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

thats right, i like intel's cpu because they are faster. if you want me define my "faster" then its ipc per core at the same frequency. no need to be nit picking when you're in the wrong lol its a bad habit of yours hmscott.

 

First, GAMES DON'T MATTER ON HEDT! Makes me wish I could point to that rule I removed. Second, the microcode, bios, and OS updates, combined, showed improvement in many programs WITHOUT THE SOFTWARE ITSELF BEING UPDATED! THIRD, THIS IS TALKING ON HEDT USES, which are talking about applications optimized for Xeons that are highly multithreaded, not light multithreading and IPC optimized, except for professional programs like Photoshop and certain other that benefit from higher IPC with certain aspects of multithreading (many of these see benefit from 7700K@5GHz, thereby begging the question). Fourth, anything made for professional graphics card acceleration will likely see more benefit with the higher PCIe count and card support. Keep it on topic!

Sent from my SM-G900P using Tapatalk

 

That means certain use cases where that matters. It is important to define to avoid fights. You are correct on your definition. @hmscott has truth as well on his definition. Those different scenarios are what we are here to explore, not attack each other over!

Sent from my SM-G900P using Tapatalk

 

It's not a bad habit, it's good work practice, to find and rationalize all the details about something before making a decision instead of only on some single item that probably doesn't mean much in the face of all the details.

AMD is faster overall, cheaper overall, and has a wider scalable architecture with Infinity Fabric - from enthusiast 4 core through 32 core Professional CPU's, up through 64 core x 4+ CPU Enterprise solutions.

Intel isn't freaking out and running around like a chicken with it's head cut-off for nothing.

Intel know's that faster IPC per core isn't going to save them, and they have nothing even on paper that will save them.

Intel can muster a lot of resources, and I predict Intel will have several competing solutions brought out to compete with AMD, in 18-24 months.

 

Faster? Nah...

Cheaper? Yeah... (in every sense of the word).

AMD has 8 core right now - Intel far surpasses that in delivered and purchasable products.

Intel isn't freaking out, imo. AMD needs to shut up and deliver what it has been taunting for far too long with too little details...

In 18 to 24 months? I agree and also predict Intel will be competing with Intel once more in 'highest performance possible'.

For the 'value' crowd, AMD will be there. But value and performance are not the same thing at all.

 

Yes and no. Certain programs are still written for prosumers, made to utilize faster speed and higher IPC. Those cannot and should not be ignored. AMD can do certain things better, but I won't go so far as saying faster (not without qualification). In fact, Jay 2 cents showed it did premiere slower, although they didn't have certain things on the platform ready yet (still had growing pains; plus so close that he also said no practical difference and that he enjoyed using it during his 30 day trial, did not miss Intel at all). But, Adobe is widely used and cannot be ignored, while fitting @ole!!! 's criteria.

Sent from my SM-G900P using Tapatalk

 

In the past recompiling and rebuilding using optimizations for new CPU architectures was common place.

Back when AMD was an irritant and motivator for Intel to make real improvements in their CPU releases we actually looked forward to new releases that required rebuilds to see improvements.

It's been so long since Intel has had to actually change the architecture and create a situation where rebuilds are required to see the full performance of a new CPU release, most people don't think about it.

Ryzen is different. Real performance gains are being made to CPU intensive applications, games with rebuilds and tuning to take advantage the new CPU.

Games were first getting improvements from rebuilding and optimizing for AMD Ryzen, but regular Applications are just as likely to see improvements.

Jay recently said now he can go to Ryzen full time, only 3 months later, due to quick improvements by AMD and Motherboard partners improving fast memory compatibility and fixing instruction issues.

AMD needs to keep this close interaction up with hardware and software developers moving forward, it's going to be a long hard struggle to get everyone optimizing their releases for AMD.

Enterprise and Professional solutions will have even more fun, as most build from source, so they can quickly take advantage of optimizations for the new AMD CPU's

 

I think we all need to calm down a bit. As it stands Intel has the 10 Core 6950 out so it is the fastest you can buy right now! Soon Epyc will be available and then AMD may be the fastest single or dual CPU you can purchase. Hopefully shortly after, or there abouts, ThreadRipper will be available and so long as Intel does not have their 14 or so core version out, TR should become the fastest consumer level CPU until the 18 Core rears its head.

