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.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
@PulsedMedia
What a bunch of BS that's not even worth to be quoted and responded to in detail. Plus, of course, as always when ammunition gets scarce ad hominems ...
"Available" != "some not small customers getting 3 processors". "Available" means "officially launched, can be purchased in many stores".
Some youtube channel as reference, uhum, sorry we are playing in different leagues it seems.
Me - allegedly - painting Ryzen 3000 worse than Xeons ... What a pile of mentally deranged BS!
In fact I'm a big fan of the Ryzen 3000 (and happily using some 3000 based VPSs from NexusBytes).
I simply don't see why anyone with a Ryzen 3000 based server should sell the 3000 and buy a 5000 at a loss of ca. 1 k€ in order to get "19% more single core performance" unless he really, really desperately needs it - and that, well noted, based on your assumption that just unplugging the 3000 from and inserting the 5000 into the mainboard socket is all that's needed.
Plus, you seem to not know the Ryzen that well after all, it seems. Hint: The significant change in Ryzen and memory speed was when AMD stopped hard-linking mem. and Ryzen speed a while back. I still remember when I got the Ryzen 1700 and (fast clocked) 3200 MHz RAM and, on a good brand name mainboard, didn't get more than 2400 MHz out of it. It took more than a year and multiple BIOS revisions - some of which created havoc - to finally get it stable and running at least at 2800 MHz.
When I later got a Ryzen 3000 things were much better and went much smoother but still one had to carefully choose a board with the right BIOS to later have the 4000 supported too.
https://wccftech.com/amd-ryzen-5000-zen-3-desktop-cpus-running-on-a320-x370-motherboards-b450-support-added/
Support coming for Zen 3 even on some A320 boards.
You are funny fellow, it's you who started with those ad hominems etc. as you cannot refute anything, and objective beyond shadow of doubt are lying as fast as you can.
Input to google "Ryzen 5000 reviews", and you get a huge list:
Some random youtube channel only? https://www.youtube.com/results?search_query=ryzen+5000+review
Wow, what a goal post move. So now we are talking about disassembling existing server, selling the cpu and plugging a new one in just because new one is available! X)
You are full of shit, you know that right?
For new servers, Ryzen 5000 series is an obvious choice when performance matters but you want to stick to Ryzens instead of EPyCs. If you just need the memory channels and PCI-E lanes, Zen 1 EPyCs like 7551 is quite an obvious choice for it's price. but a Ryzen 5950x is likely going to be faster than EPyC 7551 despite massive core disadvantage, in both multicore and singlethread, while consuming much less power (~30% less)
It all depends on what you need.
Ryzen 5950X is what 800€, and a 3950X would fetch a nice chunk of change too, so the cost would probably be more like 500€. Of course it is not worth it, when has it been worth it since the 90s when doubling of performance in under 2 years was the norm?
No, but you pick a 5950X over 3950X for new servers, when comparing apples to apples, new to new. New 3950x is 670€, while 5950x is 800€. A mere 130€ difference in price. For total system cost we are quickly talking a "rounding error", well atleast for us. a X570 server we'd like to put 6 or 8x 16TB drives for example. 670€ or 800€ for CPU becomes quite irrelevant difference at that point, what matters is the lifecycle. 5950x will have 1 extra year at the end of it's life cycle, and that 1 extra year might bring back 3500€.
To bend this from iron girder for you, a low end 6 drive system could cost something like this for a server provider and yield something like this:
Total without CPU: 3105€
Total w/ 3950x: 3775€
Total w/ 5950x: 3905€
Cost difference: 3.44%
3950x:
Lifetime electrical 5years: 9015MWh @ 0.12€/kWh (Hetzner Finland) 1081.90€
Lifetime Housing cost 5years: 1U @ 347.37€ (Hetzner Finland)
Total lifetime OpEx costs (sans bandwidth, support, failures etc.): 1429.27€
Total cost over lifetime: 5204.26€
5950x:
Lifetime electrical 6years: 10 818MWh @ 0.12€/kWh (Hetzner Finland) 1298.16€
Lifetime Housing cost 6years: 1U @ 416.84€ (Hetzner Finland)
Total lifetime OpEx costs (sans bandwidth, support, failures etc.): 1715€
Total cost over lifetime: 5620€
Cost Difference: 7.99%
Lifetime Difference: 20%
Assume you calc for 12month ROI, and ignore rest of cost, system end user prices:
3950x: 314.58€
5950x: 325.42€
Difference still same 3.44% for end user, for a juicy 20% performance boost for CPU related tasks.
As a owner however, if the best case happens, servers remains at original price for the full lifecycle for one customer (happens surprisingly often with dedis!)
3950x gross 18 874.80€ net 13 670.54€
5950x gross 23 430.24€ net 17 810.24€
Difference: 30.28%
Guess which system as a DC owner i'd like to put online more? The 5950x
Guess which system a potential customer likes to have with such a small price gap? The 5950x
I would say PEBKAC was your issue, and there is plenty of evidence 1st gen works just fine on 3200+ memory.
Personally i've built quite a few 1st gen systems, with various quality motherboards -- every single one of them has clocked to XMP ratings not a problem, if you remain with 2 modules. With Zen 3 this issue has been fixed. The infinity fabric and mem clocks were "separated" in Zen 2 indeed, but not completely, that's why you need high speed ram. You needed to apply bios update if you had issues which was released like 1 month after Ryzen 1st gen release, but never experienced any issues on release date bios myself tho. First chip tested was Ryzen 1200 lol, that OC'd to 4.15Ghz 24/7/365 100% load stable at hot environment, using Ripjawz 3200 ram kit which was not supposed to work at all, used it at garage for years before selling and upgrading.
