Ryzen vs i7 (Mainstream); Threadripper vs i9 (HEDT); X299 vs X399/TRX40; Xeon vs Epyc

I expect Intel prefer play the Mobile game first. A lot money in this segment. The rest will come afterwards. They won’t let AMD play alone in the backyards. Remember Intel come after the mainstream chips and cleaned up with more performance vs. cores (both 8700K and 9900K). Just look at newest 8 cores from AMD vs old Intel’s tech. Max out cores clocks and AMD still struggle with Intel’s old. But newest chips out will have an advantage. It’s newer. Newer can’t be worse than previous gen? But yeah, in several things new ain’t always better Just look at firmware, software, drivers and OS. + Turds.

 

I have outlined my workflow many times here, an 'exact' workflow I can't post. Please search my most recent posts in the past year or so to find the most updated version. I really do not have time right now to find it for you.

But please, don't get sidetracked with what my workflow is. That is just a red-herring that is being thrown by the AMD fanboys that has nothing to do with my questions to them and AMD about the 'many-core' push AMD is making and recommending to everyone, blindly.

My workflow is not a consideration at all in the current posts I'm making and which started from when Ryzen was originally discussed.

I am asking for the last couple of years at least what an AMD platform advantage has over an Intel platform for most consumer workloads. Not you or I here, I mean people who buy a computer as an appliance.

From the direct answers I got from people who built AMD based platforms; there is none. There are probably even regressions still (with Ryzen 3) because of drivers, firmware and software niggles compared to Intel platforms which show no such glitches. From what I've read. I don't need to post those links here, we're all aware of them and I don't want to discuss that (that is not the topic).

My question still stands. Years after I originally posted it. And after the third-gen of AMD hardware has been released, there is still not one sign that can claim AMD's marketing BS that more cores = moar, for those users.

AMD has never been competitive for me personally. Precisely because of the glitches with software, drivers and firmware since ~2001 when I started to seriously test them again at that time. Again, that is my experience (wouldn't even install the software I was using then) but I have seen many examples where an Intel system would just work and a fully updated and freshly installed AMD platform would fall on its face, in one way or another. These were not my systems; they were clients of mine that eventually converted to Intel too. That is not a great user experience at all for AMD. I have seen this improve in the last few years, but still not to an 'it just works like it's supposed to stage'.

We, or at least I, am not talking about mission-critical work here. Mass-consumers with amateurish needs.

And for the mission-critical work I am doing, we have found our own solutions to ensure the security of our IP. Don't forget, a platform is just the base you start off with. There are many more pieces to the puzzle to make it work for a corporation.

In a nutshell, I am calling AMD out that 'many-cores' is to be blindly recommended for most consumers. They will be in a tech disadvantage for the rest of their platforms' life.

In the last three years, AMD and their followers kept stating that bringing the many-core hardware will allow developers to flex their many-core program skills, I have seen not one example of that.

In the meantime, Intel has a much more balanced and holistic approach to the products they release and introduce. With real, day-to-day, improvements for each generation.

Many don't want to see that platform improvement. Their loss. I not only see it, I fully exploit it as needed too. These are after all just tools, not the holy grail.

An example is the Ice Lake platform that will be showing up any day now. Easily worth waiting for even for users of FB and twitter. If just for the battery life improvements that are promised.

Please note that my statements above don't imply that I upgrade on every single generation. I may or may not depending on what improvements are offered from my current platform.

But for people that 'need' a (mobile) computer today (yeah; this forum is still NBR)? The latest Intel offering with 16GB of RAM or more, a 1TB or larger SSD and the most powerful processor you can afford is an easy and reliable recommendation, still.

For myself, when I 'was able to play' with the most recent AMD Ryzen desktop platform (3600, 16GB RAM, PCIe x4 SSD) for a few minutes what I missed most was the instantaneous nature of my Intel-based systems.

That is an instant loss for AMD in my books. Laggy is so 1990...

 

That video is useless. It doesn't impress me at all either.

That looks like a very well executed upgrade for a gen2 product (10nm).

