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

Sounds like the logic of desperation, and throwing away more money to "save money"

The cost doubles from $600 for the 7820x to $1200 for the 7920x, $600 more is double as much for 50% more cores, from 8 cores to 12 cores, that's a bad deal.

You could get 2x 7820x for $1200, and have 16 cores.

Or, spend $850, save $350, and get a Ryzen ThreadRipper with 16 cores. and 64 PCIe lanes.

If you compare the Intel 7960x, the Intel 16 core sku, vs AMD ThreadRipper 16 core, you would pay $1700 Intel vs $850 AMD, you'd again pay 2x as much!! You'd lose $850!!

You could get 2x AMD Ryzen 16 core ThreadRipper = 32 Cores for the price of 1x Intel 7960x 16 core!!

Wow, Intel is screwed

 

imo your arguement is a bit biased. it is the price to pay for some of the higher end, latest cpu offering from intel. if people truly wanted core and best value, they could easily go for haswell xeons thats only 6% lower IPC than skylake and pay for anything in between 18-24 cores for way less, especially when both ryzen and intel offers cpu greater than 14-16 cores, they cant overclock much past 4ghz due to heat anyway, cheap xeons would be a nice option.

the reason to go for intel would be blatantly obvious, brag rights and also have the fastest single threaded IPC performance which would benefit a lot of legacy software out there, which we still use daily btw.

ryzen on the other hand is an excellent value but there are things thats lacking, single threaded performance, no turbo boost 3.0, poorer storage performance those are taken into consideration as well. according to new x299 looks like i can raid two intel ssds in raid0 without having to go through chipset, that would free up additional SATA ports, really all comes down to how you use the features. for me intel benefits me more, gotta be at least 12c or higher.

 

You would need to get 2x the performance throughput on your Intel 7960x to match the value of the AMD ThreadRipper 16 core.

If Intel is 10% faster than AMD, or 20% faster, or even 50% faster performance - which of course is very unlikely - it wouldn't be fast enough.

Intel 7960x would need to be 100% faster in throughput than ThreadRipper 16 core to be worth it's price, 2x faster than AMD to break even, just to catch up to AMD.

You could put 2 ThreadRipper 16 Core CPU's to work for the same price as the Intel 7960x 16 Core CPU, and get the same work done for the same cost.

Intel is hurting very badly, there's no way Intel is going to be able to match AMD's performance per dollar, no way.

 

Intel will never be the value of AMD offerings with Ryzen. That is not the point. Intel, for now, has the fastest platform. So if you need the fastest, Intel is the way to go for now. This may change with the release of TR, we shall see. AMD being quiet is just handing it over to Intel.

 

That's completely ridiculous, there's no way I'm spending an extra $1000+ in build cost for Intel 16 core because it has a few percent performance advantage over AMD 16 core.

You'd have to be a consummate Intel fanboy to buy an Intel 16 core at a 2x price difference compared to an AMD 16 core CPU, even if it's measurably faster.

Measurably faster isn't an automatic win for Intel at all, especially when it costs 2x as much.

Unless Intel is faster in a truly beneficial way compared to AMD performance, in a totally "worth 2x the cost" way, which is impossible.

 

To be honest, Intel's 16ore/32th CPU could end up being somewhat faster than AMD Threadripper... however, I sincerely doubt that Intel would be able to get as much as 50% increase in multicore performance even with 18 cores/36threads (they might get up to 40% due to higher clock speeds though).

I can see Intel having perhaps a few hundred Mhz advantage at turbo (if not operating on same frequencies at higher core count), but I doubt even they can bring it up to 4.5Ghz on 16c/32th... let alone on 18c/36th (and keep the speed there) and keep it at 165W.
We'll have to see what Intel releases.

Also, there's multi-core scaling. AMD might be at an advantage here because of Infinity Fabric which allows near perfect scaling.
How do Intel cores scale?
In 3d studio Max for example, 6900k is about 9% faster than 1800x (and those results are before all the BIOS/microcode updates for Ryzen - which in various games closed the gap with 6900k to a mere 5% - but no one apparently bothered to retest 3d Studio Max for example) and 6900k costs DOUBLE.

