Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


Zen 3 / Ryzen 5000, my view
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Zen 3 / Ryzen 5000, my view

jsgjsg Member, Resident Benchmarker

I just read Anandtechs long article about the new Ryzen 5000. Here is my impression and thoughts.

TL;DR front-up: if you are a gamer, hurry to get the new super-processor with 19% higher IPC!

Now, on to addressing the reasonable people with a working brain ...

First, you pay double for those "19% higher IPC", which btw. highly likely will turn out to be more like 15% or even a bit less in real life use. You pay the "AMD premium" of about $50 that AMD introduced with the Ryzen 5000 series, or in other words, now that they feel they are not the underdog anymore. Plus you pay in terms of electrical power consumption because obviously those 5+ GHz top frequency gobbles up power and not modestly.

For gamers that's of no concern, of course, but for hosters, and by extension for their customers, but also for companies with dozens or even hundreds of PCs that is of concern.

What do you really get with the new Ryzen 5000? Let me put it this way: I personally will stick to the Ryzen 3000 and 4000 series. They are cheaper, there are plenty of battle proven mainboards available and they are IMO at the real sweet spot. Let me explain: Zen 1 to Zen 2 really was a significant step. Not only do Zen 2 processors blow their Zen 1 siblings out of the water in terms of performance but they also incorporate what AMD learned from the mistakes and the experience with Zen 1. In other words, the Zen 2 are the grown up and quite enhanced Zens. The Zen3 on the other hand (a) has been designed by another team, and (b) seems to be centered around and focused largely on one point: to conquer the one area where intel still is/was the king of the hill, single core performance - which also translates to gaming.

Gaming, that must be seen, is a very attractive market segment for both AMD and intel because the clientele is very easy to manipulate and also (comparatively) easily pays premium prices. So I can certainly understand AMD's efforts to conquer that market segment.

On the other hand though the hosting market as well as a vast share of the desktop market is driven by other factors. To name two: power consumption (If you have 100 systems saving 20W per system translates to real money) and the availability of a large choice along with well established products.

The big one though is the simple fact that almost nobody needs the levels of performance a Ryzen 5000 offers, especially not when looking at the cost (both of purchase and of operation) vs benefits/performance ratio.
Fact is that most VPS customers buy - and pay for - way more performance, traffic volume, etc than they need. Fact is also that for 95% of PCs in offices even an 8 core Zen2 processor is way more than what's needed.

At this point the interests and needs of offices/hosting customers fork though. As a hosting customer or someone in an office you have a certain need, and that's it. "Good enough" usually does the job. As a hoster though, many feel pulled to join the number race; if e.g. my benchmark shows a result of say 450 (for processor and memory) you don't want to be one of those whose systems achieve "only" 410.

But frankly, I think you shouldn't be fooled into thinking that Ryzen 39xx are 'the cornerstone', the ruler along which everyone must be measured. The reason is simple: Even e.g. Xeon E5-26xx v3 with "only" results around 250 or 300 are way more powerful than what the vast majority of your customers really need - especially when considering that your customers do see $$ too and usually tick in terms of "power for x amount of $".

Practically speaking and with my feet on the ground, I could choose between an E5 2600 v3 and a 2450L ... and I chose the 2450L dedi because it's more than damn good enough to run my largest and most demanding applications and web server plus some VPSs. For reference: it's number in my benchmark is about 160. 160! Plus, the 2450L is very poor in terms of multi-core performance; the E5-26xx v3 and even the "stone age" v2 is much much better - but still, it does it's job plus it still has plenty of reserves.

But I also have some Ryzen 39xx based VPS from NexusBytes@seriesn and, yes, they are much faster than the 2450L ... but at the end of the day only one factor counts: how fast do my visitors see a fully loaded web page and how much do I have to pay for that performance?

The answer is simple: More cores (even slow ones) are better than fewer cores, faster disk IO is better than slower IO, and especially more RAM is much much better than less memory. If you need to make pretty much any server app faster, usually adding more memory will provide far more bang than a faster processor.

But those are factors in pricing VPSs too! RAM doesn't come free nor do faster disks. A quick glance over any providers pricing table will confirm that.

So, my advice to providers is two-fold:

  • explore and use other and cheaper ways to increase bang for the buck than purchasing new Ryzen 5000 systems
  • The sweet spot, at least for some years, is Zen 2 (Ryzen 3000).

I personally am not at all enchanted by the Ryzen 5000, especially not for hosting. And frankly, a closer look at the data shows that - as everywhere - there are tradeoffs and frankly IMO the Ryzen 3000 is the more attractive choice for hosting. In case you feel threatened by the "better" Ryzen 5000, just put in some more RAM and your Ryzen 3000 based VPS will easily kill a Ryzen 5000 based VPS with less RAM. And keep in mind what customers want at the end of the day: "how fast does my server serve content? And how much do I have to pay for that?".

TL;DR2: do not be fooled by the Ryzen 5000. It is largely a gamers processor.

Thanked by 3bdl RedSox lovelyserver
«1

Comments

  • thanks dad very cool

  • I didn't read anything of it, but I'm very sure that LET is not blogger. So please bugger off.

