DDR3 v2 VS DDR4 v4 nodes

[teisie]

Hello, Id’e like to raise a poll if DDR4 v4 should be getting more rewards than DDR3 v2

I’m not discoursing ddr3 nodes, but I think the difference should be as also in the price it’s a 3/4 fold difference.

My opinion:

Looking at companies like AWS, iCloud etc it’s very likely all these DataCenters they have is massively build in the newest hardware. Als for this reason they can build a stable and fastest services. Looking at TFT we’re having ddr3 v2 nodes, v3, ddr4 v3, v4. These are all different nodes, with different investments. Also of course knowing your places to be will also reflect much in the price.

But on the explorer there is 0 difference, this means when a user deploy’s a server for price 10TFT(for example) and get’s a ddr3 v2 and user 2 get’s a ddr4 V4 unit with the newest Proliant for 10TFT.

I believe this is not the way to build a stable and trusted network, when deploying a server online somewhere it’s also very clear what kind of hardware your deploying. Here on TFT it could be anything from even DDR2 to DDR5.

I don’t want to piss of the people running ddr2 or ddr3 nodes, but i paid a fair price for DDR4 360 g9 proliant units and I think it’s a bit unfair you het the same for an older unit that’s 1/3 of the price (like on als HP or dell unit)

I really hope all the people react understanding this is not cuite helpfull for ddr3 users, but if we wanna keep TFT alive I think this a must as also bussines la person I wouldn’t use this not knowing what kind of hardware I’m deploying. Especially if it’s at peoples home, it could be Even ddr1 or 2.

Summary

This text will be hidden

[Dany]

Hey @teisie,

I get your point here. But I think it’s not only the performace of RAM that needs to be taken into account. The overall performance of a node should be rewarded when providing newest high-quality hardware. But additional rewards should only be granted when reliable and optimal farming conditions are met too. That requires datacenter conditions like redundant power distribution/source, redundant ISP connections at high speed, secured uptime of >99% and so on. This is already in progress here GEP Gold Certified Farming Specs but does not differentiate hardware performance yet. Maybe the certifed farming programm can be split into different levels (like silver, gold, platin) where the highest level would also take HQ-hardware performance into account.

[weynandkuijpers]

Hi @teisie. ThreeFold welcomes older hardware. Older hardware which is given a second chance after serving a first cycle in Enterprise or major cloud setup is very sustainable. The speed and energy consumption improvements achieved are still relevant for some use cases (physical space and power restricted setups like traditional datacenters) for for a decentralized model like the ThreeFold project refurbished hardware are a perfect solution to run workloads.

For people looking for speed and the fastest machine out there we might present a little bit more information is for example the explorer. I have been working (personal project) on creating an flist that runs a few tests and send the result back to me. This way I can figure out what network speed is available in a certain node, what location it is in and some basic information about performance of the hardware. This way people consuming grid capacity are able to find the best possible hardware for their use case.

I think it’s fair to say that there are two main camps in both grid builders (farmers) and grid consumers: one side like sustainability and is happy with having refurbished hardware, the other wanting to be using the latest and greatest hardware for their use case. I am not sure that providers should earn different as they are serving two different audiences, who are specifically looking for an emphasis on sustainability for their use case, or performance.

[teisie]

Hé @Dany, indeed yes. The CPU benchmark would really play a role in what kind of node you’re running.

I think you’re silver, gold, platin idea is really good, only this really brings in unbalance between farmers. But anyway this is already the case, i think it’s a perfect idea everybody knowing upfront what they need to have minimum for a certain level.

[FLnelson]

I saw your specs on your other comment. It may be the highest quality stuff on the grid. Once sorting by hardware becomes a thing, I would assume your hardware will be utilized first. If that is set up correctly, you will get TFT for that network traffic.

[longstreet]

Mixed feelings here. I think, as would be expected, @weynandkuijpers is pretty on point. The tier idea for certified builds isn’t a bad idea either, but would be validated not only by the credentialed builder but by a certain set of benchmarks as he mentioned he is starting to test. That said, I am a big fan of reusing enterprise hardware that somebody who can write off upgrades replaces on an accelerated cycle. I run some decent personal workloads on DDR3 based systems. Usually dual Xeon, specked to the max proc and front side bus speed the MB can handle. But again, that would be reflected in some form of benchmarking. For non-enterprise budget type buyers, NIB cutting edge enterprise hardware is obscenely expensive. The ratio between cost and performance difference is not a good one in my opinion, based on a general use case. I could go on forever, as I have a lot of thoughts on this, but this isn’t the time, especially as my phone battery is about to die. Lol. Cheers to all!

