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

I'd spread some blame around too. Can't really let off board manufacturers for not testing OCs on boards that are meant to be OC'd.

 

They only had 10c/20t CPU'S during development, at best, then came Intel surprise announcement that changes everything - adding up to the 18c/36t CPU load...

 

I edited my post to show 'more than 10 or 20 Cores'.

Still doesn't change anything. TDP only goes to 165W from 140W.

The work the platform can potentially do goes up dramatically; yeah. The TDP requirement is less than ~18%.

Not acceptable in any way shape or form (and certainly not Intel's fault from my point of view). The M/B manufacturers knew where their boards would fail with the 'surprise' announcement in any event. Yet they still delivered them.




See:
http://forum.notebookreview.com/thr...99-xeon-vs-epyc.805695/page-195#post-10628855

 

I fix it for you, bruh Remember TDP is just a number... Not even valid. Aka None of the processors have accurate TDP.

"However the Core i9-7980XE goes above and beyond (and not necessarily in a good way). At full load, running an all-core frequency of 3.4 GHz, we recorded a total package power consumption of 190.36W. This is a 25W increase over the TDP value, or a 15.4% gain. Assuming our singular CPU is ‘representative’, I’d hazard a guess and say that the TDP value of this processor should be nearer 190W, or 205W to be on the safe side. Unfortunately, when Intel started designing the Basin Falls platform, it only was designed to be rated at 165W." Mind you, this is with stock clocks.

 

TDP becomes particularly important when it is limited. They should just make all CPUs with unlimited TDP and unlocked multipliers. Then you only need to decide between buying good versus garbage. Problem(s) solved.

 

At least they should test aka max out all their unlocked processors in all form and shape. They have all the resources to do it properly. If they can't... Hire people who can manage the task!! Not put all the blame on MB manufacturers and overclockers.

Intel washing their hands... As usual


See also How Dell cripple performance

 

You're seeing this from the wrong side of the fence though.

The facts are closer to:
Intel makes a product. TDP=165W at stock clocks.
Enthusiasts want to O/C. Intel allows it for some of it's products. Everyone wants to make $$$$$$$.
M/B makers make a product to support Intel's processor. They fail miserably.

Whose fault is it?

The M/B manufacturers, period. But like I said; they can say 'oops', give people working products (whether it is firmware upgrade, if possible or via a trade-in program - doesn't matter to me). As long as they fix their mistake.

Intel didn't dupe anyone here as far as I can see. They delivered what they promised.

Time for the M/B manufacturers to do the same.


And I have no doubt Intel max-tests all their processors - unlocked or not - but they're not obligated to share every part of their work/sweat with the world.

Finally; I don't see Intel putting the blame on overclockers either though?

 

Up to 205w vs. stated 165w TDP ain't what I can see delivered as promised. See... (and certainly not ONLY Intel's fault from my point of view). What would happen if Intel did the same for thin and flimsy and the OEM's have made finish their JUNK? Aka same for their BGA Junk intended for -1" Jokebooks? How many overheating problems has been brougt up on the web? No, You have to bring in INTEL in the mess as well.

 

From your own post just above:

See:
https://www.anandtech.com/show/11839/intel-core-i9-7980xe-and-core-i9-7960x-review/14

So, AnandTech knows that TDP does not (have to) equal actual power used. Why don't the M/B manufacturers?

It's not so much that I'm sticking up for Intel... the point is that who is responsible for their products (the M/B's) is the M/B manufacturers. After all, I'm pretty sure they just didn't design these on 'specs' and then had them built and shipped without touching the processor they were supposed to be supporting.

Is Intel to blame too? Maybe. I don't get to see the spec's that someone building a M/B might. But like everyone else on the planet; they can change their definitions of what TDP is (as they already have in the past, a few times) as they wish. Their product. Their rules.

In my business, the products I deliver are very specific in nature too ('simple' digital images...) - and I make doubly sure that I deliver exactly (or more than) what my contract says. And still; there are clients that will argue otherwise.