    Thanked by 1kalimov622
  • 15% of a 100 hour single threaded job is still 15 hours, I'll take it

  • jsgjsg Member, Resident Benchmarker

    @NewToTheGame said:
    15% of a 100 hour single threaded job is still 15 hours, I'll take it

    Sure - but then, how many percent of hosting customers run single threaded 100 hour jobs?

  • I mean if you watched the presentation it was literally marketed as a gamer cpu. They weren't really hiding that. Side note anyone wanna buy a 3900x from me? xD

  • NewToTheGameNewToTheGame Member
    edited November 2020

    @jsg said:

    @NewToTheGame said:
    15% of a 100 hour single threaded job is still 15 hours, I'll take it

    Sure - but then, how many percent of hosting customers run single threaded 100 hour jobs?

    I do, so there are others, we might be in a minority, but we exist.

    I do run other threads at the same time, I run 6 different tasks of similar types of difficulty.

    I can see where you are coming from, I welcome the time they become available though

  • jsgjsg Member, Resident Benchmarker
    edited November 2020

    @NewToTheGame said:

    @jsg said:

    @NewToTheGame said:
    15% of a 100 hour single threaded job is still 15 hours, I'll take it

    Sure - but then, how many percent of hosting customers run single threaded 100 hour jobs?

    I do, so there are others, we might be in a minority, but we exist.

    I do run other threads at the same time, I run 6 different tasks of similar types of difficulty.

    I can see where you are coming from, I welcome the time they become available though

    Well, I couldn't possibly consider all use cases and I do understand that for you the 5000 might be attractive. But for easily 95% of all VPS users - most of whom frankly do not even really need a 3000 - what (little) the 5000 offers over the 3000 just isn't worth it.

    Consider this: To switch to the 5000 will translate to switching the main board (or even the whole system) too, to at least for some time serving basically as a lab rat (until the 5000 has it's quirks ironed out in the firmware), etc. Plus, one also needs the OS, the libs, and the apps to actually make use of may 5000 features.

    We saw how it went with the Zen 2 which had a hard time at first because one (a provider) couldn't simply buy another cheap 2nd hand box but needed to buy a brand new box (ca. 10 to 30 times the cost), etc. Also keep in mind that those Ryzen 39xx boxes are just at the beginning phase of amortization and many providers with Ryzen 39xx boxes will not just take the losses and spend thousands and thousands of $ again for a 59xx..

    This is especially true for us here in the Low End segment which - for good reasons - still is largely based on old 2nd hand Xeons ...

  • @jsg said:
    Plus you pay in terms of electrical power consumption because obviously those 5+ GHz top frequency gobbles up power and not modestly.

    Well the TDP works out the same for the 12C/24T and 8C/16T CPUs, the 6C/12T has a lower TDP on Ryzen 5000 series - with improved IPC, you're effectively getting more performance per watt - which if systems are mostly idle, it also means it can spend more time in C2, thus actually give you an overall lower power consumption compared to 3000 series.

    If you run your 5000 (or 3000 series) in performance mode, it's not like it's gonna run at the TDP constantly :) It will sit in C2 state as much as possible. The same is the case for EPYC 7002 series - an 7402P (which is a 180W TDP CPU) sits at 55-60 watt in performance mode when the system isn't doing much (~5% load), around 75-80 watt when it's 20-25% loaded.

    Having a better IPC, means overall more time spent in lower C states, thus saving power.

    Whether the improvements are worth it, despite the "premium" obviously is up to the individual providers to decide - and obviously the improvement differs a lot depending on the type of workloads. Some workloads will have far better improvements than 19%, some will be lower.

    The answer is simple: More cores (even slow ones) are better than fewer cores

    Obviously depends on the workloads, thus the answer is not simple. Optimize for your workload, and choose the hardware that makes the most sense.

    Providers have to look at the price/performance ratio, performance/power ratio as well as the revenue point of view. If they, for a small increase can yield the same performance but higher revenue, and possibly at same or lower power consumption, then it's a win for them.

    If people get the most bang for the buck when it comes to old gen xeons, then let them do that :-D We all have different needs and requirements, both as providers and as customers.

    Obviously everything you've said is your view of things, luckily.

    I think providers may have a different view than you do.

  • serv_eeserv_ee Member
    edited November 2020

    @jsg said: You pay the "AMD premium" of about $50 that AMD introduced with the Ryzen 5000 series, or in other words, now that they feel they are not the underdog anymore.

    Well, they don't "feel". They know they aren't. Not trying to put words into your mouth but saw the same kind of comments on AT earlier and people are "upset" because AMD raised prices...they are a company after all who needs to earn profit to please the stockholders (WHY ON EARTH DID I SELL AT $6?!). They know they have the "top dog" so they must to ask $$$ accordingly.

    Same goes for the GPU department. As soon as they are on par with nvidia you can bet your ass they will raise the price.

    Side note and my personal opinion. As soon as AMD has enough free cash flow they should just buy back glofo and get some life into it with their own fab instead of using TSMC. (Yea yea I know glofo is far being with their nodes - to be fair I don't even know on what node they are now since I haven't checked for a long time)

  • Please run Autodesk 3DS Max to test the real performance

  • Biggest thing is that these will make the used market obsolete just look at 3100 and 3300x beating a fucing 7700k with lower power usage

  • jsgjsg Member, Resident Benchmarker
    edited November 2020

    @Zerpy said:
    [power consumption/TDP]

    You see me mistrusting numbers that still are largely based on marketing info. The way I see it (until proven otherwise) is based on two quite reliable rules:

    • Higher frequency ~ higher power consumption and maybe a bit higher performance
    • smaller gate size ~ lower power consumption or higher performance

    Now, obviously both AMD and intel choose a mix, but still, the gate size was and still is 7 nm but frequency has increased. Also note that AMD has said that the total power envelope stayed the same (142 W); they didn't say that power consumption generally stayed the same.