-Robert

[teisie]

Indeed I agree DDR3 is working and there diferent options between nodes.

But my question: if there diferent nodes to choose from so DDR3/DDR4, there should be a price difference right. If there is not, everyone will go for the best ofcourse like @FLnelson says.

But I would love to see price difference’s like all other dedicated server companie’s, and that wil also meen there is more room to give ddr4 more tokens for in use nodes.

I just wan’t something to know that i’m not 100% wasted money for buying ddr4 and not ddr3 having 3 times more units. Besides that it’s newer ofcourse and easier to get rid but that’s not what I’m for in this project ofcourse.

Because right now a $1900 ddr3 v2 server get’s as much as a $4000 Ddr4 v4.

I still think the best thing is to set difference prices ranges for diferent performance on clients side. Because as you say there diferent kind of nodes performance wise for different kind of usecases. This will result in DDR 4 getting little more on in use rewards.

[jeroenburjs]

To be fair, i don’t think we can make a reward system based on system cost, only on value provided.

in that sense it would be logical to have 2/3 tiers. and every year/2year you should add a tier because there is new hardware available.

Than the grid users will indicate to the farmers which systems they prefer by using them…

[teisie]

Yes true agree, this also keeps the grid from users view up to date at all time.

Maybe move this thread later to the DOA

[TheCaptain]

This could prove to be an unpopular opinion but as of right now, based on the metrics you’re rewarded on, it makes absolutely no sense from a CapEx perspective to be building out ddr4 nodes at scale. The fact is farmers are rewarded on quantity over quality and to some degree this makes perfect sense - if you go to AWS or (insert cloud provider) you’re typically given the choice of how many cores and how much ram you’d like, not how fast those cores are, their micro-architecture/die size, ram speed etc. The specifics behind the technology employed arguably has more of an impact on the provider (in terms of energy efficiency, warranty etc) than the client (who is often going to be limited by network bandwidth before they notice any bump in compute speed).

[FLnelson]

Yeah, sorting by CPU is 100x more important than DDR3 vs DDR4.

[teisie]

if you go to AWS or (insert cloud provider) you’re typically given the choice of how many cores and how much ram you’d like, not how fast those cores are, their micro-architecture/die size, ram speed etc.

Thnx for the response.

I just looked this up. And this is AWS but every other cloud company shows exactly the opesite. They show everything from:

• NVME or SATA
• Processer type and gen with Ghz speeds
• Internet speeds
• Garunteed uptimes (SLA)
• Free bandwidth
• DDR3 or DDR4

You’re right on AWS but are we AWS? People trust in them so prob don’t need to see the specs. Company wise i would like to see and know the specs of a rented machine. Because every other cloud company is also showing them.

i’m talking about comparing, and comparing to websites like:

Dedicated Linux servers: veel power & lage prijzen | STRATO

Intel® Servers with Xeon® processors | IONOS

https://www.ovhcloud.com/en/bare-metal/prices/

They all show really clear what there providing. They have options between DDR3 and 4, but majority is all DDR4.

Then there is TF grid where i could be deploying a ddr2 to machine and big change a ddr3. You get the point, i understand you with

based on the metrics you’re rewarded on, it makes absolutely no sense from a CapEx perspective to be building out ddr4 nodes at scale.

But there must be a reason for poeple to put on better nodes otherwise this grid will forever be at the same level and i understand you because you (Probably) have ddr3 nodes as seen in the chat (Correct me if im wrong) but look it from the perspective in the long term. For how long will this ddr3 be good? And for how long do we want to farm.

I see the point of recycling etc but there just must come a reason to put on better hardware, seeing all these companies majority all running ddr4 and we never can compete at that level is just like not giving everything and staying here and accept we won’t run any better and lose lots and lots of users. I think still the majority of people using want new stuff and no recycled.

[scott]

I just looked this up. And this is AWS but every other cloud company shows exactly the opesite.

You’re looking at dedicated nodes. VPS offerings tend to include much less information about the underlying specs. Typically there are a few varieties that are optimized for different things. There might be a minimum CPU speed on the CPU optimized instances or certain SSD type/performance level specified. RAM speed is typically not specified for a VPS.

Anyway, I agree that we should differentiate between hardware with different performance levels, for the purposes of farming rewards and also allowing Grid users to choose the best hardware for their needs. To me, it’s less important which generation of technology is in service than the actual measured performance it produces. We’ll know more about that when our benchmarking tool comes online.