Doesn't matter how they get a performance increase (in actual productivity, battery life or any other usability aspect), as long as it is increased.

The Ferrari I will order when I win the lottery won't be based on what engine is inside, it will be based on what it can do on the track.

 

Circular argument, the Ferrari purchase may not be based on the engines inside, but what it does on the track most definitely will!

 

NO.

The spec's of the engine can only guess what the car will be capable of in the real-world. Or, they may be made up specs too.

The track times don't lie.

The circular argument is that spec's and 'scores' predict what the actual results achieved will be.

 

Well, with due respect, I haven't got the time to go digging through an entire year's worth of posts just to find your 'workflow'. Until you qualify exactly what you do with your computers—which, incidentally, is hardly difficult nor time-consuming: for instance, I develop programs in C, C++, Java, and Python; use Mathematica to model and view scenarios for my assignments; typeset documents in LaTeX (specifically, pdfLaTeX); transcode media with ffmpeg and edit videos with DaVinci Resolve; browse the Web with Chrome; view and edit PDFs with Acrobat; for entertainment, watch Blu-Ray rips in PotPlayer or stream Netflix, and play some of the more demanding AAA-studio role-playing games like The Witcher 3 and Assassin's Creed: Odyssey—I will consider it a vague, hand-wavy word that has no real meaning, and no context.

This is a pattern I've noticed in some of the posts here, especially in those attacking AMD and Ryzen: isolated, anecdotal arguments, and wilful ignorance of 'synthetic benchmarks' in favour of 'real-world performance', with no context given for the latter, either. These arguments are highly unconvincing, when I find that a straightforward benchmark does what it says on the tin: put differing hardware under an identical, repeatable, fixed computational load, so that they may be objectively compared. Synthetic benchmarks are as scientific as they get, and 'real world' performance is an iffy, vague, murky, muddy and generally unclear term used to justify a non-argument.

What is this platform improvement you mention? All I see from Intel is a half-decade-long stagnation in computational power and core count; nickel-and-diming; and gigantic security holes that result in significantly worsened performance once said holes are plugged. Many more. I apologise, but the way you put it, one would think you're being paid to see the Sun shining from every orifice of Intel's.

Easy. Web-browsing, with webpages loaded with ECMAScript applications, heavy use of 'interactive' CSS (see any of Apple's recent product webpages, for instance), have increasingly become heavily multi-threaded. I daresay the everyday user needs multi-threading more than ever, because the every-day user is multi-tasking more than ever. Video games have increasingly advanced AI and offload rendering draw-calls to idle cores, and said video games see a marked improvement in performance with CPUs with more cores.

This automatically negates your argument, because you are using an anecdote—from a mere handful of minutes of experience.

Moar cores=moar power is absolutely true, which is why my university's computer science department has recently opened modules pertaining to parallel programming, GPGPU computation, multi-threading and such. Or are you going to say this:

I still am curious (and you haven't answered what your workload and 'workflow' is—I tried looking for it over the course of typing up this reply and I gave up, because I see increasingly vague references to some pie-in-the-sky) what it is that you do with your computers. Do you write code for the nuclear missiles or something?

 

Not all can and or will show or talk about their workflow. That has to be respected. In general if you need more cores you go after that. But not at the expense of whole workflow. Some need both from both worlds. You can’t get that always from one chip manufacturer.

 

yeah, he works primarily with photography and programs related, which are still heavily serialized as far as processing goes. Photoshop and I believe after effects are included in the software suite he uses. So, for him, depending on the filters applied and how much time is spent in lightroom (he does use that one) and possibly after effects, along with a couple other things, there is an honest chance that a mainstream 9900K is the best for his specific workflow, and considering the 9900K has been out 9 months about, give or take a month or so, there may be very little reason for him, specifically, to consider upgrading the systems.

But, he has this problem of not giving context, not acknowledging where AMD is winning, etc. He makes ambiguous marketing like statements, like platform benefits, rather than digging deeper. He doesn't even acknowledge if you are a content creator, those new PCIe 4 drives seem pretty exciting.

Also, even when I discuss his workflow at the programs he uses level, he then snaps back that I know nothing of his workflow, further creating ambiguity without reason.

