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

TANWare, I certainly don't understand your stance here? You don't have to change your mind about me if I really am spreading FUD as you say/think; I know I won't change one way or another (I'll always seek the actual truth; not what is spoon fed to me).

The actual words don't have to be used to be inferred, correct? I am more than capable of applying previously gained knowledge to this seemingly 'new' situation and seeing the similarities - even if you can't. Again; I'm not asking you to continue this discussion if you don't want to... but no matter the complexity and the cleverness of the design used to make non-monolithic chips communicate as if they're one; they were still attached together somehow. 'Glue' is as good a word as any to suggest how it was done.

I have no agenda to push other than presenting factually all aspects of the topic of discussion. This is what I do in each/any thread I participate in.

I can't and won't admit to resorting to FUD when the truth is so far from it. Facts are facts. A monolithic die is different than four dies in a single package. Use any word you like to describe how they are attached to each other - and I'll agree to it.

But what I hear you saying is that Naples is a monolithic die when it's not.

That was my whole point and you're trying to take it somewhere that I didn't (and still don't) want to go...

I don't care if any company (which I view simply as a vendor) appreciates it or not: the appreciation I get is simply from telling the truth even when that happens to not be the popular opinion at the time...

Take care.

 

ajc9988, I think you've gone a little left field of the point I was making (please see my reply to TANWare)... not disagreeing with anything you've said below except the last paragraph.

It's actually kind of funny to see how defensive people get with their favorite ('anythings')... and yeah; of course I'm guilty of it too.

But my point is not something that can be misinterpreted. Naples simply is not a monolithic die. Period.

We can argue about the semantics of the words used to describe it's construction - and that is not what I want - but you can't argue how it's actually put together; 4 discrete dies on a single package (with infinitely smart interconnects: yeah; a pun for you).

That is all my post was trying to highlight. No FUD, no faux pas, nothing else. Just that single fact. sigh...

 

No one anywhere, ever, that I am aware of claimed a Ryzen anything is a monolithic die. So anyone using this is just an invalid argument for anything. Now we have not seen mobile or reaserched it so TBH I am not sure yet if that is monolithic or also CCX's as well.

 

We're not talking about Ryzen; Naples is multiple Ryzen (based) dies on a single package.

 

Even with CCX or not (I'm betting with), the 8 core die wouldn't be considered monolithic. It is getting into the mid and high core counts (12+) that it becomes one die so large it really goes beyond the easier construction. Six and eight core counts cannot be argued in this category because AMD did it on Ryzen. So, the critique is to attack epyc and TR.

Once again, fine to differentiate, not to lie or poo on something that obviously is not how it is described by the competitor.

Sent from my SM-G900P using Tapatalk

 

Did you read my comment on my switch from buying Intel $1000 chip to AMD this round. Review that for the favorite part.

Sent from my SM-G900P using Tapatalk

 

No time to track that. Link?

 

http://forum.notebookreview.com/index.php?posts/10565373

Sent from my SM-G900P using Tapatalk

 

The point is Zen or Ryzen does not simply just glue like it did with Bull Dozer and again Intel did with Pentium D dual and C2Q. Now wording such as "using IF to electronically stitch multiple CCX complex's together" No one could have argued it.

On that note too I would like to see Epyc clocked a bit higher as it seems for now well below TDP on power usage and since OC is not allowed for base memory speeds to get a speed increase. The primary issue I see though is some of those really high density DIMM's are slow. This could be an issue for Epyc compared to Intel.

 

The speed issue is why some reviewers are waiting for bios updates for better 2667 compatibility. But it is a great point to state you need 2400+ ram to get the best out of the product, which adds cost as 2667 is only now supported and will charge a premium on high densities.

Sent from my SM-G900P using Tapatalk

 

CPU monkey has adjusted to the new skews for 1920x and 1950x.
http://www.cpu-monkey.com/site.php?...annels=&memory_ecc=&pcie_version=&pcie_lanes=

 

naw man u know i can't. i need something thats more than 4.6-4.7ghz. also from those review IPC isn't a problem, its the new cache or mesh design losing performance but i have no idea if it'll boost in the area i use or i'd lose performance in the area i use.