    The answer is simple: More cores (even slow ones) are better than fewer cores

    Obviously depends on the workloads, thus the answer is not simple. Optimize for your workload, and choose the hardware that makes the most sense.

    For most workloads in our area what I said is true.

    Providers have to look at the price/performance ratio, performance/power ratio as well as the revenue point of view. If they, for a small increase can yield the same performance but higher revenue, and possibly at same or lower power consumption, then it's a win for them.

    ... modulo the fact that they need to buy brand new and hence much more expensive systems ...

    Of course, in theory you are right, but in real life, you will have a hard time selling new and expensive systems to (especially low end) providers, particularly when they just recently shelled out tens of thousands of dollars to buy some Ryzen 3000 nodes.

    If people get the most bang for the buck when it comes to old gen xeons, then let them do that :-D We all have different needs and requirements, both as providers and as customers.

    Obviously everything you've said is your view of things, luckily.

    I think providers may have a different view than you do.

    I don't think so, especially when looking at low end providers.

    Don't get me wrong, Zen3 is not somehow sh_tty, nope, it's a very nice processor and in fact a bit faster than the 3000. My point is that, unlike the 3000, it is not that much better than its predecessor that providers, especially low end providers - and that's what LET is all about, after all - will hurry to buy new and expensive systems.

    My guess is this: Very few LE provider will even consider to buy Zen3 systems now. Plus I doubt that the usual cycle "Some buy new ones, others will then buy their old ones 2nd hand" will work well, because the 3000 systems haven't amortized yet and still are expensive even 2nd hand, way more expensive anyway than a 5 year old Xeon system.

    Finally I'm speaking based from my own experience. Maybe you and me can afford to play the "what the heck, I'll pay $10/mo and get a hotter VPS" game but I guess most users here(and in the LE market) are paying more attention to "what do I really need" and to "bang for the buck" than to high-end performance.

    Thanked by 1RedSox
  • jsgjsg Member, Resident Benchmarker

    @serv_ee said:

    @jsg said: You pay the "AMD premium" of about $50 that AMD introduced with the Ryzen 5000 series, or in other words, now that they feel they are not the underdog anymore.

    Well, they don't "feel". They know they aren't.

    Fair point. I worded it badly.

    Not trying to put words into your mouth but saw the same kind of comments on AT earlier and people are "upset" because AMD raised prices...they are a company after all who needs to earn profit to please the stockholders (WHY ON EARTH DID I SELL AT $6?!). They know they have the "top dog" so they must to ask $$$ accordingly.

    I'm not upset and at least for the time being IMO one still gets a better deal than at intel with Zen3. my point is whether that price increase is worth it. For 16 core system it may seem small but for e.g. a 4 core system +$50 isn't that small anymore.

    Same goes for the GPU department. As soon as they are on par with nvidia you can bet your ass they will raise the price.

    I can't reasonably comment on that because I do not care about graphics cards and GPUs. At all.

    Side note and my personal opinion. As soon as AMD has enough free cash flow they should just buy back glofo and get some life into it with their own fab instead of using TSMC. (Yea yea I know glofo is far being with their nodes - to be fair I don't even know on what node they are now since I haven't checked for a long time)

    Hmm, I'm not sure but I seem to remember that AMD actually had and possibly still has the "janitor" (IO etc) chiplet made by GloFo; it is, or at least was, 14 nm. Only the core chiplets were/are made by TSMC in 7 nm.
    As for buying GloFo back, nuh, I don't think so. Frankly GloFo is a bit behind. I think that AMD for quite a while will be better off (and way safer) staying a customer (with heavy weight and lots of influence).

    @2gouzi said:
    Please run Autodesk 3DS Max to test the real performance

    Certainly not. If and as soon as I get my hands on a Ryzen 5000 system I will run my benchmark and possibly some MySql and nginx benchmarks. That is the relevant load here.

    @codelock said:
    Biggest thing is that these will make the used market obsolete just look at 3100 and 3300x beating a fucing 7700k with lower power usage

    "Obsolete" is probably too big a word, but yes I also think that the normal cycle we all know from Xeon based systems is shaken.

  • jsgjsg Member, Resident Benchmarker

    Supplement: It seems the Zen3 Epyc is a much more attractive processor which isn't targeted at the gamers but at the server-clientele.

    I just came across a report that shows a 2 socket Zen3 Epyc (64 cores) system pretty much on par with four intel Platinum (28 cores) processors! And that within a typical Epyc power envelope.

    Very attractive. I'm looking forward to get more information on the Zen3 Epycs.

    Thanked by 2vimalware coreflux
  • this turn like NFS 1 , 2 , 3 , 4 , 5 , 6 , 7 .... i lost the number ... finnaly is only cars.