Also at that time built dozen or so Ryzen 1200 based GPU mining systems which also CPU mined. All of them with 3200 to 3600 ram kits, and all of them 24/7/365 100% load stable.
We've not had a single issue, hell we even run ASRock A300s as mini servers, built custom fan shrouds etc. for them, all of them are rock solid at XMP ratings, from Athlon 200GEs to 3400Gs, using SO-DIMM with no heat spreads no less. They are very stable ... if they just had AST2000 chip in there ....
We don't build that many Ryzen/EPyC systems tbh, but i believe in the past 3 years it still has to be 100 systems or more i've personally built.
I am sorry to tell you this, but you clearly have no clue, no understanding of the tech. or business requirements.
65W holds true on what exactly in the "real world"? Actual power draw? If thats what youre implying then I got some bad news for ya.
5950X TDP is 105W. It will use more than that at about 120W or maybe a bit more under full load. BUT it still uses less than 3950X
But I guess we were talking about the 5600X, it's pretty damn near 65W at a full 100% blender load
120 isn't 105 is it now. And by others it actually goes around edit: 215 from the wall under load.
That wasn't really my point tho. Saying 65W on the box doesn't mean 65W. Hell, I'd be glad if the 9590 was actually "only" 220W but it's not even close to that. TDP /= power consumption.
Totally agree with that, but in this case 67W and 65W are pretty damn close.
TDP is used as a rating for heat output for you to know what cooler you should get, not as a metric to know how much amperage you should hire when putting the server in a DC.
Hey, I wasn't the one implying that in the first place lol
That the CPU roughly draws that in real world production. At least that has been the case for us. I know TDP comes with about 50 of *, but it's been roughly that for us.
It's not like intel where you have ~double it to get close (or take measures to get it back down to advertised ratings, and nerf the already weak performance further). Well at least on all the systems i've seen or owned for a decade now.
I also measure from the wall when doing comparisons of what uses what, as that's ultimately what matters.
Exactly this -- needs actual measurements. My point is that what AMD says tends to be closer to reality than one would expect.
Like 5600x says 65W and takes ~67W from the EPS rails, that's pretty darned close don't you think? 105W TDP and 120W consumed is a bit off indeed, but it's still not 250W or so what would the Intel counterpart of "105W" be.
@PulsedMedia
It's rare that I call something a wall of text but your last post is, and frankly I simply skipped it after a bunch of questionable assertions and attempts to insult me (or utter lack of self-discipline).
Example for totally bending things: it was YOU who painted it like "Just pull the 3000 and plug in the 5000 and be done". I merely (and with good will) based some of what I said on that, YOUR, assertion.
Anyway we'll see who got it right. If it's you then we should rather soon see a flood of Ryzen 5 (I don't hold my breath ...).
It probably doesn't hold true under all circumstances like e.g. under full load but generally speaking AMD has been much more honest with their TDP numbers than intel. It seems though that the TDP numbers for the 5000 series, particularly for the higher end ones are quite a bit off (or "preliminary"), so for colocating (where power consumption is a major cost factor) I'd suggest to first run tests with your particular work load.
I btw. do not really hold that against AMD as the 5000 is targeting the gamers who seem to care little about power consumption anyway. They seem to care much about peak IPC however and that is also what AMD underlined in their marketing material.
I am sorry but sounds like english is not your native language. Re-read what you read, because that is not what i said.
I am not sure you know what IPC stands for, sounds like you think IPC means Instructions Per Core. It stands for Instructions Per Cycle. That's static, there's no peak. Different OPs have different performance, but it remains the same constantly. What peaks and fluctuates is clock rate.
By few days I read this thread. It is interesting in a way. Unfortunately, has come a bit rude now.
Why don't you buy some 5950x or 3950x or ... 798735999x (i'm joking with this one) and post some really nice offers here to make all is happy? Low end prices, please. And some free VMs, why not.
Marketing is marketing. Plus 19% performance, 125W or 65W ... Thier factory tests can be very different than ours, in practice. Of course, the marketing has to give a plus to one of them, to pull down the other competitor.
Black Friday is around the corner. Just buy some 5950x and give us a reason of joy. I will not pay more than 7 USD per month 😜 for another idling VM. It must be low end price.
(No offense for anyone, of course)
my Intel Core2 Quad 2.83 GHz (4 cores) with TDP 95w (now overclocked to 3.2 GHz) bought in 2010 is still working fine and hosting multiple mid-traffic WP sites. Will be keeping this machine for say another 4-5 years!
Cool story, bro!
Too long, didn't read, LOL, so all I can say Intel sucks hard in desktop, even harder in servers and now with the new Epycs round the corner it's gonna be embarrassing
What do you hope to achieve with your glaring misinterpretations? I think his statement is perfectly understandable, but if you are having difficulties, maybe you should ask people to clarify rather than making things up.
Regarding your claim about IPC being static, well that's not true at all. IPC depends a lot on the use case, instruction sets, pipelines, memory and overall interaction between different components of hardware and software. It is only a measured average.
Regarding this thread's actual topic, Ryzen 5000s are a huge leap forward in my opinion. We have seen a lot of horizontal scaling in the past few years, but little vertical scaling in comparison. 64 cores are relatively cheap nowadays. Unlike with vertical scaling, scaling to multiple machines (e.g. 4x 16 cores instead of 64) is also feasible for multi-threaded workloads. Therefore, one is much more likely to be bottlenecked in a single-thread scenario compared to multi-threaded.
Thanks to Ryzen 5000, game servers or other software with tight "main loops" or poor load spreading are finally seeing improvements. The same holds true for many desktop software, so I am very pleased with the new Ryzens. But I agree that for the vast majority of providers it does not make sense to invest in Ryzen 5000 at this time (reasons have already been mentioned, typical usage patterns in contrast to niche single-threaded workloads).