I can always check out turbo boost 3.0 since no review seem to want to do an extensive test about it and this is the hardest part. 1st time building desktop, with so many things i wish to know like VROC/TBM3.0 i'd need a perfect mobo.. the enthusiast life test and spend own money.

@tilleroftheearth @TANWare
hey peeps, tiller's got a point, though i dont want to bring too much hate onto both side let me just settle with this. i recall clearly back then when intel first made some CPU skus, they'd have chips like 2500k and 2600k being of the same design, ie just have a better silicon chips named as 2600k and the rest of 2600k disable HT with microcode then becomes 2500k.

IIRC this is a way to save money really just smart business but back in the day i spread hate about intel being cheap and using cheap tactic to make more, same with what they did with consumer HEDT which is crappier xeons. fast forward to today AMD bringing idea of availability using multiple die and nobody is calling them cheap, you guys see the difference?

tho they aren't exactly same i'd say both sides are doing similar things, just how it is.

 

also i just found out theres no E-ATX boards yet.. wtf.

when E-ATX board comes out theres got to be way more surface area to put stuff on, that might help solve VRM cooling area over regular ATX.

so far found a board that does 8x8x8x4, MSI not asus though, with onboard ac/bt/audio. just need onboard VGA too, https://www.newegg.com/Product/Product.aspx?Item=N82E16813144053

 

You suggested waiting for the 18 core previously, which is closer to 4.0 all cores. It is almost always 100-200mhz less with 2 additional cores. I don't know if it acts as a straight line or exponential so that it has a sharper drop off with additional cores (we are in uncharted territory). But, only if having 2 core boost is needed, and that boost can continue as more than 2 cores is used, will you see it be of benefit. If you lock it to two cores by affinity, will you have a dual core, effectively, which gets beat by a quad. The all core has more effect, and the 10 is tops, with reductions on larger. This is also why I put the 12 core in a sweet spot of having the cores needed and the speed. Above that, the speed reductions will get it so that the multi core performance of TR-1950X wins until you switch to having to have higher single IPC with the larger core count.

And, no, Intel didn't, nor did AMD. Unlike prior attempts, AMD redesigned hypertransport for their chips. Because of that, there is no analogy to be made. That is the problem with Intel saying glued, or trying to say cheaped out. They redesigned an entire standard!

Sent from my SM-G900P using Tapatalk

 

You have that wrong, even back then with C2Q, which I have in my P79, no one was calling Intel being cheap. The problem lied, as in all of them, the two dies were running as independent CPU's essentially just sharing memory bandwidth. Without scheduler tweaks there was a 10% or greater performance hit along with power and temp issues.

Ryzen and IF have changed the course of using multiple dies. Nvidia did not look at the C2Q and say "wow, lets do that!". The fact of skewing dies is fine, I just worry the 18 core variants just will not get that high of a clock.

So if you need 5 GHz as a base you may need to forget about any HEDT for a while. Then again once processes is to where HEDT can yeild 5.0 GHz the quad cores may be close to running 6.0 GHz. At least for now you have to decide you are in need of a core hungry HEDT monger, or you need a zippy quad core fighter, in the end to each their own as nothing wrong with either.

 

thats the thing, no one testing them only way to find out is to test it myself. thats why im getting a 8c-10c and 18c will follow through if im happy with what TBM3.0 can do on 2 cores. 8c is the lowest sku to have TBM3.0

silicon lottery got plenty of 5ghz 6c/8c, tbh i'd happy with just 4.8/4.9 it will be so much faster than my ivy at 4.4ghz, around 30-35% performance boost thats only from the CPU, along with 2-4 more cores if i opt for 8/10c.

 

Have you tried writing to reviewers on their corporate email asking for it to be reviewed. Could save you money in the long run.