  • ZerpyZerpy Member
    edited November 2020

    @jsg said:

    Now, obviously both AMD and intel choose a mix, but still, the gate size was and still is 7 nm but frequency has increased. Also note that AMD has said that the total power envelope stayed the same (142 W); they didn't say that power consumption generally stayed the same.

    Regardless of anything, higher IPC, means lower utilization to get the same calculations done - thus you can idle more.

    The same was the case between zen1 and zen2.

    Luckily the TDP on AMD systems actually tend to be fairly realistic, unlike Intel.

    For most workloads in our area what I said is true.

    You define "our" area as everyone on this forum has the same workloads. They don't. If you believe so, then you're simply silly.

    ... modulo the fact that they need to buy brand new and hence much more expensive systems ...

    They don't have to buy brand new systems.

    Of course, in theory you are right, but in real life, you will have a hard time selling new and expensive systems to (especially low end) providers, particularly when they just recently shelled out tens of thousands of dollars to buy some Ryzen 3000 nodes.

    Obviously providers only spend the money if it makes sense - whether they go for 3000 series or 5000 series is up to them. Some people may have motherboards laying for expansion (I know there's a decent amount of providers here that does), if they've not yet bought the CPUs, they can easily justify the 5000 series.

    If someone bought 3000 series 6-12 months ago, they're unlikely to go buy 5000 series and replace their 3000 series CPUs. That's not how providers work.

    Don't get me wrong, Zen3 is not somehow sh_tty, nope, it's a very nice processor and in fact a bit faster than the 3000. My point is that, unlike the 3000, it is not that much better than its predecessor that providers, especially low end providers - and that's what LET is all about, after all - will hurry to buy new and expensive systems.

    There's plenty of providers here with good hardware, that targets more high-end users than you - you're assuming everyone on LET, actually only buy low-end stuff - which also is a false assumption to make.

    Finally I'm speaking based from my own experience.

    Yes, that's clear.

    but I guess most users here(and in the LE market) are paying more attention to "what do I really need"

    No, they really don't - rather the opposite.

    and to "bang for the buck" than to high-end performance.

    Again, depends on what people use their VMs for - some people might have multiple requirements, thus looks for different specs.

    Thanked by 1TimboJones
  • jsgjsg Member, Resident Benchmarker

    @Zerpy said:
    Regardless of anything, higher IPC, means lower utilization to get the same calculations done - thus you can idle more.

    The same was the case between zen1 and zen2.

    Well, that is one way to look at it, but it isn't mine, among other reasons the fact that with plently Ryzen 3000 systems around, there is no urgent need.

    Luckily the TDP on AMD systems actually tend to be fairly realistic, unlike Intel.

    Yes, indeed.

    For most workloads in our area what I said is true.

    You define "our" area as everyone on this forum has the same workloads. They don't. If you believe so, then you're simply silly.

    I didn't say that everybody has the same workload. (a) I said "most" - not "everybody", and (b) sure, one can turn and bend it but fact is that 85%? 90%? over 90%? of all low-end VPSs run some kind of web stack and/or some other kind of server software like chat, email, etc. The situation probably is different, albeit not really by that much, on more expensive VPSs.

    ... modulo the fact that they need to buy brand new and hence much more expensive systems ...

    They don't have to buy brand new systems.

    Great! Kindly provide a link to a 2nd hand Ryzen 5000 server.

    Oh, you mean, the socket is the same as the 3000 socket? Right, but does the BIOS, firmware, etc. support the 5000? And why exactly would a LOW END provider - and it's those we talk about here - shell out about €1500 (or more) 5950 and sell his 3950 with about €1000 loss to gain 10% to 15% performance? And that while his system isn't yet amortized?

    Obviously providers only spend the money if it makes sense - whether they go for 3000 series or 5000 series is up to them. Some people may have motherboards laying for expansion (I know there's a decent amount of providers here that does), if they've not yet bought the CPUs, they can easily justify the 5000 series.

    Yes, indeed. IF BIOS, firmware, etc support the 5000. Plus, of course, they must be ready to pay about €1000 premium fee on top.
    Frankly, I don't see many low end providers doing that. What I do expect is that they wait for the 3000 (2nd hand) getting cheaper due to the 5000 being what most run after now.
    Which is nice for us because it translates to more Ryzen, albeit "only" 3000 based VPSs becoming available.

    If someone bought 3000 series 6-12 months ago, they're unlikely to go buy 5000 series and replace their 3000 series CPUs. That's not how providers work.

    Uhm, thanks for agreeing with me.

    There's plenty of providers here with good hardware, that targets more high-end users than you - you're assuming everyone on LET, actually only buy low-end stuff - which also is a false assumption to make.

    BS! This forum is called low end talk for a reason. And btw, you again (intentionally, it seems) mis-interpret what I said. Of course, there are some not low end customers here, but those are a quite small minority I (reasonably) guess. And, yes, of bloody course many providers here do not have low end products only but according to the rules and reason they mainly sell low end products here.

    but I guess most users here(and in the LE market) are paying more attention to "what do I really need"

    No, they really don't - rather the opposite.

    and to "bang for the buck" than to high-end performance.

    Again, depends on what people use their VMs for - some people might have multiple requirements, thus looks for different specs.

    Pardon me but you get boring. I'm talking the "85%" and you again and again talk about exceptions - which I did not exclude.