I also don't know how the load to measure the power used was generated. Was this an actual workload or a meaningless synthetic run? I am pretty sure Intel tests actual workloads when finalizing their products - not power viruses (even if the 'specs' are presented as synthetic bm 'scores').

 

I know, but they still claims on 165W And all know what will happen if the MB manufacturers follow Intel's specs as the Devil


Intel know what's power viruses is out there. Just test it!!

 

Lol...

See from the post above:

And voila! They're below 165W (for the cores). The devil is in the details...

 

I really hope Intel don't put finalize specs under such circumstances / tests (If we remove this from the load power numbers)

 

I totally get what you are saying. As a general rule, Intel makes good products that work exactly as intended, or better. Intel CPUs seldom ever die. They are closer to being bulletproof than anything else in the technology world.

Motherboard manufacturers--especially with laptops, which usually suck--run the gamut from always half-assed trash to almost always excellent products, with quality tiers in the middle. We end up with whatever they feel like giving us, whether it turns out right or not. Most of them do not own any of their mistakes. Buyer beware.

It merely exacerbates the problem when end users cut a corner or two to save a few bucks. Then they want someone to blame instead of reflecting on the fact that expecting wine and cheese on a beer and pretzels budget is what got them into a pickle.

 

ah the beauty of first world countries. expecting 5 star quality services while paying pennies.

 

thats ddr speed? or ddr2? so damn fast

 

Where's the link?

The reads are almost spot on DDR2...

 

just some share of few quick test from reviewers lol. honestly though no point really raiding optane, the 4k will drop too much in exchange for barely used sequential performance.

 

There... I fixed that for you.

 

So, not an optimal solution (i.e. you need to spend more, if you want this issue fixed/addressed...). But at least M/B makers are acknowledging that their current products are not up to snuff.

For a $33 premium right now over 'normal' (i.e. defective) boards, this is acceptable for anyone that hasn't bought into this platform yet. I still think that they should replace with these new boards anyone that brings in beta v1 boards back to them though - I'd even agree to pay that $33 difference to do so.

See:
https://www.anandtech.com/show/12007/asrock-releases-xe-edition-of-x299-taichi-and-gaming-i9

 

ASRock Releases X299 Taichi XE and X299 Gaming i9 XE: Big Boy Heatsinks
by Joe Shields on November 9, 2017 11:00 AM EST
https://www.anandtech.com/show/12007/asrock-releases-xe-edition-of-x299-taichi-and-gaming-i9

"This week ASRock has released “XE” editions of two of its popular X299 motherboards: the X299 Taichi XE and the flagship X299 Professional Gaming i9 XE.

The XE versions have a larger main heatsink, and an additional power delivery heatsink connected via heatpipe located behind the back panel VRM. Along with being larger, there are more fins cut out increasing surface area and heat dissipation capabilities. The original(s) had a single, smaller heatsink to cool the 13-phase VRM underneath and not a lot of surface area.

Both boards now also add a second 8-pin EPS 12V connector for additional power to the CPU.

In order to fit the updated cooling solutions, the shroud covering the back panel IO and audio sections have been removed."

ASRock X299 Professional Gaming i9 (before)

ASRock X299 Professional Gaming i9 XE (after)

ASRock X299 Taichi (before)

ASRock X299 Taichi XE (after)

It's good to see these upgraded power and cooling motherboards coming out so quickly

"In the weeks after the Skylake-X and Basin Falls (X299) chipset release, concerns were brought up with the platform’s VRM and cooling. Igor Wallossek from Tom’s Hardware looked into this issue in depth. In the end, it was found that these 140W+ CPUs, such as the i9-7900X, can use around 250W when clocked around 4.5 GHz using all cores (through burn-in testing –a worst case scenario). All that power creates heat, both at the CPU and the VRM level delivering voltage to the FIVR. This was also a problem when motherboard manufacturers were implementing their own turbo modes over the top of Intel specifications.