No, it does NOT need respected. He is making bold claims without any facts or data to back it up. He hides behind an ambiguous concept of his "workflow" without giving context, then pushes a product, which by many objective standards disagrees with the existing data, while having ZERO nuance given to when choosing AMD may be a better choice for the purchaser.

At least you bench, give screenshots, etc. He gives none of that. We know you own your product, you show the exact performance you are getting out of it, then say "show me an AMD 8-core that can do this." So you are giving context and data, and doing a fine job of overclocking (your hobby). He gives none of that and has amorphous statements.

He doesn't even have to talk about the plugins for the apps, etc., he just needs to state the apps themselves so that people know what type of workload is being done, not specifically what he is doing.

And even you acknowledge, as do I, that Intel is still better at some things, or that having different machines with chips from each mfr would be better. I whole heartedly agree. But that also isn't what tiller says. There is no nuance to his statements. That is what is being attacked along with a lack of information for people to properly weigh the truth or falsity of his statements.

 

Playing devils advocate but unless he uses programs that have a major advantage of multi core AMD may not benefit him as an upgrade from a 9900K.

 

Or and he use an mobile platform(soldered hardware) in his workflow.

 

Tiller and I had a little discussion about this a while back (on a public forum on this site so no qualms about posting it). The way he describes it it sounds like a lot of it involves highly parallelized humans looking at images to find the best ones, with precious little processing applied to the images. Think human pattern recognition.

For something like this, lowest latency for viewing (and maybe minimal processing) sounds like what's needed, but more processing power may not be too helpful (although there's at least some research on parallelizing JPEG decoding, http://modul.repo.mercubuana-yogya....urnal/OpenJournalOfBusiness/customer/92_2.pdf). But really tuning that would probably better be done on a GPU than on a CPU anyway.

Essentially, it's an embarrassingly parallel MIMD kind of problem, but that scales with the number of humans, not the number of CPU cores/threads. For something like that, sure, I can believe that i7-9700K or even i5-9600 might be ideal, maybe with a decent GPU if he has the image decoding so implemented. He might not be using commercial apps at all, so talk of apps and plugins might not even be relevant. Screenshots again might not be relevant even if he could display them without giving away his game. Photos of his work site would probably be more to the point.

So I think I have enough information to take him at his word on this. But what relevance this has for anyone else -- this is a very specialized workflow -- is questionable.

 

A big part of the problem is that tiller projects that his workflow is applicable to everyone else.

 

Thank you, I couldn't remember where it was.

Now that that is cleared up, how about you compare that to his statements on AMD in the past couple pages, or saying AMD is not right for consumers at core count or price point and using ambiguous language on platform, etc., to try to tell people don't get a third gen AMD?

 

As I said: I consider this to be a very specialized workflow, and it's unclear to me what broader relevance it has. Tiller is responsible for his own opinions.

 

As I mentioned if we are having the production people just having to load an image or simple page a 3900x will do little to help, but it will not hurt either rather just being overkill. With fast nVME dives any Zen2 will be fine and older Ryzen may be slightly slower, less snappy, than an Intel system.

That is the main thing I am looking forward too. TR3 with its faster ST performance will have lot more snap in its step. Again can not wait to see the 32 core in action.

 

Although...with, say, a 3700X (or even a 3600), you could use a PCIe4 NVMe; if you're throwing around really big images, that might actually make a perceptible difference.

 

not really. all i see him talk about is workload, buy off based off workload. when did he ever say his workload applicable to everybody.

 

Tillers only problem is he claims he, and others, are better benefited by Intel offerings. His statements seem to be the majority if not all others too. This could have been arguably true pre Zen2 but not anymore.

 

dude, where did u get your hands on a quadro RTX mobile card? details please

sorry about the OT guys, carry on

 

Hehe, it wasn’t MXM, it was a DGFF card I got. Details here and in later replies.

 

This was a recurring theme in discussion of SSDs where tiller would say that over provisioning by a minimum of 30 percent is necessary to keep a drive from going bad. In a data center or with a heavy, specialized workload this might be needed, but pretty much anyone else can get away with half that, if any at all.