As to ivy, I'm going from a 6700K running 4.8 to the TR@4. Why? Because I need more cores for my purposes and 4GHz will be fine for the less threaded tasks, especially since we are seeing continued Multi threaded optimizations that even make the FX 8000 & 9000 chips beat SB chips. Getting something like the 18 core, expect 4.0-4.3 all core, if not less. The IPC benefit will be very little compared to TR, with double the price (you could pick up a 2P board and 2Epyc chips for that). In fact, you could get the 1950X, the most expensive board, and trident Z 4000+ x2x16GB kits for the price of the 18-core. I'm seeing so little benefit above 12-core.

Sent from my SM-G900P using Tapatalk

 

When I said quad core I was being a bit facetious as we know better than a quad can do 5.0 GHz. I was just setting the extremes to look into/at.

 

yep i have. for turbo boost 3.0 they said maybe in the future, which means a no for now. as for VROC, i know a few individual has gotten full height add in cards that can hold 4 x M.2 PCIE SSDs but those are not what I wish to find out. I wanted to know if optane SSDs will work without the add in cards, or any intel PCIE SSDs for the matter, also the performance of those add in cards.

what they've told me is intel is still working with oem so its best not to get x299 right now until its all reviewed. i might be forced to buy a 10c cpu due to a mobo that has VROC might not support the 28 lanes split i wanted which is 8x8x8x4..

all in all im still waiting LOL. and hope i can get a 5ghz from silicon lottery and i can simply downclock it by 200mhz for best temp.

 

FYI. You may have seen this before... But Siliconlottery have now changed from Coollaboratory Liquid Ultra over to Grizzly Conductonaut for their delidding service. Damn, Intel screwed us all with the move from solder to thermal paste. And they know the chips run hot

 

lol yeah, they said the new liquid metal is better than CLU, well in terms of material and mwk w/e that unit value is for transfering heat.

my only concern is how long that new material lasts.

 

I have used both. Grizzly will last minimum same as Coollaboratory. And Liquid metal under the lid lasts longer than between lid and the heatsink.

 

its so hard to get a x299 laptop damn so much wait, so much tears, so little hope

 

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/13

Also wanted to point out, Ryzen had around 130ns between CCXs with 2400 ram. Epic, with the same speed ram, has 89-95 and ram latency was at 96-98. Intel's ram latency was 89-91ns but used 266mhz faster ram. This means TR will likely react very well to fast ram (if you can get it to run at the much faster speed, which may also be improved considering the large improvement in latency).

Sent from my SM-G900P using Tapatalk

 

AMD Threadripper 1920x vs i9-7900x cinebench R15 benchmark

 

I do not really care what they get per 100 MHz for the score after all it is a 12 core VS. 10 core. the other data though is relevant.

 

Given the cost of the singular object, the CPU - ignoring cores - the value as the performance it has per 100 mhz is interesting.

One of the touted advantages of the i9 is the higher frequency operating range, which makes up for having less cores. This measurement shows that isn't true.

 

I did a pre-lim analysis on it already, which it shows about 400 points per 2 cores for both Intel and AMD, with a slight edge to Intel. I also discussed that the overclocks on the 10 core reaching the same score, slightly better than, the 12 core achieved. Now, we don't know the score achievable by the 12-core OC, which could be between 2580 to 2800, spitballing on scaling. But that is an unknown. The OC 10 core will give better single-thread performance because of this.

To @hmscott 's point, there also is the less efficient multicore scaling on Intel, which is why it needs more MHz to achieve similar scores. It has better IPC, and receives an edge after balancing those two aspects, but has to be noted that about 200-300MHz overclock is needed, relative to the clocks on a Zen based processor, for the multi-threaded performance, at least in regards to Cinebench. This varies by task, so may not be applicable in other scenarios. So, single thread, Intel has it, multi-thread, especially when examining the 16-core, AMD has it.