Even if Intel is faster by 40% (and that's a big estimate considering than we don't know the clock rates of their 16c/32th (which might be 15% to 20% faster) and 18c/36th cpu's or how their multi-core performance will scale.

Does anyone really think that spending twice as much money for potentially less than 50% increase in performance is justifiable?
Also, what about Ryzen refresh at 14nm+ which is likely going to consume even less power, and might also be clocked higher by default (much closer to Intel) ?

I don't think Intel's prices justify the purchase. Even businesses are in the cost efficiency game. Do you think they are more likely to spend $2000 on Intel for 20%-40% increase in performance, when they can get 2 TR systems (16c each) at the same price with near perfect scaling?

 

I did not say it was the smart choice, it is the fastest. Not that it is measurably fast enough over AMD to justify the cost, it just is slightly faster. No fanboy about it, if both are available for the wanting there is no reason not to choose AMD.

Edit; the problem is today I can pre-order a 10 Core Intel but only up to a Ryzen7. I can't justify waiting as there is no info as to release dates and skews and benchmarks to compare. So for today Intel is the overall winner. This may change 8/10 if all skews are released until Intel releases theirs. That is almost two months for x299 to mature and saturate the market.

 

Are people really this impatient?
What's going to happen if you were to at least wait and see how TR performs?
Is civilization in impending danger? Well, it is from climate change which is driven by a ridiculous ideology of infinite growth on a finite planet (Capitalism and outdated monetary based economy), animal agriculture which contributes more to climate change than all planetary transportation combined, followed by unsustainable and ridiculously outdated methods of production and energy generation (and of course enormous waste on land and in oceans which could be harvested for raw materials and returned to Earth to repair the damage we caused as well for production of goods and services instead of mining fresh resources which takes more energy, time and money).
But that's a different topic.

No, seriously, what exactly would a person 'lose' to just wait a bit longer for TR release and see how it compares before they arrive at a decision which system to invest in?

Intel has more resources at their disposal, that's why they can churn out a product a bit faster than AMD.
And there have been minor talks that AMD might be giving motherboard manufacturers time to adjust and avoid shortages like what happened with Ryzen.

If multicore scaling is taken into account between 1600x and 1800x (both of which operate on same frequencies and turbo speeds), the 1800x is between 25% and 30% faster.
The 10core option from AMD will likely do the same and be about 5%-10% behind 7900x if current benchmark results are anything to go by (of course, this is just an estimate based on what is out... and we don't know how TR will actually perform).

Oh well, its not my money or decision.
Its yours.
I just don't get why people would still opt to go for Intel when their products are far more expensive while marginally better, even if they are available a bit sooner.

 

All the 'cost analysis' in the posts above are missing some major points. TANWare gets it: Intel is faster right now. IF you need HCCP right now; Intel is the only option, regardless of how the value may seem to drop once TR ships or is actually available.

First off; a CPU is not making a system 'double the cost' if/when maximum performance is needed and the rest of the platform around it is considered too.

Second; that 'rest of the platform' for me can be upwards of $4K to $6K - there are many, many others like me that don't really see a 'savings' of even $1K on the CPU as relevant in the long term (i.e. the lifecycle of the platform in question).

Third; when a platform offers anything greater than 5% improvement in actual productivity - and that system is used minimum X hrs / day to make $XXX with - then, buying anything that saves you a one time cost of even $1K is a waste of money.

Fourth; as many are finding out more and more: in a strict gaming scenario (today) - the highest level of performance attainable is not needed for 'good enough' game play. If you do more than game with a single system; then go ahead and spend more than $xxx on your CPU purchase.

Fifth; when maximum performance is required: everyone knows expecting that performance to scale with 'value' is an infantile idea. The peeps buying Ferrari's, Lambo's, etc. to race them aren't looking at value; they're looking to 'win' over the next guy who has a slight disadvantage in hardware... even if that 'win' is just 0.001 seconds better.

Sixth; to justify a purchase on paper is doing everyone, including yourself, a discredit. Buy to test - not to brag about spec's. If the testing proves a platform worthy (regardless of what IPC, marketing and fanboys may indicate...) - it is worthy (to you). There is nothing wrong with spending $$$$ when you have it. Just make sure you know what it is being spent for...

