A Look at the AMD Zen 2 Core • Hacker News Board

Re: A Look at the AMD Zen 2 Core

#31

Post by garkin » Sat, Jul 06, 2019, 5:57 PM UTC

jchw wrote: post #21

garkin wrote: post #18

Zen 2 is very good in number crunching and synthetics. But it has a problem - terrible memory latency. 70ns with 3600cl16. ( https://www.userbenchmark.com/UserRun/18168279 ) It distills to a not-so-good gaming frame times. It's 64mb L3 cache ( https://en.wikichip.org/wiki/amd/ryzen_9/3900x ) helps only partially. Few games will suffer greatly from it, but there are several titles with RAM bottlenecks, like PUBG and F…

Not sure why your original comment disappeared. I was kind of curious what the latencies might be for other contemporary processors/builds, and I'm not sure 70ns is actually really outside the normal margins. Here's an i7-8700k build that is already pretty close to 70ns: https://www.userbenchmark.com/UserRun/18173216 - Also acknowledging that, this is not the 'best case' performance. But seems to be not so unusual ei…

I will distill your post to: most user builds are bad balanced to begin with and they wont see a difference and would had a better price / more cores. Valid point, i agree with it. Still could be argued about a need for a better memory for Ryzen. This equalizes total build cost and you need to be informed about this platform trait beforehand, which will results in even worse average build balance. Imagine prebuilt PCs from wallmart with 2400 leftovers.

But my point is more about advanced user builds and press benchmarks.

[0] - 2400 cl16. It's much worse than 3600 cl16, used in referenced Ryzen 3000 test. It would be interesting to see how Ryzen would perform with it. May be zen2 comparison with 2400 vs 3400 (3600 is very rarely achievable there as IMC trait of the platform) would be a good illustration.

[1] - 3200 C16. But it performs way bellow expectations. Modern Intel chip should have ~55ns with it. With 3600 cl16 intel will have ~40ns.

Re: A Look at the AMD Zen 2 Core

#32

Post by jefft255 » Sat, Jul 06, 2019, 5:58 PM UTC

Der_Einzige wrote: post #17

bogomipz wrote: post #5

The post states: >"Perceptrons are the simplest form of machine learning and lend themselves to somewhat easier hardware implementations compared to some of the other machine learning algorithms." Can someone explain what is it about perceptrons that make them easier to implement in hardware?

Wait, how is K-NN harder to implement than a perception... Even in hardware? What about linear regression? What about decision trees? What about naive Bayes?

Performing nearest neighbors search is much more expensive than a simple multiply-add. I don’t know about the other, my idea is that a perceptron is much easier to implement in hardware/low-level operations.

Re: A Look at the AMD Zen 2 Core

#33

Post by vbezhenar » Sat, Jul 06, 2019, 5:59 PM UTC

jchw wrote: post #23

demilicious wrote: post #20

Earlier quoted context omitted.

As far as I know, these processors are not yet released. What's the confidence level that this user benchmark will be indicative of real life expected performance? It seems implausible that this user benchmark is a good indicator. The Zen 1 architecture exhibited nothing of the sort[0] -- it would be an order of magnitude performance regression. I expect we'll start to see more accurate tests once the processors are…

Well, to be devil's advocate: just because AMD improved their memory latency does not mean it was state-of-the-art. Indeed, looking at some random user benchmarks of Zen 1, it looks like the memory latency is similar with Zen 1. (See my reply to parent for a couple links, but I think you could also just find random Ryzen 2700X benches and see the same.) Of course, the effects of ~20ns more latency on system memory ac…

But Zen 2 has HUGE L3 cache. I think that this cache would compensate for that latency.

Re: A Look at the AMD Zen 2 Core

#34

Post by kkmx » Sat, Jul 06, 2019, 6:00 PM UTC

jefft255 wrote: post #32

Der_Einzige wrote: post #17

Earlier quoted context omitted.

Wait, how is K-NN harder to implement than a perception... Even in hardware? What about linear regression? What about decision trees? What about naive Bayes?

Performing nearest neighbors search is much more expensive than a simple multiply-add. I don’t know about the other, my idea is that a perceptron is much easier to implement in hardware/low-level operations.

Same with Naive Bayes. That's very expensive compute-wise.

Re: A Look at the AMD Zen 2 Core

#35

Post by overcast » Sat, Jul 06, 2019, 6:09 PM UTC

garkin wrote: post #18

Zen 2 is very good in number crunching and synthetics. But it has a problem - terrible memory latency. 70ns with 3600cl16. ( https://www.userbenchmark.com/UserRun/18168279 ) It distills to a not-so-good gaming frame times. It's 64mb L3 cache ( https://en.wikichip.org/wiki/amd/ryzen_9/3900x ) helps only partially. Few games will suffer greatly from it, but there are several titles with RAM bottlenecks, like PUBG and F…

I think people get a little over zealous with CPU requirements on games. If my eight year old Xeon 5676 can handle every past and present game I throw at it, I think you'll be beyond spec for years to come with a Zen 2(or any modern processor).

Re: A Look at the AMD Zen 2 Core

#36

Post by eloff » Sat, Jul 06, 2019, 6:09 PM UTC

vbezhenar wrote: post #33

jchw wrote: post #23

Earlier quoted context omitted.

