New Ryzens and DAWBench.

[Kaine]

Load of benchmarking this week, results here:

http://www.scanproaudio.info/2018/05/02 ... testbench/

[Keith99]

Very interesting! I am due a new machine soon and I had been looking at the 2700X as opposed to the 8700K as other benchmarks tend to show the 2700 just behind on IPC but better on multi-threaded. So my reasoning was the 2700 would be more future proof as software makes better use of more cores however I had not seen any comparisons based on music making (my main PC use) and the results here seem to suggest the 8700 is still way ahead of 2700X. I was not sure if the tests were carried out with the new 470 chipset? I also wonder if given time a solution for the multi-core boost speed problems will be applied.
Food for thought - thanks

[Kaine]

The board was an ASUS X470 series model.

[Dip200]

Thank you for the benchmarks. Disapointing but also slightly weird results.

http://www.scanproaudio.info/wp-content ... 8-Q2-1.jpg

At 256 buffer the 4.1 Ghz 2700X can only open 3 more SGAS than the 4.0 Ghz 1800X ? That contradicts everything I have read about this chip.

Also interesting: The overclocked 4.7Ghz 8700K can open 177 SGA instances.
However, in this benchmark http://www.scanproaudio.info/wp-content ... ssic-1.jpg it can only open 317 RXC compressors vs. 355 by a 1700X at 3.8 Ghz.

Do you have any idea how that is possible ?

[bmrzycki]

Do you have any idea how that is possible ?

AMD has focused (in general) on more cores/threads with slightly slower instructions per cycle. Intel is the exact opposite. For highly serialized workloads IPC wins.

AMD also focused on memory throughput problems from gen1 to gen2. Unfortunately that doesn't really help us, and these benchmarks verify that.

The digital audio segment of computing tends to be bound by serialized problems which is why DAWBench and VI show almost 20% between the AMD 2700x vs Intel's 8700k.

Thank you Kaine for running these benchmarks. They help those of us who build our PCs and are looking to upgrade to make informed decisions.

[starflakeprj]

So the i9-7940X is the best choice, when only taking performance and not price/performance into the calculations. Let's hope the price will go down ~200 USD until the summer.

[Dip200]

Do you have any idea how that is possible ?

AMD has focused (in general) on more cores/threads with slightly slower instructions per cycle. Intel is the exact opposite. For highly serialized workloads IPC wins.

AMD also focused on memory throughput problems from gen1 to gen2. Unfortunately that doesn't really help us, and these benchmarks verify that.

[...]

The 1800X 11% faster than the 8700K at the RCX benchmark at 256 buffer. So the performance seems to depend on which plugins are used.

Would love to see a RCX benchmark with the 2700X. If it behaves like expected and all benchmarks and reviews say it should be able to open as many instances of RCX as the 7820X (roughly 10% more).
1800X = 355 instance + 10% = 390 instances (7820X 392 instances) (8700K only 317).

http://www.scanproaudio.info/wp-content ... ssic-1.jpg

[Dip200]

So the i9-7940X is the best choice, when only taking performance and not price/performance into the calculations. Let's hope the price will go down ~200 USD until the summer.

In this scenario the 7980X is the best choice. But I would wait for the next generation which should not be to far away. That generation is supposed to have Spectre fixed on the hardware level, maybe soldered heatspreaders again, better IPC and maybe new prices incase the 9700K will be an 8 core which rumors say it will. And then the 10 core maybe costs under 600 and so forth.

[Scotty]

Huge thanks to Scanaudio for doing these tests and making them public. We all benefit. I am holding on till November or December to build a rig to replace my 3930K.

[Kaine]

Also interesting: The overclocked 4.7Ghz 8700K can open 177 SGA instances.
However, in this benchmark http://www.scanproaudio.info/wp-content ... ssic-1.jpg it can only open 317 RXC compressors vs. 355 by a 1700X at 3.8 Ghz.

Do you have any idea how that is possible ?

As I say, the two sets of benchmarks are pretty much uncomparable, although a difference in how it handles the load balancing between plugin's could also show up some result differences.

Differences in test bench install:

3 major service packs
Meltdown and Spectre patches
BIOS builds
Cubase Revision change
Reaper Revision change
Whole Benchmark change.

It's why I and the other firms who test, wil tend to freeze the setup at first install to try and ensure a fairly level playing field.

I'll see if I still have my old install anywhere to do an RXC test.

[Kaine]

The 1800X 11% faster than the 8700K at the RCX benchmark at 256 buffer. So the performance seems to depend on which plugins are used.

Now I think about it, I don't recall ever doing that chip test on the RXC benchmark; are you using Tech Reports figures?

Would love to see a RCX benchmark with the 2700X. If it behaves like expected and all benchmarks and reviews say it should be able to open as many instances of RCX as the 7820X (roughly 10% more).
1800X = 355 instance + 10% = 390 instances (7820X 392 instances) (8700K only 317).

http://www.scanproaudio.info/wp-content ... ssic-1.jpg

2700X has small gains on the 1700X, sits in the middle of the 8700K & 7820X still.

64 - 308
128 - 342
256 - 359

[Dip200]

I meant 1700X, not 1800X, sorry.

Thanks for doing the RCX test but these results are strange. A 2700X at 4.1 Ghz (I assume) can only open 4 more instances at 256 than a 1700X at 3.8 Ghz ? Shouldn't even a 1700X at 4.1 Ghz (if it was possible) be able to open more than that ? The clock is almost 10% higher.

[Kaine]

Not really, as the 1700X results are pre spectre/meltdown fiasco, and the 2700X is afterwards.

The 2700X outperforms the 1800X in the new test.
The 1800X outperformed the 1700X in the old test.

A lot has changed here as I say, none of the old results are comparable against the newer ones. If I run the 2700X on the old bench then I need to flash the BIOS to do so, similarly to run the 1700X on the new X470 bench the BIOS updates are already baked in.

But, as I've still got the bench up and I could locate a chip...

1700X @ 3.8GHz all cores - SGA 1566 DB Test

64 = 124
128 = 131
256 = 140

Which looks roughly where it should be given the 1800X results.

So, for comparison, I rebenched the 1700X using the RXC test on the same platform.

1700X @ 3.8GHz all cores - DB classic RXC Test

64 = 239
128 = 277
256 = 302

Which are all far lower than the original benchmarking although it's also worth also noting at this point that the current benchtest is running at 24/48 for the SGA test (and this classic test this time), where as it was 24/44 with the older benchmarks which should explain a chunk of the above. The rest of it will be down to changes to the platform in the interim.

[Scotty]

It seems a shame with all of these cores being available, and with a good clock speed that the maximum logical core issue doesn't seem to get much attention. This combined with the the maximum unique .dll issue makes building a monster PC or even upgrading harder to justify as much of the power will be inaccessible to daws.

Kaine do you have any more information on these topics?

[acYm]

I'm not so sure that a x299 successor will show up anytime soon. I've been looking around, and NOBODY likes this platform except us who are just drooling, it's far from critical success for Intel. if I were them I'd consider killing the HEDT line and just have i7's and xeons.

I'm still planning on going for a 7940x this year - gotta consider that the next platform, whatever it is, won't support W7 for sure, and I'm not letting go of it until January 14 2020.