, wkdpaul said:

Actually, Haswell to Broadwell IPC improvments are just above 3%, not 1% ... and less than 3% from Broadwell to Skylake, so around 6% from Haswell to Skylake.

 

Source;

http://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/9

 

To further add to the Broadwell/Haswell info pool: https://us.hardware.info/reviews/6215/2/intel-five-generation-ipc-test-broadwell-haswell-ivy-bridge-sandy-bridge-and-nehalem-results

 

Results seem very similar to what Anandtech saw. As for the Skylake vs Haswell argument, it's hard to find people that test DDR3 on Skylake to remove the variables.

1 minute ago, djdwosk97 said:

How come ryzens latency within CCX'S is so much lower? 

Check out stilts benchmarks from zmeuls IPC thread. It seems to indicate a roughly 14% IPC advantage for kaby lake. Although, there is also a lot of variance in test runs (in general), so it kind of makes me question the way we generally test IPC -- you really need to average out a lot of runs to minimize variances in testing, and I'm not sure how many tests reviewers end of averaging.

I've seen that information. I am not entirely sold on it's validity. 

[79wjd]

1 minute ago, MageTank said:

To further add to the Broadwell/Haswell info pool: https://us.hardware.info/reviews/6215/2/intel-five-generation-ipc-test-broadwell-haswell-ivy-bridge-sandy-bridge-and-nehalem-results

 

Results seem very similar to what Anandtech saw. As for the Skylake vs Haswell argument, it's hard to find people that test DDR3 on Skylake to remove the variables.

I've seen that information. I am not entirely sold on it's validity. 

It still points to a lot of inconsistencies between testing runs. Id just really like to see someone do a lot more in depth IPC testing taking into account testing inaccuracies. One thing I noticed, when I was digging into Stilt's results, is a 10%~ variance in CB runs, that's a pretty large variance when trying to compare two very close CPUs.

[Hunter259]

52 minutes ago, gabrielcarvfer said:

Still think that MS can mitigate the cache latency between CCXs by adjusting their scheduler.

 

  Reveal hidden contents

 

 

https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/page-19

Not likely. That isn't a scheduling issue. Ryzen chips are built in a similar form to how the Core 2 Quads were made. They take a Quad core and connect it to another on the same package. The latency is the fact it is having to move data outside of one chip to another and then out. That is simply an architecture problem. Nothing microsoft can do.

[zMeul]

1 hour ago, DocSwag said:

so generally for content creators I'd still recommend Ryzen.

you mean workloads that can be accelerated by GPUs?! it makes no sense to buy Zen

[zMeul]

1 hour ago, gabrielcarvfer said:

Still think that MS can mitigate the cache latency between CCXs by adjusting their scheduler.

very very bad idea

if they adjust the CCX scheduling that will create issues further down the line for the 6 core and 4 core Zen CPUs, R5 and R3 ... if memory serves

[zMeul]

1 hour ago, wkdpaul said:

Actually, Haswell to Broadwell IPC improvments are just above 3%, not 1% ... and less than 3% from Broadwell to Skylake, so around 6% from Haswell to Skylake.

 

Source;

http://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/9

those tests have proven to be quite wrong

we had a lengthy debate over the Zen IPC thread and found out the results from Anand's testing do not match with other tests and my very own - the very specific Cinebench R15

 

one other thing: if some of the tests were done with W7 and others with W10, it invalidates the results

[Doobeedoo]

W10 scheduler aside, specially now. I'm more interested to see the effect of 'optimizations for Zen architecture' for software that doesn't leverage it to it's potential.

We've seen numerous games where it performs very good and some where it performs very odd. It all comes to how many cores each game use and API but still.

Though shame it can't have frequencies like latest FX CPUs though. That'd be great.

[Tomsen]

53 minutes ago, djdwosk97 said:

How come ryzens latency within CCX'S is so much lower? 

Its quite simple. Because each CCX has its own L3 cache.

So the cores in CCX[0] can't see the cache in CCX[1]. [0] being the first and [1] being the second CCX cluster.