The concern here there was that some motherboards do not have enough VRM cooling to dissipate the energy lost as heat from the power delivery. At base frequencies, in well ventilated cases, this doesn't appear to be a big issue, however some motherboard partners enable turbo-boosting features such as Multi Core Enhancement by default, raising clocks and voltages, and can be too aggressive with its enhancement. Of course, cooling a CPU using that amount of power isn’t as easy as strapping down a $30 air cooler on it and calling it "OK" either. The point is some motherboards in certain situations can throttle the CPU at the VRM level due to the heatsinks not being able to dissipate such loads."

"Pricing on the XE versions of the motherboards will cost a little bit more to compensate. The Taichi is currently priced at $289 on Newegg while the Taichi XE is $322. The Fatal1ty Gaming i9 is priced at $389 with the Gaming i9 XE at $422. Both boards are asking a $33 premium. If pushing 140W processors or the 165W flagship 7980XE hard is in the plans, one may want to consider the XE versions and pay the premium, if only for peace of mind and cooler operation.

ASUS has a Strix XE as well with a larger heatsink."

 

I feel like this has become worse lately, especially in entertainment and related fields.

 

yee, well on the other hand laptop cpu is doing the reverse. charging premium for mobile bga while giving garbage.

anything that use marketing markup u know something is wrong. similar thing for seagate/wd for their external drives selling cheaper than bare drive, taking full use of the enclosure and its purposely designed firmware on usb 3 controller to limit the read/write speed of USB 3 in order to hide HDD performance defects. in a way its still better than intel, charge more with garbage, seagate/wd charge less with garbage.

 

Yeah, as we posted above

 

I was more referring to unrealistic consumer expectations about what quality they can expect at bare minimum price. (or no price). But you make a good point there, they talk like the product is in a higher class than it is.

 

Nvm

 

Wow, great scores, is that at stock or OC? How are the temp and power draw on the VRM's?

 

VRM is fine. Non AVX on AIDA64 is below 165w.

 

Cool, Thanks!, glad to hear it, please let us know how it OC's and what the power draw is as it ramps up in multi-core OC limits.

 

16 CPU Core Comparison - AMD 1950X Threadripper vs 2x Intel Xeon E5-2670

 

http://hwbot.org/benchmarks
Download the wprime from here! Scores are better with it and more comparable.

 

5 year old platform vs. 2017. Not a good comparison, imo?

 

https://corona-renderer.com/benchmark/27
Compare my score on there versus a dual processor 2680v3.

 

Hot DAMN!!! That is some sexy numbers! Post that over here: http://forum.notebookreview.com/thr...unge-laptop-owners-welcome-too.810490/page-12
Also tag @bloodhawk and @Mr. Fox

 

If you don't have it:
https://hexus.net/gaming/news/pc/111887-watch-dogs-free-pc-gamers-13th-nov/

 

Haha, I am still running a lot of testing at the moment. This is a very complex processor to be running.

Testing higher cache clock right now. Its hard to get 4.2ghz stable and worrying about AVX offsets. I have never overclocked something this complex.

 

Tuning always is! I always leave plenty of time to try to nail it down before I put it into commission, normally. Only reason I'm using it now is it is good enough until I can reseat the block. But, once you go through all the testing a couple times, it starts to become second nature. I mean, it took me a fraction of the time to set the ram up in this build compared to the 6700K build, which was somewhat quicker, somewhat slower (because of more control over timings), than flashing my SPD for my laptop ram. Plus, they slightly change the game on OCing settings each gen, so learning the nuances of each platform does have its own curve. If I could spend a week or two with your machine.... I'd return it tuned for you! (plus the benchmarks for me)...

 

Hahaha. If you are in Edmonton sometimes, we can go out for a cold one and tune some PCs together. This is probably the most complex processor I have got my hands on with respect to overclocking. So many different aux voltage, somewhat new ram kit, and etc.