Sent from my iPhone using Tapatalk

 

its not untrue to what he said though and that is especially dependent on the ssd and it's flash and also how soon people decide to replace their drive. with QLC coming out that becomes even more important. 30% is great because a crappy drive will give less performance issue when some cells start to go bad.

if people are okay putting up with performance degration then they dont need provisioning.

 

Those discussions took place long before QLC came out. I want to say it wasn't long after TechReport did their extended SSD endurance test, which showed that all the drives they evaluated lasting well beyond their official specifications.

 

Oh, yes, I remember these I recall I was planning to upgrade my sister's old notebook's HDD to an SSD (never ended up buying one, as the laptop was sold), and asked for suggestions, and Tiller suggested I not go for the then-cheapest BX100, and over-provision any SSD I got by 30%. Nope, not happening on a consumer SSD.

Tiller suggests that Intel is reliable, when every single media outlet, technically-inclined or otherwise, suggests that Intel CPUs are a bad buy right now because of huge security holes. That is the definition of not reliable.

AMD's manycore mindset will definitely bear fruits in the not-too-long future, because software, from media processing to gaming, to IDEs and others, is becoming parallelised. There exist plenty of embarrassingly-parallel operations that will greatly benefit from things like 12, 16, 24, 32 cores. And needless to say AMD has been pushing the price-performance boundary as far as it would go. If AMD had released a viable platform for notebooks, I would've bought said notebook straightaway.

 

TLC isnt exactly great either. if you're moving back in time then firmware was junk back then compare to what we've got now.

worse firmware with TLC
better firmware with QLC

both got share of it's problem. like i mentioned, what he said isn't untrue and as a storage maniac myself i am all about storage performance so i agree with him.

 

New doesn’t always mean better... But all depends on what’s the usage.

Your Next SSD Might Be Slower (Thanks to QLC Flash) - HowToGeek
22. jul. 2019 · Newer isn't always better. Recently, SSD manufacturers have begun to trade off speed and reliability in the interest.

 

I almost bought that Asus laptop with the Ryzen 7 3750H and GTX 1660 Ti, but the fact that the CPU is a 4-core/8-thread part put me off. I ended up getting an i7 8750H/GTX 1050 Ti one on clearance for less than half the retail price of the Asus anyway. I won't be playing modern AAA games on it, but for titles up until around 2014-2015, the GPU is more than up to the task.

 

If you really care about absolute maximum write performance, fine, go ahead and overprovision like that. That's a tradeoff individuals have to decide on for themselves. But it could be a rather expensive tradeoff for a lot of people -- one that really isn't worth it in most non-data center cases, in my view.

 

Clock speed and IPC improvements are rapidly diminishing in returns; the only other ways to improve performance are custom hardware (e. g. special purpose FPGAs for crypto mining) and parallel programming (which includes GPU use -- GPUs, in addition to display hardware, also contain massively parallel SIMD attached processors).

Whatever exactly Tiller does, he's essentially (on his own description) doing some kind of parallel image recognition and minimal processing, but with the parallel element being done by people rather than a computer. I rather suspect that at some point either he or somebody else in his field will come up with some kind of deep learning to recognize images, and that will surely be a highly parallel operation at a per-image level, at which point those cores will come in handy.

His call, obviously, but high IPC, low core count isn't likely to be long-term the way forward for that kind of thing.

 

actually its both read and write. ssd performance is constantly getting slower as you use it until theres almost no good cell left then decline becomes much sharper.

 

What is the mechanism for read performance getting worse with decreased free space? I could see greater fragmentation (regardless of how full the drive is) requiring more I/O ops to read the same amount of data from a file, but it doesn't have the same issue as writing.

SSDs normally have a fair bit of overprovisioning built in to start with, and between TRIM and garbage collection, things are a lot better than they used to be.

 

Please start, or use, a separate SSD thread if you wish to continue that discussion.