 

34 minutes ago, Hunter259 said:

Not likely. That isn't a scheduling issue. Ryzen chips are built in a similar form to how the Core 2 Quads were made. They take a Quad core and connect it to another on the same package. The latency is the fact it is having to move data outside of one chip to another and then out. That is simply an architecture problem. Nothing microsoft can do.

Except for the fact that core 2 quads were using MCM whereas ryzen is a single die.

It is not impossible for microsoft to optimize for zen architecture. For example, keep all the processes spawned from one program on a single CCX.

[Techstorm970]

6 minutes ago, zMeul said:

you mean workloads that can be accelerated by GPUs?! it makes no sense to buy Zen

With that logic, you could say that it also makes no sense to buy Intel Extreme processors.

[Tomsen]

4 minutes ago, zMeul said:

very very bad idea

if they adjust the CCX scheduling that will create issues further down the line for the 6 core and 4 core Zen CPUs, R5 and R3 ... if memory serves

You are dead wrong. The same scheduling optimization will work on the 6 core, and the 4 core wont even have the issue to begin with. Your memory is bugged.

[mattebad]

5 hours ago, zMeul said:

not that simple

Puget Systems has done some real world test workloads (not games) and concluded, in two cases, the 7700K is the better buy

 

• https://www.pugetsystems.com/labs/articles/SOLIDWORKS-2017-AMD-Ryzen-7-1700X-1800X-Performance-908/
• https://www.pugetsystems.com/labs/articles/Premiere-Pro-CC-2017-AMD-Ryzen-7-1700X-1800X-Performance-909/

I enjoy how ridiculous those Premiere benchmarks are considering they didn't give them all a solid clock to hold for the entire test nor did they mention what they ran at during the entirety of the benchmark other than, yeah, here's the turbo range, figure it out, but here's a solid clock for Ryzen. I'll take the other 20 reputable benchmarks from places like guru3d, gamernexus, anandtech, and pcper over pugetsystems, a company who I didn't even know still existed and is not a professional reviewer, clearly, from how they conduct their benchmarks.

 

The 1800x and 1700x consistently beats the 6900k in single thread benchmarks in Cinebench R15 as well as multi-thread, however that is just one benchmark. On average in rendering tests, the 1800x performs as follows:

 

-3% Corona Photorealism 1.3
-10% Blender 2.78
-5% Luxmark C++
-27% Luxmark OpenCl
+11% POV Ray 3.7
+4% Cinebench R15 Singlethread
+10% Cinebench R15 Multithread

 

All percentages were done comparing the 1800x at stock to the 6900k at stock using Anandtechs rendering benchmark scores.

 

So if an average of 10% loss in performance in some rendering tests is of concern to you and you need to spend another $520 for 10% more performance for your render, then well, you be you. If you are not insane however like @zMeul, and are an actual consumer looking for an amazing processor for the price for rendering, then this processor is for you!  

 

In addition, if you are playing at 1440p or above in a majority of games, you should expect parity with other processors including the 6900k. However, if you are gaming at 1080p and below, well I would say that you should hold off for a bit until the Ryzen platform matures a bit, if you seriously need a CPU like immediately because you can't wait any longer to make that sweet build, then I would recommend you go for the 7700k which overclocks like a beast or alternatively, if you wanted a good Quad core with no hyperthreading, I personally recommend the 7600k over the 6600k which is what I am running because Kaby Lake seems to overclock much better than the latter. Now if you are buying second hand, you could go for a 6600k/6700k or you can take it a step higher if you can find a good 5820k those overclock pretty easily and the extra 2 cores are great for more cpu intensive games like GTA V or The Division.

 

Edit: Benchmark links that were used

 

Anandtech: http://www.anandtech.com/show/11170/the-amd-zen-and-ryzen-7-review-a-deep-dive-on-1800x-1700x-and-1700/18

 

Guru3d (Page 16 on): http://www.guru3d.com/articles_pages/amd_ryzen_7_1800x_processor_review,16.html

[zMeul]

2 minutes ago, techstorm970 said:

With that logic, you could say that it also makes no sense to buy Intel Extreme processors.

who the fuck even buys Intel Extreme?! you are that rich .. go ahead

 

if you have workloads that prefers i7-6950X, then it should be no problem paying for it - normal people don't have those types of workloads