The way I see it, between the 1920X and the 7900X, many will be served better with the Intel chip until the software industry switches over more programs for n-core scaling. But, 7900X vs 1950X, the equation shifts to the AMD chip for the same price with WAY more multi-threaded performance at the same price, unless you MUST have the higher single-thread performance. This mostly holds true with the 12-core chip as well, although we can expect around 2500-2550 on CB15 score base, with overclocked achieving as high as 2900 (probably in the 2800-2900 range, more realistically, unless the heat limits the speed more than 100-200MHz, and then I would revise that lower). That is 100-200 points lower than the 16-core at stock, plus gives the benefit to single threaded apps. After OC on AMD, I'd still say 3200-3340 on the 1950X, which would shift all who need SMT back to AMD. Due to heat, I'm predicting a decline in clock speed on Intel, which will hit the multithread inefficiency of Intel, thereby making it harder to justify the 14 and 16 core chips over AMD. Once you hit the 18-core, you are only buying this on the basis of multithreading as all single thread benefit of Intel is destroyed. But, it will be faster than the 16-core AMD, just at double the cost. So, with known information, that is why I recommend ONLY either the 12-core Intel chip or the 16-core 1950X. Those are the stars on performance on HEDT this round, with all others making much less sense unless you have A SPECIFIC NEED FOR A SPECIFIC FEATURE!

 

As far as the total score, yes it did make it up being OC'd. Not that it matters as other than gaming it looses else where. Now about the only real question is what are these x399 boards going to run cost wise?

Edit; I am a bit more skeptical for the 18 core troucing the TR than I was before. Unless x299 tames the power concerns, the chips getter better than the paste cooling they now have and we look at higher TDP's to get those clocks up, I just do not know. I am not sure a 3.5 GHz 18 core could actually take on the TR, at least in R15. Also TR 1950x OC's and with ddr4 3200 plus, well?

Edit 2; Pure Speculation but the 1950x by CPU monkey list all core turbo as 3.5 GHz. if this is true and 3.5 GHz= 3062 then an OC of 4.0 would be 3500 not including faster ram too as this was only 2666 memory.

 

I've been wondering that myself! Now, going off of Ryzen 7, we saw an increase in board coast compared to prior sockets, but it was still lower than equivalent Intel boards (hard to eval equivalent as one has TB3 and optane, but we can get close enough to see a slightly lower price on AMD boards). But then comes laying the extra PCIe lanes and a couple other features. I'd say Intel minus a small amount, but not a lot. Expect to pay between $300-500 per board for the higher-end feature sets. Unfortunately, we have only seen three boards to date on it, although siggraph will likely be the reveal on the board lineups (as well as more demos, including, potentially, another LN2 demo).

 

You do have a point. I'm going off the assumption that the all core was only 100MHz slower than the max boost on a single core, as seen on the Ryzen 7 series (So 3.8GHz). If it was 3.5, even a 3.8GHz would overcome my stated estimate.

On the power consumption, I have a feeling you are right, but am assuming custom water cooling for the 18-core, including either a VRM cold plate or a full cover block. But you definitely have a point. We know AW is doing 2933 ram, but custom clocked in custom built should be able to achieve higher than that, like the 3200 you mentioned. Who knows above that at the moment.

 

[Rumor] Threadripper has Watercooling as a stock cooler - Hermitage Akihabara, Japan
https://www.reddit.com/r/Amd/comments/6nprym/threadripper_has_watercooling_as_a_stock_cooler/

 

But that was done with the 9000 fx series and is more a sign that to run as constructed with decent temps, an AIO is needed. It is to make people use other than air to cool it, which it is an 180W chip...

Sent from my SM-G900P using Tapatalk

 

I'm calling bull. They had a similar thing with Ryzen 7. They showed 5GHz on the marketing material, hype happened, some were let down. So, no, I don't believe that.

I'm trying to prevent it from being over-hyped. 7nm, that is a different story. Here, most is 4.3 if the stepping change added clock ability.

Sent from my SM-G900P using Tapatalk

 

We don't know anything about stepping 2, or the cooling, or the effect of the large cooling area of the new package.

Given Ryzen's lack of OCability, I'd be happily surprised to see 4.3ghz too...but our experience is with Stepping 1 so far, Stepping 2 is all new

 

They are basically the same cores, just more, so it should not be a surprise. Everyone expecting more would whine later for too much hype. If the rumors for AIO are true - kudos to AMD once again for supplying proper cooling solution.