If AMD pulls the crown out from under Intel they will fight for it back, and we as consumers win in this battle. We get to sit on the sidelines to watch all the fun and benefit from it. Ain't life just grand right now.

 

Intel’s 28-Core Xeon Platinum 8176, 24-Core Xeon Platinum 8168 and 16-Core Intel Xeon Gold CPUs Spotted And Get Cinebench Benchmarks Leaked-Wccftech.com


Intel 28-Core Xeon Platinum 8176 Dual-Socket Server Rocks Cinebench Benchmark With 112 Threads

"Perhaps most interesting is the fact that a lower-end Xeon Platinum seems to perform better (see shot above) than the highest-end one, at least according to Cinebench. Where dual 8176 chips scored 6525cb in Cinebench, dual 8168 chips hit 7212cb, despite having fewer threads (96 vs. 112). This is thanks to the different clock speed, 2.7GHz vs. 2.1GHz (both stock, non-Turbo). The smaller of these two chips features less cache, but that's to be expected given the fewer cores."

 

While great information, again please calm down and keep this on the topic of platform comparisons.

 

I am as a customer rejoicing.

More of this, please. Still waiting for AMD to enter the high-end GPU market, let's go!

 

I love this comment below
Pcworld.com

 

if all you do is game, other than streaming, at this point you do not care about HEDT but you may later on. Point of this thread, comparing platforms. One of the few good things I can say about x299, you can break into HEDT later. The bad there is if you are a gamer on a 4 core only looking at 16 PCIe lanes could be an issue.

 

That is hilarious! With programs for editing video switching to graphics acceleration, that's a joke, while gaming tips at 6 cores until they change the coding, unless you are a game designer, running multiple VMs to bug check...

Sent from my SM-G900P using Tapatalk

 

That's the problem, you either jump to HEDT to get more PCIE lanes and spend a bunch more $$$, or go with AMD and save money and get way more PCIE lanes at any CPU core count.

What am I saying, why go with Intel then?, AMD is the only rational choice, amiright?

 

A rational choice is not between more or less PCIe lanes. A platform that you will depend and use for the next 18 months to half dozen years is much more than that.

Why go with Intel? Or AMD for that matter or anyone else that may show up to convince you to part with your hard earned cash?

Make a list of Pro's and Con's for both - including longevity of use and compatibility of O/S, programs (and programmers).

In this list don't include price as a major deterrent - because 'price' is only a one time payment (hopefully) - rather; include 'cost of ownership'. This cost of ownership is the cost of using the chosen platform and the benefits or detriments of the performance (productivity) hits you'll get each and every day it's in use.

This 'cost' includes any performance deficits' one platform may have over the other, each and every day you use it.

This 'cost' includes any waiting for fixes that may never come in terms of platform 'optimizations'.

This 'cost' includes any design decisions that may impact real world workflows/workloads and actual performance (productivity) vs. any marketing inspired bm 'scores' that look great on paper (and may even have theoretical benefits...) but little in the way of actual, day to day workflows for workstation type uses (i.e. single user).

At this point, we have no way to make the above comparisons (products are MIA...). The rational choice is to wait and see.

See?

 

Guys, @TANWare 's take on this is that this is a .VS. thread now.

Only discussions comparing Intel VS. AMD as explicitly listed in the Title, minus the Vega / Volta GPU's in the original title.

So the Title is now being taken literally with the "VS" the crux of the discussion.

We can't discuss beyond Skylake-X, or any unreleased products - I'm kinda fuzzy on this as Epyc, i9, and ThreadRipper are all unreleased, but announced.

We can't talk about Cannon Lake, Coffee Lake, or Ice Lake as they are unreleased - confused as they are just as announced but unreleased as the rest, but I guess they weren't mentioned in the Title.

I tried to reason with @TANWare that all those Cannon Lake, Coffee Lake, and Ice Lake are i7's too, as in the Title, but he didn't agree, and refered me to the Site's TOS - what he say's goes.

We also therefore can't discuss Intel VS Intel, or AMD VS AMD, as those aren't explicitly mentioned in the Title.

Rather than the original intent to expand the discussion to the new CPU and GPU's being released, which was perfect for current events going forward through all the new architecture releases from Intel and AMD moving forward.

@TANWare feels its more important to limit discussion than to embrace expansion to cover system's discussion.

I tried to explain that in my role as a Systems Architect I need to take into account all the options for parts that can be assembled for new systems in an overall design.