Well, to be devil's advocate: just because AMD improved their memory latency does not mean it was state-of-the-art. Indeed, looking at some random user benchmarks of Zen 1, it looks like the memory latency is similar with Zen 1. (See my reply to parent for a couple links, but I think you could also just find random Ryzen 2700X benches and see the same.) Of course, the effects of ~20ns more latency on system memory ac…

But Zen 2 has HUGE L3 cache. I think that this cache would compensate for that latency.

Only partially. There are diminishing returns on cache size.

Re: A Look at the AMD Zen 2 Core

#37

Post by jchw » Sat, Jul 06, 2019, 6:13 PM UTC

garkin wrote: post #31

jchw wrote: post #21

Earlier quoted context omitted.

Not sure why your original comment disappeared. I was kind of curious what the latencies might be for other contemporary processors/builds, and I'm not sure 70ns is actually really outside the normal margins. Here's an i7-8700k build that is already pretty close to 70ns: https://www.userbenchmark.com/UserRun/18173216 - Also acknowledging that, this is not the 'best case' performance. But seems to be not so unusual ei…

I will distill your post to: most user builds are bad balanced to begin with and they wont see a difference and would had a better price / more cores. Valid point, i agree with it. Still could be argued about a need for a better memory for Ryzen. This equalizes total build cost and you need to be informed about this platform trait beforehand, which will results in even worse average build balance. Imagine prebuilt PC…

You seem pretty certain this will manifest as a noticeable performance hit. Can this kind of thing be easily measured in practice? I don't know if my workloads would be memory latency sensitive and worse, I don't even have a clue how I would find out.

I'm not too concerned though, since I already use Zen 1 and it looks to be around the same. If I had to take a shot in the dark, I'm guessing it's just a consequence of the chiplet designs of Zen and Zen 2 that enabled them to scale so well. If it is such a tradeoff and there is not future gains to be made on latency for Zen platforms, I can accept that.

Though it does make me curious if Intel will ever stumble upon the same problems.

Re: A Look at the AMD Zen 2 Core

#38

Post by transpute » Sat, Jul 06, 2019, 6:19 PM UTC

mtgx wrote: post #30

I wish AMD enabled Secure Encrypted Memory by default for all devices, and also bring Secure Encrypted Virtualization (or something similar) to consumer Ryzen chips, too. Even Microsoft has made VM-sandboxing a consumer feature. It's time for AMD to join the bandwagon and help along the way, not keep the virtualization/other security features exclusive to server customers.

SEV is claimed to be present on Ryzen Embedded, https://en.wikichip.org/wiki/amd/ryzen_embedded/v1605b#Featu...

Available on UDOO Bolt, https://www.kickstarter.com/projects/udoo/udoo-bolt-raising-...

At least AMD made DRTM (SKINIT) available on consumer CPUs, where Intel still keeps DRTM (TXT) limited to business vPro chipsets. ECC is available on some CPU/motherboard/bios combinations for Ryzen, although ECC and low-power for Ryzen APUs is segmented to "Pro" models only available to OEMs.

It's still near impossible to buy devices with Ryzen Embedded, despite several system announcements from ASRock, but SEV is available today on Supermicro Epyc 3000 mITX motherboards.

Re: A Look at the AMD Zen 2 Core

#39

Post by ivalm » Sat, Jul 06, 2019, 6:29 PM UTC

Der_Einzige wrote: post #17

bogomipz wrote: post #5

The post states: >"Perceptrons are the simplest form of machine learning and lend themselves to somewhat easier hardware implementations compared to some of the other machine learning algorithms." Can someone explain what is it about perceptrons that make them easier to implement in hardware?

Wait, how is K-NN harder to implement than a perception... Even in hardware? What about linear regression? What about decision trees? What about naive Bayes?

A single layer perceptron IS linear regression + some non-linearity of the output. I think in reality they could have used the word "logistic regression" but chose perceptron because it sounds better.

Re: A Look at the AMD Zen 2 Core

#40

Post by BubRoss » Sat, Jul 06, 2019, 6:29 PM UTC

garkin wrote: post #18

Zen 2 is very good in number crunching and synthetics. But it has a problem - terrible memory latency. 70ns with 3600cl16. ( https://www.userbenchmark.com/UserRun/18168279 ) It distills to a not-so-good gaming frame times. It's 64mb L3 cache ( https://en.wikichip.org/wiki/amd/ryzen_9/3900x ) helps only partially. Few games will suffer greatly from it, but there are several titles with RAM bottlenecks, like PUBG and F…

That benchmark only has a few samples on an unreleased cpu. Typical ballpark numbers for memory latency are usually 100ns. There are also a lot of factors that can influence random memory access such as whether the lookup is going through virtual memory that isn't a page in the TLB cache. I'm skeptical this benchmark tells us much.

Also games that are sensitive to memory latency are very poorly written games. This is not a new or contraversial opinion of PubG, which is know for being terribly written (and that is supposedly a reason for fortnite taking its audience so easily). It is also written using Unity and C#, which don't require hopping around in memory, but do make that the most obvious path, which is a trap for people who don't know better.

Farcry is not a game I would expect to have these problems. Ubisoft seems generally technically sound, despite releasing wildly unstable games, but it will take real numbers to get a clearer picture.