Seventh; for a platform to be worthy - it needs to be reliable and dependable (from day one). Intel has proven itself for over a decade now in this regard and at the highest performance levels that have been possible up until this time - this is worth extra $$$ in and of itself. A platform with high productivity and low reliability/dependability is worse than a much worse performing option - when $$$/Hr are involved; a single hour/day/week of downtime offsets even $1K CPU differences very, very quickly...

Eighth; I agree that if you need 8 or 10 or 'xx' cores today at 'X' performance; buy a CPU with at least a little more for room to grow with. Tech workflows and workloads (O/S + Programs) don't stay the same over the lifecycle of your purchase - if you buy what 'fits' today; you may find yourself looking foolish in as little as 3 to 6 months down the road (when if you had spent even up to 20% more over the entire platform; you would have 2 to 10 times the useful life out of your purchase).

Ninth; don't be stuck on one metric or 'score' when evaluating a platform. Your workflow is king (to you). Ignore this point at your risk.

Tenth; Don't fall into the trap of sunk costs. Once you've parted with your $$$$, stop. Everything you do now is independent of that past action. Act accordingly.


The most important idea here? Don't make decisions on products that haven't shipped yet with performance/productivity that can't be confirmed! (Yeah; the above points allude to that - but it still needs to be made perfectly clear).


What I am observing lately is that 'gaming' isn't the segment of the market that is pushing tech anymore (or; at least not to the degree it once was). The tech available has far surpassed gaming requirements for a very long time. Will systems (O/S' and programs) need more than 8C/16T in the future? Of course. But the reality today is for most consumers; the answer is still no.

Regardless of what marketing (on either side; Intel or AMD) may tell us.

The best bang for the buck is almost always the lowest performance part available - for a given platform. The highest performance part is worth whatever manufacturers can actually charge - even for a 1% improvement...

What it is worth to you can only be decided by your testing alone. Just make sure your reasoning is logical after that testing is complete - you'll only fool yourself if it isn't.

 

Ant the guy Talha Amjad in segmentnext.com didn't mention that next mainsstream Core i7 chips will have 6 cores. Buying 4 core's processors now or a bit later is pure stupid. Same if you later go for old tech. Only a good advice if you shall keep your old rig. Not for your next computer. 4 core chips is OLD!! 6 or more is the NEW. He also does not mention streaming that is becoming more and more popular. No wonder people are doing stupid purchases.

"While having additional cores is great, there aren’t many games out there that take advantage of more than 4 cores and 8 threads so paying the extra money for more cores would not make sense seeing how the Intel processors cost more than $1000. You would be better off buying the mainstream Ryzen 5 or Kaby Lake CPUs."

Btw.
Intel Core-X Series ‘Skylake-X’ and ‘Kaby Lake-X’ CPUs Overclocking Detailed – Core i7-7740X Can Easily Hit 5 GHz, Liquid Cooling Recommended For All Chips-Wccftech.com

 

You're wrong on storage. With 64 lanes, I've seen nothing requiring amd to go through the chipset, nor do you need a raid key for better raid setups. They are making raid cards to hold 4 nvme m.2. Just had to correct the storage crap you said. Intel has Optane. That is all that can be afraid currently, as well as you needing a raid key for Intel.

Sent from my SM-G900P using Tapatalk

 

Every two cores, reduce Intel's top speed by 100-200mhz. Considering 4.6 seems common and right for the chip, you are likely to get 4.3 tops on Intel's 16 core, maybe even less. That means the speed advantage dissipated with the core count.

Sent from my SM-G900P using Tapatalk

 

Yeah, Intel have Optane, but we will may see Samsung's Z-SSD technology meet up.

 

Your edit is false. True pre-sale starts tomorrow, sale on the 26th for Intel. There is information, just not what you want as proof. Also, Epyc is released on the 20th and will give insight. So, saying no use in waiting is actually contrary, considering the X299 firmware has been proven under developed, you'll have growing pains, the 10 core isn't worth it if you truly need the core count, the 12 core is Intel's value lineup (speed per core, plus number of cores).