 

a comparison; https://www.forbes.com/sites/erikka...l-i9-9900k-vs-amd-ryzen-9-3900x/#2f274bdd54b0

 

So, AMD Epyc servers just took 80 World Records!
https://www.amd.com/en/processors/e...rldrecords&utm_medium=redirect&utm_source=301
Edit: This is on the heels of Intel saying next year they will have a SOCKETED 56 core Cooper lake. Hmmm.... World records and performance with an upgrade path, or bet on Cooper lake on 14nm and potentially the Ice Lake chips whenever they appear. Hmmm.... Intel tried to steal AMD's thunder, but with their non-appearance on the 56-core Cascade CPUs, versus AMD having actual 64-core chips in OEM and ODM systems without such restrictions on availability, with the Cascade-AP chips being BGA, enough said.

Edit 2: Mellanox working with AMD to bring 24 PCIe 4.0 NVMe partnered with their 100 and 200Gbps network adapters. Yum!
https://www.businesswire.com/news/h...gxNjXHjsNmKaXEz4koEK2KQOELXF1Klv27yoigQe38g==

Edit 3: Benchmarks on storage for a Giga server on storage:
https://www.storagereview.com/gigabyte_r272z32_amd_epyc_rome_server_review

 

@ajc9988 eypc clock speed skus https://wccftech.com/analyst-amd-to-gain-10-server-market-share-with-rome/

 

For deep dives into the Intel Xeon vs Epyc debate, I present to you ServeTheHome and Anandtech's reviews (spoiler, AMD did more than amazing!!!).
https://www.servethehome.com/amd-epyc-7002-series-rome-delivers-a-knockout/10/

https://www.anandtech.com/show/14694/amd-rome-epyc-2nd-gen

Edit: Phoronix data:

https://www.phoronix.com/scan.php?page=news_item&px=AMD-EPYC-7742-2P-Summary

https://www.phoronix.com/scan.php?page=article&item=amd-epyc-7502-7742&num=1
https://www.phoronix.com/scan.php?page=article&item=amd-linux-epyc-7002&num=1

AMD saying they are working on open-source Bios/Coreboot
https://www.phoronix.com/scan.php?page=news_item&px=AMD-EPYC-Coreboot-Open-Austin

 

you dont need to post so much to show AMD is ahead. just this graph is enough to shows how powerful epycs are.

https://www.anandtech.com/show/14694/amd-rome-epyc-2nd-gen/14

avx512 intel vs just avx eypc, just wtf intel?? their only advantage of avx512 is now gone no more reason buying intel in enterprise space.

 

I do because each show it in different ways. Serve the home is the most in depth. Anandtech is in depth, but is NOT serve the home. Moreover, Phoronix focuses on Linux in their own way.

Did you see serve the home showing the 100Gbps NIC running 40% slower on Intel? Or the storage throughput with that system?

Believe it or not, even though some overlap, each of those is unique in their coverage.

It is NO DIFFERENT than showing the coverage from multiple sites for the dropping of an HEDT or mainstream product.

While Anand is good and in depth, their coverage isn't the same as STH, which dives really deep, showing a 2Ux4 node 2017 Intel system can now be replaced with a single AMD node.

So, read the coverage before saying such things.

 

all people needed to know is that intel's only advantage in avx512 is gone. but sure you can link them to phoronix as they do linux stuff, which has nothing to do with us consumers unless you plan to run linux and purchase eypc.

now we shall wait to see how and what zen4 bring to desktop might be able to beat intel, or intel can make a return with tigerlake when zen4 becomes available. in consumer space intel technically has the advantage due to monolithic design for overall cpu + ram latency as well as high clocks, for 14nm++ anyway. tigerlake should be on 10nm++ so i expect fierce competition between amd and intel. highest clock + highest ipc is the way to go, cheaper and more cores and extra power saving i consider them as big bonus too.

 

I titled this thread as showing, also, Epyc vs Xeon. And I don't think you understand exactly how much these details matter to those interested in the topic. The only difference is I did not editorialize or analyze the articles, instead allowing them to stand on their own.