    Thanked by 1lovelyserver
  • Well, instead of writing another blog, I'd summarise it like this:

    1. if you have Ryzen3k/4k & not looking for max performance, stick with it. Not worth to upgrade to 5k (now with with these intro-prices)
    2. if you are buying cpu now, AMD is better option when it comes to price/performance/power (but it might change in future)
    3. as everything new, Ryzens5k are expensive (AMD can afford it, as they delivered something really good). It you do not need new cpu right now, wait a few weeks
  • @jsg said:
    Great! Kindly provide a link to a 2nd hand Ryzen 5000 server.

    Not in that way - but you can use Zen1 and Zen2 mobos, assuming bios has been updated - which is the case - thus not brand new.

    Oh, you mean, the socket is the same as the 3000 socket? Right, but does the BIOS, firmware, etc. support the 5000? And why exactly would a LOW END provider - and it's those we talk about here - shell out about €1500 (or more) 5950 and sell his 3950 with about €1000 loss to gain 10% to 15% performance? And that while his system isn't yet amortized?

    There's Ryzen sellers here - if they have 3000 series, they have 3000 series, if they're buying more machines for expansion, they can put in 5000 series.

    There's also people who already have mobos etc laying - thus just need the CPU.

    Yes, indeed. IF BIOS, firmware, etc support the 5000. Plus, of course, they must be ready to pay about €1000 premium fee on top.

    There's no premium of €1000.

    BS! This forum is called low end talk for a reason.

    Righto. BS for not knowing people.

    And btw, you again (intentionally, it seems) mis-interpret what I said. Of course, there are some not low end customers here, but those are a quite small minority I (reasonably) guess. And, yes, of bloody course many providers here do not have low end products only but according to the rules and reason they mainly sell low end products here.

    They can still have Ryzen systems for low-end products.

    Pardon me but you get boring. I'm talking the "85%" and you again and again talk about exceptions - which I did not exclude.

    You are boring too - because you get all sour when someone disagrees with your view. But that's typical you anyway.

    Someone disagrees with your view, and you get all offensive about it ^_^ Kinda funny mate.

    Have a beautiful evening sunshine :kissing:

    Thanked by 1TimboJones
  • jsgjsg Member, Resident Benchmarker

    @Jarry said:
    Well, instead of writing another blog, I'd summarise it like this:

    1. if you have Ryzen3k/4k & not looking for max performance, stick with it. Not worth to upgrade to 5k (now with with these intro-prices)
    2. if you are buying cpu now, AMD is better option when it comes to price/performance/power (but it might change in future)
    3. as everything new, Ryzens5k are expensive (AMD can afford it, as they delivered something really good). It you do not need new cpu right now, wait a few weeks

    Add to that that we only have some first preliminary tests and reports based (at best) on engineer samples.

    @Zerpy

    I'll cut it short. Reality is different from what you seem to think.

    We can reasonably expect that many (maybe even most) manufacturers will have a working BIOS and firmware when the 5000 is officially launched in a couple of months. Then, according to experience, it'll take another one to three months for problems being ironed out. Then we'll have reliable mainboards to bet ones business on.
    Similarly, it'll take up to half a year for compilers to really support the 5000's new features. And then another half year (or more) for the distros to really support the 5000's new features.
    Until then the 5000 is but a slightly pimped up 3000 - but with an ugly twist: Having been designed by another team the 5000 may look like an "AMD Zen" on the outside but there more likely than not will be quirks that will only be seen once thousands of software packages are running on it.

    Re "no €1000 premium" - well that's about the difference between what you'll pay for a new 5950 and what you'll get for your 3950.

    You will pay more for a Ryzen 5000 dedi (and I'm not talking about 10%) as wrt to us in the low end market about the only thing one really gets to know about a VPS vCore is that it's usually not a Ryzen core.

    As of now providers will calculate like this: They can add 10% to 15% vCores but for that they'll have to either buy a brand new node, or take a loss of about €1000 when upgrading their 3000 based node. To be able to sell a few more vCores that's not worth the trouble.

    Later, much latter (as in 2 or 3 years later) your POV will be right because by then 2nd hand Zen2 systems will be available in reasonable quantities. Yet 3 years later those systems will be cheap (as in sub $1k) - and that is where the vast majority of low end nodes come from.

  • @jsg said:
    As of now providers will calculate like this: They can add 10% to 15% vCores but for that they'll have to either buy a brand new node, or take a loss of about €1000 when upgrading their 3000 based node. To be able to sell a few more vCores that's not worth the trouble.

    Later, much latter (as in 2 or 3 years later) your POV will be right because by then 2nd hand Zen2 systems will be available in reasonable quantities. Yet 3 years later those systems will be cheap (as in sub $1k) - and that is where the vast majority of low end nodes come from.

    No, you really don't get it :)

    Thanked by 1TimboJones
  • jsgjsg Member, Resident Benchmarker

    @Zerpy said:
    No, you really don't get it :)

    Well, time will tell ...

  • msg7086msg7086 Member
    edited November 2020

    If you are paying $50 more for the CPU it's not going to increase the whole system cost by a large portion. $50 more on a $250 CPU sounds a lot, but on a $700 system it's slightly less important. On the other hand, if it performs better and will end up saving on electricity cost, it may still be worth it.