 

See:
http://www.tomshardware.com/reviews/intel-amd-die-fabric-slides,5125.html

Like I mentioned a few pages back and TANWare in particular took exception to; Intel wasn't simply bashing AMD with the 'glued together' slide. Out of context; you can make anyone look bad.

When the whole article is read and understood for what it is; Intel was merely sizing up it's competition with the information it had at that time. Comparing it to what it had to offer to it's customers. Of course there is an Intel twist to it - wouldn't expect otherwise - but it is not a simple slam or faux pas as was suggested by some here against AMD. There was (internal) research and testing behind it too.

Pretty accurately (and fairly) too, how Intel sized up it's competition at such an early stage (with AMD giving info in a piece meal fashion at the time) considering it didn't have actual Naples silicon to play with.

I think the article from Tomshardware allows the emphasis to be put back on actual performance differences between the companies (hopefully in tests with your actual workflows/workloads and not just BM 'scores' that may or may not be modified and/or twisted to fit marketing's or management's hopes).

Still waiting to see more real world usage info and comparisons come forth from the AMD/Epyc camp (like many others surely are too).

 

The article from Tom's is BS. The WAN show (LTT) already showed the surrounding slides for context and talked about it being 30+ slides, some being comparison, and many gross mischaracterizations. Intel pooed the bed on this one. We've all done it at some point. The question is why are you bringing it back up when we would otherwise just move on. It will not be rehabed. Many didn't cover the full context. Then you have most that did saying Intel messed up. Then you have Intel apologists.

Sent from my SM-G900P using Tapatalk

 

I must say this. Mudslinging is a sign of weakness and if Intel were truly confident about its wares, they would let the products do the talking. Take it from me. I am heavily invested, both personally and professionally, into Intel's ecosystem and you could say I am a fan of theirs.

Although let's call it what it is, a knee jerk reaction to both Epyc and TR. AMD certainly got Intel's attention. Come this time in August, I will have my first AMD CPU based build in several years and that was due to one flaw...

 

ajc9988 and Rage Set,

I can't believe your responses and more so, I can't believe you read the article fully either.

No problem, the truth is out there somewhere - for each of us. At least both sides have representation now on this thread (and not just merely my 'biased' opinion).

 

First, for you to infer I am on the side of any one company is laughable. I've consumed various multimedia coverage of both brands and their wares, the pros and cons of each platform. I have come to my own conclusions for my personal computing, however, I still have testing to do with regards to business.

Professionally, I will have time to evaluate both Intel's and AMD's offerings. Not through media coverage but in actual test beds and use cases. Ultimately, what my company offers will be based off my clients needs and budgets.

On a personal level, I have built many rigs with Intel procs for friends, family and myself. In the past I refused to even acknowledge AMD's products. Nevertheless, Intel's new HEDT platform is not appealing to me. There are no excuses for Intel to use paste on their "unlocked" HEDT products that will retail upwards to $2,000. You may not overclock your procs but a great segment of the HEDT demographic does and the reported temps, even at stock, are downright terrible.

If you want to put on blinders and claim anyone that doesn't agree with your assertions as representation of the other side, that's your prerogative. I say the same for anyone that solely for AMD. At the end of the day, I will vote with my wallet as you will. At least for the next year, AMD's procs will be in my personal computing stack.

 

I just ignore anything accusing me of bias at this point. I gave my reasons pushed back, explaining my switch while giving recommendations on both sides. Also, calling a company out for a bad presentation is nothing new. Even though we all are putting it behind us, the comments with the slide deck are not what changed my opinion on anything. It just was bad taste. My comment on how others will interpret Intel as seeing AMD as a threat and might test it after that is looking at psychology and how people react to info. Overall, that is why I'm unconcerned. I've got no team, but do have reason to celebrate AMD's return. 1950X shall be mine.

Sent from my SM-G900P using Tapatalk

 

iirc software have long been optimized on intel's platform so intel has huge advantage here in terms of performance. AMD has to bring more cores, which they did, also better efficiency which they also did very well as well, now all it really comes down to is the performance in actual workload.

there has been some early benches i can't find where now but HPC is terrible for AMD and that covers a huge area in server side. but theres also a factor to consider with the new cache/mesh design i wonder how that will fair with intel.