Also, when weighing cascade lake-X versus 7nm, we look at the performance gap being closed, AMD having higher logic density as the 7nm is expected to be just under Intel's 10nm logic density, but cascade will use 14nm++, meaning they are favoring IPC. But, if Intel delivers 10-20% (on its promised 30%IPC jump, as they always over promise), and AMD gets the expected 40% from the node shrink, potentially, AMD is positioned well next year.

Sent from my SM-G900P using Tapatalk

 

Ok, a day off. I meant it more as pre-orders are essentially here. I am not saying the 12 core is more desirable I am saying the 10 core is faster than Ryzen 7 1800x and Intel will then have the faster solution for 2 months. Epyc might give a better idea but not the actual numbers for TR.

I am sure x299 and x399 will have growing pains. By your analysis I should wait 2 months for x399 then 2 months for growing pains to settle a bit?

The market for PC's just does not stop and wait for no reason and AMD is not giving one. Again official information/benchmarks with firm dates and costs. Blame people not waiting on AMD themselves.

 

Yes, but the reason I'm going for AMD (I'll admit it) is Intel FUBARing up their release schedule moving forward. They have no clear path. They are doing Cascade-SP for Xeons next year (14nm++ to give extra IPC, because 10nm is lower IPC than 14nm++ by a good ways), then they say Cannonlake-X (10nm) for HEDT next year, then Coffeelake (14nm++) for mainstream and cannonlake (10nm) for mainstream and mobile? Meanwhile, Ryzen+ (14nm+) will be mainstream and pro, Zen 2 (7nm) for Rome (EPYC successor) and TR successor, then moving everything else to 7nm, then to 7nm+.

But, 10 core is not a great value. If you need the higher core count, then 12 is better, hands down. I'm not saying give TR the same benefit of developing as Intel gets, I'm saying don't buy it, then later sell it in the secondary market to get what you should have in the first place. You get stability off the bat with less to no growing pains. Then, only if TR by AMD hits a grand slam do you change course, but you get what you want/need without the extra hassle or waiting a year to get what you had your eye on this year.

Personally, I just see the 10-core as a stupid decision with the lineup. Benching, fine. Those that don't need the core count but want the 44 lanes, fine. But if you need the cores, it is ridiculous not to get the 12 core.

Edit: I forgot to mention, Ice Lake is taping out now, with supposedly second half of 2018 potential availability. Considering that is for Server and HEDT first this round.... So what are they doing?

 

Agreed that Intel is jumbling everything to keep the edge of the market. This may cause AMD to re-assess the Epyc lineup and cost as now with newer high end Xeon's (platinum etc.) the pricing structure will change. Pricing and competing against old Xeon's is most likely not going to work.

Since Intel is bench marking that 26 core like it is meant for a super high end workstation maybe we need to see Epyc ported to x399, it looks like x390 was a single TR and x399 can handle a dual TR, or converted Epyc?

 

to flat out say im wrong on storage is simply jumping to conclusion LOL, there are more to storage performance than the simple PCIE lanes involved but i dont feel like arguing so i'll leave it be, as tihs isn't a storage discussion.

 

I'm sure AMD is addressing regularly! But, you can't compare if you don't have official numbers, so you go with current market availability. Same is true with Intel comparing the HEDT to Ryzen.

As to TR and EPYC, THEY CAN NEVER BE INTERCHANGED! EPYC has the chipset on the chip, TR has the chipset on the board. That means no chipset would exist if you put a TR into a server board, while you'd have two chipsets trying to control an EPYC chip (although, you could potentially disable one of those to allow in a X399 board, but then you lose 64 PCIe lanes and go from 8 to quad channel memory, a huge downgrade). What I'm hoping is EPYC is unlocked and the 8 and 16 cores are able to be overclocked, if they exist as presented. Then, have the server manufacturers do a gaming/OC board like supermicro does with superO line on X99 and the upcoming X299. Maybe Tyan might even do that. Then, crossfire 6 cards at 16 lanes, leaving 32 lanes for storage, networking, etc. That would be a beast, especially if it hit 4GHz! Just saying...