    For 99.9% users, it's not as useful. We used to run Athlon X2 220 in our office about 10 years ago and people still do their daily work fine. Today you can perfectly run a Ryzen 3200G for all kinds of office work. Mid to high end Ryzens are for gamers and enthusiasts. Like, who needs 6 cores to watch youtube or write posts?

    For "web" hosting, epyc would be a much better option. People use Ryzen may use it for encoding-like work.

  • jsgjsg Member, Resident Benchmarker

    @msg7086 said:
    If you are paying $50 more for the CPU it's not going to increase the whole system cost by a large portion. $50 more on a $250 CPU sounds a lot, but on a $700 system it's slightly less important. On the other hand, if it performs better and will end up saving on electricity cost, it may still be worth it.

    For 99.9% users, it's not as useful. We used to run Athlon X2 220 in our office about 10 years ago and people still do their daily work fine. Today you can perfectly run a Ryzen 3200G for all kinds of office work. Mid to high end Ryzens are for gamers and enthusiasts. Like, who needs 6 cores to watch youtube or write posts?

    For "web" hosting, epyc would be a much better option. People use Ryzen may use it for encoding-like work.

    Yes, that pretty much is my view too. As for the Zen3 Epycs: Yes, those seem to be really attractive, yay! AlthoughI need to note that most Zen VPSs seem to be Ryzen based, probably for the higher performance/core (whereas Epycs are more about many cores and high total system performance. But frankly, I think that even Epic cores (even Zen2) would be bloody good enough for VPSs that blow Xeon nodes out of the water).

  • I was hoping/waiting for 3950x to drop in price to upgrade my aging Xeon 2689 workstation. But seeing the performance chart maybe 5900x is a better buy.

  • jsgjsg Member, Resident Benchmarker

    @Faizi said:
    I was hoping/waiting for 3950x to drop in price to upgrade my aging Xeon 2689 workstation. But seeing the performance chart maybe 5900x is a better buy.

    If you need to buy a mainboard anyway and if you really need maximum performance, in particular single core performance then, yes, the 5950 is a better buy than the 3950 for you.

  • PulsedMediaPulsedMedia Member, Patron Provider

    So many issues with OP that barely got to 2nd paragraph before i could think "sigh another Intel shill or clueless".

    AMD achieved real world average 19% IPC gains. This is ALWAYS dependant on your task.

    The CPUs barely touch 5Ghz, and clock frequency does not equal power consumption. They achieved simply by microarch design advancements and node maturity this. Zero extra power consumption. When AMD says it is 65W that typically holds true in real world.

    "Not" Battle worn / tested motherboards: Dude, they are literally the exactly same motherboards as before. Literally. Take old motherboard, apply BIOS update, insert new CPU. Srsly, this is not Intel. First gen Ryzen motherboards will be curious which ones support, but it's upto the mobo vendors if they want to go out their way.

    @jsg said: TL;DR front-up: if you are a gamer, hurry to get the new super-processor with 19% higher IPC!

    There are single thread bound, latency sensitive server workloads too. I know, because our core #1 piece of software is one of those. Barely any threading, and always blocked by the main thread ...

    @jsg said: Let me explain: Zen 1 to Zen 2 really was a significant step.

    And core per core this is just as significant step.

    @jsg said: The Zen3 on the other hand (a) has been designed by another team, and (b) seems to be centered around and focused largely on one point: to conquer the one area where intel still is/was the king of the hill, single core performance - which also translates to gaming.

    So your grief is that they focused on the single area they were not #1? All righty then.

    Another team: This is what AMD has told you constantly, they have 2 teams each leapfrogging each other. This is not news, nor a bad thing. This methodology allows AMD to get so much done with so few people.

    Now 6 cores roughly matches 8 cores, and 8 cores matches close to 12 core performance in variety of tasks. This is a bad thing exactly how?

    @jsg said: Gaming, that must be seen, is a very attractive market segment for both AMD and intel because the clientele is very easy to manipulate and also (comparatively) easily pays premium prices.

    For Intel gaming is quite small. for AMD gaming is quite large (consoles). And targeting a lucrative market is bad exactly how? It is called business 101.

    Further, DIYer / gamer chips are very cheap. Just think how many 16 core gaming chips you can buy for the price of just one 32 core frequency optimized EPyC? ... Those are about 4k each street price.

    ...

    Ok i'll stop here because my time is more valuably spent elsewhere than trying to correct misinformation and FUD.

    Then again it was fun, not everyday you see someone trying to turn positives into negatives..... Oh wait nvm, you see that constantly by the mainstream media

    Thanked by 2TimboJones easy
  • jsgjsg Member, Resident Benchmarker

    @PulsedMedia said:
    So many issues with OP that barely got to 2nd paragraph before i could think "sigh another Intel shill or clueless".

    AMD achieved real world average 19% IPC gains. This is ALWAYS dependant on your task.

    Funny. In fact all data available so far are based on engineering samples and in-house code. What you will see once the 5000 is launched and using available and commonly used compilers is a rather different story. Possibly a nice one - but as of now there is nothing "real world".

    As for "intel shill", Haha, I'm writing this on a Ryzen system and I'm known to strongly prefer AMD over intel, I'm also known though for not blindly and fervently favoring either.