 

[email protected] Water (Custom)


[email protected]

 

You madman. Haha finally got it hunh?
I'm pretty much set on the 1950X though, will explain over steam why I dropped my plans for the EPYC/7980XE.

 

lol @|BloodBrother ... did the heart warming thought of eight hunge in Aisle 40 have sumthin to do w/those plans ... ... ... fitty dolla per core's lookin' good

​

 

Neah, got remote access permissions to use the office's render farm from home

 

In Hindsight: Some of the Worst CPU/GPUs Purchases of 2017
By Steven Walton on November 13, 2017
https://www.techspot.com/article/1523-worst-cpu-gpu-purchases-2017/

The Worst CPU & GPU Purchases of 2017

 

https://www.notebookcheck.net/Qualc...m-server-processor-Centriq-2400.263806.0.html

qualcomm server cpu, looks promising wonder how amd will be able to make good money with this CPU in place. going from anandtech's review, qualcomm is doing what amd is doing by providing massive number of cores while focusing on power efficiency while have a better value on top of AMD's eypc.

amd has more cache and pcie lanes though, and they specifically target dual socket system. @ajc9988 @tilleroftheearth what you guys think

 

It's an arm based chip, and Intel is more under siege than AMD. But, with the different architecture, you have to utilize a different OS, have less inter-compatibility than between Intel and AMD, need programs that can operate on an arm os, etc. Until there is a larger adoption of arm servers, which may come in a couple years, it isn't really a concern yet.

To put this in perspective, Intel has 98% share in the server market. AMD has less than a percent, while arm has even less. So, it depends on use, ability to shift over, etc.

 

They'd have to make it really easy to switch out to make it worthwhile.

 

Yeah; I admit it looks promising.

But if I'm hesitant to move to Ryzen/TR/Epyc platforms (even if they met my performance/productivity criteria) because of concerns such as stability, availability and compatibility with my current software and O/S setup, imagine how the customers that need 48 cores 24/7/365 will feel about taking a good/known setup and throwing it out the window to start 'new'.

Even taking into consideration the 3x performance offered and the 5x less cost...

Qualcomm will be in a much better position sometime in 2020+ when and if the O/S of their target customers, along with the software, of course, is offered, proven and then gives more performance at that time than what their rivals/competition offers then too.

Today? This would be great for a startup that is looking to get into this type of server at the lowest cost. At the same time; that lowest cost is going to cost them a lot in time to setup exactly as they need it (if it can be done at all, too).

I trust Qualcomm to have researched their target market and wish them all success in selling every one of these systems they can build.

Are any of the other players worried about them today? N0.

Because they'll have 10nm platforms out in due time too. With none of the drawbacks and limitations that 'arm' brings with it today.

Not to mention their proven history of dependability and compatibility with what the world runs today and into the near/medium foreseeable future.

 

Yeah, I couple years back, I hoped for quicker adoption, but after seeing current offerings, have less faith. With that said, 7nm TSMC arm chips at 4GHz have been demoed, with huge core count like this. GloFo, if they achieve 5GHz x86 on 7nm (still an if, not a when; but likely to get into the 4GHz range for 7nm HPC), we could see more adoption, unless Intel next year gets more adoption of their FPGA platform. If you are going to pay for that change, giving a real look at Qualcomm makes sense.

But we can't simply point to 10nm, as Intel's price cuts to partners (credits and rebates) are their best defense. This also supports your statement about small companies and startups that don't get them and pay closer to 70-80% of announced price, instead of 50-70% for larger partners. And those with no contacts and no way to get the price down, full retail is Hell, so AMD and this make WAY more sense for them.

But I do agree.

I've actually seen its server rack config. They are pretty nice, even though they can take up a decent amount of rack space. It is the same difficulty for switching out chips as with AMD and Intel. So that isn't a concern. They had some really cool showings at computex.

 

That's good to hear, it was sounding like there would be more changes needed, but if it's just adding more physical space for them, that's not nearly as big of a deal.

 

To show some of the ARM server offerings out there.