Sent from my SM-G900P using Tapatalk

 

It's not jumping to conclusions considering storage on a mass scale is where AMD is expected to make the largest inroads in the server market. So, yes, it is more than PCIe lanes, but it also shows you haven't seen the predictions either, nor comments from the execs at Dropbox which are currently testing EPYC for potential deployment. But what do I know...

Sent from my SM-G900P using Tapatalk

 

thats talking about server and their 8 channel memory, on a 2S system versus intel's 2S system.

in terms of value, AMD is the best and always has been, in terms of fastest performance, intel is currently.

when these two mixed together, i put max turbo boost 3.0 quite a bit above the others as thats what i needed most, single threaded IPC performance would benefit me and my legacy software more so thats what i will go for, as long as pricing is within what I call reasonable. almost 2x more expensive than AMD for around 1400 for 14 cores? for me thats reasonable and a trade off i'd have to pay so I can go for it.

what i could get out of it, is single threaded performance at around 20% - 25% faster than AMD if used right with turbo boost 3.0. what I didn't expect to get out of it, is toothpaste under the IHS that part turned me off A LOT and its rather disappointing really.

 

Yes, so a storage based company is choosing a system solely on ram support, nothing to do with storage capacity or ability to perform at that task. That makes sense.


Sent from my SM-G900P using Tapatalk

 

we talking about consumer part though, i certainly wont choose amd over intel with the benefits that comes with intel that its faster in storage and in cpu raw power.

 

Now we're talking + overclocking.

 

Raw power, yes, storage (excluding Optane support), I don't see that as absolute.

There is where I'm having the issue. I'll agree in IPC and higher overclock, question multicore/thread scaling (because of IPC and speed, still best performer, but some evidence shows better scaling on AMD, but Intel still, in the end, having higher performance, which may not hold true on HCC chips as the speed benefit drops on all cores). So, I only agree in part. Also, absolute GHz means nothing, otherwise we'd be using AMD old designs, which is absurd.

Sent from my SM-G900P using Tapatalk

 

+2x " overpayment".

 

The practical person in them knows that if they wait for AMD there might be no practical reason to then buy Intel.

So if they wait they can't fulfill the irrational desires of the fanboy that's bursting out from the seams right now, so they gotta buy now!

 

Worth every single penny. With phase cooling like before, I will have a fun time. Collecting stamps and old coins is also a expensive hobby

 

Collecting stamps and old coins is not only an expensive hobby, if done well it can secure the retirement future for the collector.

Not the same for soon to be irrelevant computer " overpayments" hardware

 

To be fair, HEDT is less about cost/performance in some ways, with performance being the focus.

But, @Papusan, comparing what AMD can do on DICE would be good to terse which system will give the BEST under phase...

Sent from my SM-G900P using Tapatalk

 

well intel does have some edge there, extra 300-400mhz MIGHT be possible with these 14nm++ silicon. if only i can get a cihp of that quality shown in youtube video, i'd go for 8-10c at 5ghz instead of 12c at only 4.2ghz.

nah, imho optane is absolute, since im a storage freak and i want the best 4k random read/write performance available i have to go for optane. besides, intel allows raid via cpu pcie lanes on x299, thats over anything what AMD could offer, as u'll need additional raid card if you're going for x399, and we all know about raid cards.. high sequential, $h!tty 4k random performance

when it comes to storage, intel wins atm hands down.

rofl you just have to addressed the value part

 

This is 14nm+, coffee and cascade-sp for Xeons is 14nm++. So please keep the processes straight.

Also, no information for on-board raid is yet available for X399, so I will not speak to that, nor is it known what EPYC supports with the on-chip chipset. Then, you have what papusan pointed out from Samsung. So, it is not a hands down win until Intel gets better Optane products out the door! Until then, it is a nothing-burger. But that is why I brought up the raid cards, which are also being used ON INTEL CHIPSETS (sold for both) to utilize 16 pcie lanes...

 

https://www.globalfoundries.com/sites/default/files/product-briefs/7lp-product-brief.pdf
I just found the reason I'm going AMD for sure: 7nm@5GHz target!