    The CPUs barely touch 5Ghz, and clock frequency does not equal power consumption. They achieved simply by microarch design advancements and node maturity this. Zero extra power consumption. When AMD says it is 65W that typically holds true in real world.

    Thanks for repeating the AMD marketing blabla - because that's what it is. Probably largely true but still but marketing blabla.

    "Not" Battle worn / tested motherboards: Dude, they are literally the exactly same motherboards as before. Literally. Take old motherboard, apply BIOS update, insert new CPU. Srsly, this is not Intel. First gen Ryzen motherboards will be curious which ones support, but it's upto the mobo vendors if they want to go out their way.

    Nice theory - that does however not hold up in real life. For a start Ryzens (unlike Epyc) are consumer processors and there has been plenty "catch me!" in that segment. Catch me as in e.g. "oopsie, the xyz chipset doesn't really (or at all) support the new processor or as in "theoretically chipset xyz can support that processor but it depends on your mainboard firmware support" or as in "oopsie, the current version doesn't support 3200 MHz memory but just 2400 MHz".
    I happen to know because I had to play those catch me games with Ryzen and Ryzen mainboards. That is not to say that Ryzen is sh_tty, but it means that one should reasonably expect quite some hiccups when Ryzen 5000 is launched and that many will need new mainboards.

    @jsg said: TL;DR front-up: if you are a gamer, hurry to get the new super-processor with 19% higher IPC!

    There are single thread bound, latency sensitive server workloads too. I know, because our core #1 piece of software is one of those. Barely any threading, and always blocked by the main thread ...

    How boring, yet another guy with an application that is far outside the 85+% of what runs on most VPSs. I never doubted that there are outliers, so no need to provide yet another example.
    Oh and btw, good luck with pushing those extra 15% performance out of a shared VPS.

    @jsg said: Let me explain: Zen 1 to Zen 2 really was a significant step.

    And core per core this is just as significant step.

    Well, we disagree on this one in the context of this thread which is about Ryzen 5000 in the low end hosting market and, to a degree, as a non-gamer workstation.

    @jsg said: The Zen3 on the other hand (a) has been designed by another team, and (b) seems to be centered around and focused largely on one point: to conquer the one area where intel still is/was the king of the hill, single core performance - which also translates to gaming.

    So your grief is that they focused on the single area they were not #1? All righty then.

    Nope, no grief, and I iwsh AMD good sales with the 5000. What I said was that we, the hosting market and particularly the low end hosting market were not the target market.
    Btw, if you really knew what you pretend to know you'd know that significant changes in a processor are two-sided; they can be a blessing but they also can be a curse for really speed critical software

    Now 6 cores roughly matches 8 cores, and 8 cores matches close to 12 core performance in variety of tasks. This is a bad thing exactly how?

    IF that turns out to be true with after product launch and with generally available compilers, libraries, software, etc ... it's not a bad thing. Experience shows however that every processor manufacturer shows his products in the best light which usually means that it at least takes a lot of time for code that is not custom tailored by the manufacturers team (using profound inside knowledge, special tools, etc) to even come close in performance.
    In reality btw, which I happen to know as a developer we often compile e.g. for Nehalem that is, a stone age architecture because one usually can't afford to either only care for a tiny part of the market or to maintain dozens of versions.

    @jsg said: Gaming, that must be seen, is a very attractive market segment for both AMD and intel because the clientele is very easy to manipulate and also (comparatively) easily pays premium prices.

    For Intel gaming is quite small. for AMD gaming is quite large (consoles). And targeting a lucrative market is bad exactly how? It is called business 101.

    >
    BS! The gaming market is a very significant market both for intel and AMD. And No, it's not bad to target that market - but it's bad for us.

    Ok i'll stop here because my time is more valuably spent elsewhere than trying to correct misinformation and FUD.

    Yeah, right, go and set up some new seeding rigs using Epycs. Oopsie.

  • @Faizi said:
    I was hoping/waiting for 3950x to drop in price to upgrade my aging Xeon 2689 workstation. But seeing the performance chart maybe 5900x is a better buy.

    Xeons also can handle queued jobs much better.. so, its not just performance! (PS: A gamer's processor is not necessarily bad choice, if the price is at least 60% or lower compared to a Xeon)

  • PulsedMediaPulsedMedia Member, Patron Provider

    @jsg said: Funny. In fact all data available so far are based on engineering samples and in-house code. What you will see once the 5000 is launched and using available and commonly used compilers is a rather different story. Possibly a nice one - but as of now there is nothing "real world".

    I don't know what are all those reviews and benchmarks we've been seeing then?
    There's dozens of them.

    We got a few Zen 3 in the mail right now, i believe we will receive them tomorrow.

    Further, AMD has been in the past few years been quite accurate with their claims. They do not need to lie like Intel, they are on top, so no need to lie.

    @jsg said: As for "intel shill", Haha, I'm writing this on a Ryzen system and I'm known to strongly prefer AMD over intel, I'm also known though for not blindly and fervently favoring either.

    Well you act like one, spreading FUD. Easy to disprove FUD too. Like above one, these CPUs are launched and in the wild.

    @jsg said: Thanks for repeating the AMD marketing blabla - because that's what it is. Probably largely true but still but marketing blabla.

    Certainly everything company says is more or less marketing, but that's from their technical docs. It is a new microarch, certainly based on the last few gens, but a major overhaul never the less.