 

Are you talking Epyc, where Anand had the current review that shows good performance, or TR, where the GB4 results look like a single Ryzen 7 and the chipset reads as an x370? Or are you talking the confirmed TR benches on CB15? Still waiting for the STH review on epyc.

Sent from my SM-G900P using Tapatalk

 

thats right, im talking anad's latest article and pointed out the good power efficiency and decent performance. however server side of things are more than just anand's benchmark.

 

Which is why that will shake out next year on server sales...

Sent from my SM-G900P using Tapatalk

 

I did nothing of the kind (the 'inferring' you state). I mean (only) what I said.

Furthermore, your responses (below) I'm actually agreeing with - and what I expected as a response, btw...

I am the opposite of 'putting on blinders'. Open eyes and mind will always get me (and/or anyone else) the best outcome for the long term.

To me; it doesn't matter which companies platform anyone goes with - what matters more are the reasons.

And if the only reason to ditch Intel was because of some internal slideshow that tried to summarize what they knew about the competition (at the time when AMD had posted only few details...); that is a dumb reason.

Not inferring you're in that category (or anyone else...) - but the mass response on the 'net of Intel 'stooping' so low is below kindergarden level, imo. That is not the real issue at all. Except for online rags that need to feed on clickbait 'reporting' methods.

 

No one accused anyone of bias. But your response below is a 180 from what you originally replied.

From your original response: yeah Intel did get some things wrong (they were guessing and they said so). But they also got many things right basing what Naples/Epyc would offer from a Ryzen die...

Of course Intel is seeing AMD as a threat - but that is the business they're in. For a very long time now.

Just like your analysis' of certain aspects of a platform change shape and form as new info is available; Intel was doing the same back then too. No 'slamming', no 'faux pas' at all; just business running as usual.

I'd really like to see their current slide decks...

When I turn a critical eye on my operations vs. my competitors... I don't hold back on the words I use - for them or for me - that's a waste of my time. The semantics isn't the important part here. What is important is truly identifying (and that will be proven or not later... in the near/far future...) my strengths and weaknesses vs. others in my field. And then laying down the ground work to minimize the weaknesses and bolstering the strengths.

I look forward to you getting the 1950X - and telling us all about it too!

 

'actual workload'.

That is all that matters (ever).


Well, that and reliability, stability, dependability and availability (of the platform...).

 

This isn't a change. It comports with my prior statements. Bad taste comments are faux pas. But, other than making more look at AMD, it isn't anything. It comes with everything else, such as the perception of the market leader in the eyes of consumers. But presentations, other than getting certain details, don't sway my decision. Benchmarks have more weight, but are still limited. As I've said, and others have, it is enough to do a test bed.

Believe it or not, you get more tempered responses from me when I don't have to defend attacks on unfounded ground.

Also, this is not a simple business as usual situation for Intel. They are looking at losing 20% of CPU market share inside the server industry, which, as a whole, is $16B. Losing billions hurts, is a threat, etc. What people took issue with is Intel started by ignoring, then pushing their performance. That is what we expect. Including your own comparison is nothing new. What shifts is when you have lies about partners on a slide, misrepresent information by calling it glued, etc. But, reactions and talking about it is just that, reactions. The sooner people let it die, the sooner we can move on.

Now, availability of other Ryzen chips congress in a week or two. More will be known then. Intel hasn't shown up on bringing competing publicly, but reports of strong arm tactics to lock people in on deep discounts months ago were reported. Who knows if true, and could be playing on Intel's M.O.

Overall, my prior statements agree with where I stand. Intel has good products. Now, so does AMD. Also, since optimizations started, we see improvements on the FX series, proving software optimization was lacking. That is not all on AMD. That is on developers as well, and shows using CMT, if it had been supported, AMD would have been better off. By using SMT, it does enough like Intel and performs well enough that they are finally optimizing. Because of that, the ball game changed.

So don't try flipping my words. They are there for a reason. Use the new words to interpret, not replace. At most you can argue tempered prior statements. But no flip occurred.

Sent from my SM-G900P using Tapatalk