@Papusan @TANWare @hmscott @Rage Set @triturbo

Edit: it also supports DDR4/5

 

I prefer to think of it as the practical part. The first part one considers before wasting time on the rest...

"Oh, so the Intel CPU costs 2x as much for the same core count as the AMD CPU...Intel is $1000+ more for the whole system including motherboard and full feature licenses?"

"Hey!, look!, an Ice Cream Shoppe!!, let's go over there!!. You wanna have Ice Cream, don't you?"...

 

Hopefully we will get some full benchmarks soon with Epyc. If the benchmarks from CPUMonkey are true then Intels Xeon Platinum could be in trouble even against a TR let alone an Epyc 32c. This may not be true of single core performance but no one is going to but a 16+ core solely based on that.

Optane for the most part is a slow non-volatile ram disk. It is caching reads and writes that eventually have to make it back to the storage device. So long as I/O is not sequential high rate the true storage device catch's up and there is no problem, if it can't then there can be some slowdowns. For casual use it should work wonders for a HDD, for high intensity you may still need to use it in front of a good SSD so the device can keep up. Now Optane as a cache in front of an Optane SSD would be optimal as other than bus issues it should almost always keep up.

 

broadwell is 14nm, skylake is 14nm+, kaby is 14nm++ isnt it? skylake-x should be using the same process as kaby.

and im still not sure why you kept on briing up pcie lanes, though i already explained it only benefits sequential which is not really what im after. and its not like saying intel coulnd't do the same. the people who get 64 pcie lanes on AMD x399 will have other uses rather than just "high" sequential storage lol

 

WRONG! Broadwell and BW-E are 14nm, Skylake is 14nm but different architecture tweaks, Kaby is 14nm+, Skylake-X is 14nm+, Coffee and Cascade-SP are 14nm++! So you are correct that Kaby and SL-X are the same process, but you are wrong on which process it is on!

 

ewwww https://www.notebookcheck.net/Intel-Cannonlake-may-have-to-wait-until-Q2-2018.228582.0.html

we going from Q1 to Q2 now? i guess clevo 6c laptop coming end of next yr. while 8c laptop in AMD prob available Q3 or end of this yr.

 

Cannonlake is weaker IPC than Coffeelake, but has higher logic density. So, you have to make a decision anyways!

 

why would cannonlake has weaker IPC than coffeelake? i think you meant to say the other way around. intel would use the same architecture and then optimize it further. just like broadwell from haswell, broadwell is tiny bit faster with the node shrink + a tiny bit of optimization.

 

Look at the INTEL provided post on IPC. 14nm++ is better than 10nm, meaning coffee is better than cannonlake!

 

Collecting stamps and old coins will not give me any pleasure. More like watching paint dry. I'm +50 and want some fun before I walk into the coffin.

I have said this before. I want numbers before I spend any money.

 

I agree with wanting numbers! I was just saying DICE scores on both platforms (-35--50C) would give you a good comparison of the platforms if using Phase...

Edit: Also, AMD's IMC has a cold bug, in case you hadn't heard IIRC.

 

I'm not very optimistic about the OC capacity for Ryzen. The sampling is limited, but still. I will sit on the fence and watch before... Yoo know.

 

I see your point! I'll take the hit if x399 supports ddr4/5 and just replace the CPU with a 7nm in a year, which should give parity.

This is IBM's 7nm power design, which they've put out over 5ghz chips in the past, which makes me extremely hopeful.

Sent from my SM-G900P using Tapatalk

 

LMAO you cant be serious LOL. if 14nm+ and 14nm++ is that much faster we would expect to see kabylake much faster than skylake, which review shows ipc is at 0-1%

also LMAO at that graph, 22nm so much above 32nm ivy to sandy difference in ipc is a mere 3-4%. seriously man, these graphs are likely showing performance and quality of the silicon, but you believe what you want to believe anyway.

but no, coffeelake will likely have lower IPC than cannonlake at same frequency. however 14nm++ will likely clock much higher than 10nm first gen, assuming all are from intel.

 

looks like we are into desktop this gen. good bye laptop market .

 

Definitely, if all chips were welded on the MB. Yes, then... Game over. Finito laptops.