    You keep on making yourself sound even more and more like Intel fanboy.
    Just saying you are writing this on ryzen system is zero proof of anything, you can say anything. Obviously we've seen you don't like to stick to the reality.

    @jsg said: Nice theory - that does however not hold up in real life. For a start Ryzens (unlike Epyc) are consumer processors and there has been plenty "catch me!" in that segment. Catch me as in e.g. "oopsie, the xyz chipset doesn't really (or at all) support the new processor or as in "theoretically chipset xyz can support that processor but it depends on your mainboard firmware support" or as in "oopsie, the current version doesn't support 3200 MHz memory but just 2400 MHz".

    I happen to know because I had to play those catch me games with Ryzen and Ryzen mainboards. That is not to say that Ryzen is sh_tty, but it means that one should reasonably expect quite some hiccups when Ryzen 5000 is launched and that many will need new mainboards.

    There was no new chipset release, there were barely any new motherboard releases (1 from Asus i believe). and that "catch me!" thing, what a load of shit. Support has been rolled out to more different chipsets than AMD promised for. Zen 2 chips for example work on friggin' A320 boards even tho they are not supposed to.

    Next months will show us how many 1st gen (A320/B350/X370) boards will support Zen 3. 2nd gen will support.

    There are videos of people running on a 40$ A320 board a Zen2 16core CPU.
    Still, even if 1st gen does not arrive, it's tons better than Shittel get new mobo every generation.

    3200Mhz memory -> That's actually overclock. So of course it's not on the official marketing. Having built dozens of Ryzen systems, not a single one has refused to operate 3200 - 3600Mhz memory, even at 1st gen release time.
    People are running even 4000Mhz memory just fine, infact Zen3 benefits from it -- and i hear going for 4 modules instead of 2 also increases performance, which was the achilles heel on 1st gen. Sounds like they did a major overhaul on the memory I/O ...

    Ryzen 5000 series is already launched, and there were almost no hiccups at all even under Linux. Check Wendell's video on Level1Linux channel for this.

    @jsg said: How boring, yet another guy with an application that is far outside the 85+% of what runs on most VPSs. I never doubted that there are outliers, so no need to provide yet another example.

    Oh and btw, good luck with pushing those extra 15% performance out of a shared VPS.

    You were talking about consumer chips. So you are spreading FUD by moving the goal post.

    Well i'll bite; If you need say 64 cores, you still benefit from IPC uplift.
    Again, that IPC uplift has been proven and benchmarked to be there, some actually got on average 21-22%. On Cinebench it was way more infact.

    VMs run same types of software as "non-vm regular baremetal servers", there is no magic pixie dust making software inside VM act any differently. All VMs regardless of use take a few % hit on performance on average (we actually have a customer for whom doing VMs uplifted performance by orders of magnitude -- kernel limitations were met!). With Intel you take upto 80% hit on I/O ... tho that's regardless of being in VM, but with VMs you are kinda forced to keep the security mitigations enabled. For single user firewalled server it's fine to disable them in many cases (not all).

    @jsg said: IF that turns out to be true with after product launch and with generally available compilers, libraries, software, etc ... it's not a bad thing. Experience shows however that every processor manufacturer shows his products in the best light which usually means that it at least takes a lot of time for code that is not custom tailored by the manufacturers team (using profound inside knowledge, special tools, etc) to even come close in performance.

    In reality btw, which I happen to know as a developer we often compile e.g. for Nehalem that is, a stone age architecture because one usually can't afford to either only care for a tiny part of the market or to maintain dozens of versions.

    Tested and benchmarked, all software is available etc. Ryzen 5000 series is still x86_64, that has not changed. Check Wendell on Level1Linux channel on his tests.

    These uplifts happened, they are very much real.

    @jsg said: Nope, no grief, and I iwsh AMD good sales with the 5000. What I said was that we, the hosting market and particularly the low end hosting market were not the target market.

    Btw, if you really knew what you pretend to know you'd know that significant changes in a processor are two-sided; they can be a blessing but they also can be a curse for really speed critical software

    Nope, again with the lies. You are spreading FUD and trying to discredit Zen 3 as being somehow even worse than decade old Nehalems are. or that somehow Zen 3 performs lower than Zen 1 did.

    So now you try to discredit me too. That's called an ad hominem, look it up.

    @jsg said: BS! The gaming market is a very significant market both for intel and AMD. And No, it's not bad to target that market - but it's bad for us.

    DIYers in general is something like 15-20% of their respective revenues, nothing to sniff at, but SI and DC markets are much larger. Gamers are subset of DIYers.

    However, in AMDs case, they get to sell all those chips in PS5 and Xbox series, so for AMD it's very significant over the next few years. Probably something like 20% of their revenue for the next 1-2 years will be somehow related to Gaming, nothing to sniff at but when the DC market is probably around 30-40% and rest is SI. Tho DC and SI intertwine a bit.

    @jsg said: And No, it's not bad to target that market - but it's bad for us.

    Higher performance on same wattage is never bad.
    You keep claiming that more performance for same wattage is somehow bad, but you want more performance for even less wattage. You might need to visit a doctor or therapist, because that's some serious serious cognitive dissonance.

    Thanked by 1vimalware
Sign In or Register to comment.