Hell, if you read the STH article, you might have focused on their analysis suggesting the performance gap is so great that TCO for many customers is so much lower that it even can beat Intel on per core licensing because of the energy savings, rack reductions, etc. Think about that! Acquisition costs plus costs of running it and ROI make it so that even taking a hit on per core licensing makes AMD the better deal and that argument will likely NOT save Intel. That is even more crazy than the AVX512 news!!!

 

thats the thing i didnt read STH's article, at least not yet. but its not hard to figure out that intel only had few areas left they could still be competitive and avx512 was the biggest of them all, which is now consider nulled. besides AMD went the smarter route asking highly parallel workload to be done on GPU which is a better replacement than having avx512 done on CPU.

i dont see how that cost thing is a bigger deal than avx512. these companies and actually majority of companies have spent and use to spent that much with intel, imho i dont think they care. they do understand going AMD will bring cost saving in power, performance per watt and performance on all front is eypc wins. but switching to something new may always have issue because so use to the old, and i dont blame enterprises not switching, at least for now.

example would be me switching to AMD would mean i no longer able to use throttlestop and i'd need to spend more time learning AMD OC or even possibly installing new windows with most of my software. alot goes into that and needs planning, not a matter of saving intel or not, dont honestly give a damn about that.

 

Intel's arguments include that they have the single threaded performance, so when it comes to power core licensing of software, like running windows server in certain licenses, you are better off with Intel than taking the license cost hit with AMD, even if your workload is faster on AMD with more cores. It is part of the TCO analysis.

Now, the fact you can save so much with AMD that you still can save significant amounts of money while paying the higher fees under per core licences changes the game. It isn't just one workload type being conquered, it is every time per core licensing pops up. AVX512 is a very small part of the actual server workload ecosystem. Per core licensing is much larger segment of data center and enterprise. Because of that, it means procurement officers must now redo their analyses on AMD servers entirely. That is why that is a bigger deal!

 

st performance shouldnt matter in enterprise side of thing. i guess theres probably some VERY limited workload would benefit from ST performance but be honest that should only apply to monolithic design on desktop. the mesh system up the latency drastically compare to that of desktop so their advantage on the desktop side over zen2 is basically all gone. and yeah most enterprise getting so many cpus theres no real need for ST performance at that point, AMD also have PBO to take care of that concern.

3950x laptop 5ghz thanks amd.

 

Well, you should have read it. I think that then you'd understand why presenting a single benchmark graph wouldn't do justice to EPYC.

That NAMD chart really isn't very interesting. If you're running a highly specialized AVX512 workload, then there's no compelling reason to switch. Even if you're not, the CPU performance differential really isn't all that compelling, maybe 50% tops. That sounds like a lot, but it takes a big capital expense (and organizational things, like training) to take advantage of it.

Thing is, there's a lot more to it, like the I/O improvements. Look at the charts for 100GbE performance. EPYC (Rome) simply blows Xeon clear out of the water there. And server workloads tend to be a lot more I/O-intensive than desktop workloads.

And the security improvements are nothing to sneer at. 500+ VMs can have their memory independently encrypted such that even if the hypervisor (much less one of the VMs is, or is actively hostile) is, it can't get at it.

People who run servers for a living -- and that's the target audience of STH -- simply don't care about throttlestop or overclocking. But as you note, switching vendors might mean a significant change for you. When you're talking thousands of servers, that's greatly amplified. That's why EPYC needs to be more than just better on a few top-line benchmarks. And that's what the STH piece went into, in great depth, in language that server farm administrators will understand.

 

Not sure if this was posted or not - new vulnerability was discovered:
https://www.techpowerup.com/258039/swapgs-another-speculative-side-channel-vulnerability

There is a performance impact for Intel CPUs after mitigation patch, although pretty small.

 

too much work for reading into something i'll probably never come across of using it for what i need. if it's something interesting and something i'll be using down the road then yeah but the article will be there. so when the times come i'll go back and dig it up no problem.

 

Well, OK -- but then don't complain about it being posted. It's directly on point to the EPYC vs. Xeon comparison.

 

all i see is you and ajc complaining LOL.

 

They made a post and you were the complainant. Essentially too much info, I have to agree with too much is better than not enough.

 

Self explanatory;