@TANWare 's response was that a Systems Architect doesn't need to discuss unreleased CPU's...

Of course that's totally a misunderstanding of my role as a Systems Architect, because 99% of my decisions involve unreleased products - usually 6-18 months before product release - otherwise my newly designed systems would be obsolete just as they are rolled out, every time

I always work with manufacturers evaluating unreleased - in progress of development hardware and software - and work closely with the engineers, product managers, sales and software engineers during evaluation and acceptance to get their products meeting our specifications and needs.

That is not at all unusual and is part and parcel of my job since the '70's.

IDK what to do to help @TANWare understand this need, maybe the rest of you can help and add your desire for an open place for open future systems discussion, which I think is what @ajc9988 was going for when breaking away from the limits of the AMD Ryzen thread.

Any thoughts?

 

Well, first, let's not point blame at anyone, especially a mod. They have a tough job as is trying to make the forums a great place to be, and they do a great job. They do let us stray, as needed, but help overall.

Second, it is helpful to remember that as information rolls out on either side, we may not have correlating info for comparison purposes yet, but we should still post and share that with the thread/community, that way it is available when new info comes to light. It is blind cheerleading that is an issue, not information that disagrees with us or our opinions (which is the better time to call it on this thread).

So, let's chill for the moment and see where this goes.

It is like @Papusan posting about the Xeons earlier. It showed a bias in a benchmark. Now, it would be nice to have someone multiply out the speed, cores, and throughput per core to see whether the benchmark favors speed to core count, as well as addressing whether that makes it a proper test for us to use in discussing the comparison of chips.

Further, the Xeon data should be kept as Epyc is nearly out, saving US duplicative research later for comparison.



Sent from my SM-G900P using Tapatalk

 

As I stated @hmscott if you want to discuss future systems and system architecture then please start a thread of your own in Desktop area unless you are discussing mobile chips, then please do so in this forum in your own thread. I give as much leeway as possible (everywhere) all things considered. Again for off topic comments refer to the TOS.

To all, we can not have such a general thread where everything and anything is fair game, I apologize but that is the way it is (in every forum).

 

games matter on hedt. i plan to have a desktop with 6 display game dota2 at 1080p 60fps on all of them at the same time with vmware. will do 3 cores each, this will be perfect on 18 core intel i9. its all how we use it me boys

 

That is one way to use HEDT. Not too sure TR would be a choice here. Other than to say you should look to a x299 that supports 3x video cards to be safe. 2x even 1080 TI's might be doable but pushing it. 3x 8 lanes is still only 24 lanes of PCIe so you should be fine.

One of my other points, 64 PCIe lanes are great but what do I right now need more than 44 for?

 

I can imagine many scenarios to use more than that. If you use professional graphics cards, x8 IS NOT ENOUGH! That takes 48 lanes for three cards. If you are running crazy raid setups, then 48 isn't enough. If you are doing high throughput fiber with 80Gbps or more connected with that raid setup, there goes 16 lanes. The problem is, 44 lanes limited people's imaginations. If you cannot think what to do with the extra lanes, it may not, on that criteria, be the product for you.

Sent from my SM-G900P using Tapatalk

 

for gaming, right now 8 per lane should be fine. Yeah some deep learning on crazy cards yea 16 lanes per slot may be required. Yes I can see where enough hardware can be installed but realistically where do "I" need more than 44?

 

I think you over estimate the speed you'll get on an 18-core.

Sent from my SM-G900P using Tapatalk

Edit: you need to chop off 100-200mhz, at minimum, per 2 cores added. The 10 core, without delidding and with being binned, his 4.8. That means at least 400-800MHz is gone on a binned chip. Considering the average chip will be 4.5-4.6GHz without delidding, this means 4.0 will be common, if not lower, on the higher core count, although they have to hit around that for competitive reasons with TR.

 

That is why I said that criteria might not be for you. But I've heard that a lot recently and wanted to make certain people don't repeat it needlessly or limit their build just because they refused to dream. But I definitely get your point.

Sent from my SM-G900P using Tapatalk

 

Give me as many lanes as possible. I will find a way to use it all, hahaha.

 

That used to be me, when it was my money. Now that I am retired and it is a hobby I am much more passive.

 

Just my 2 cents - There's value talking about future CPU's as long as they are going to be supported by the motherboard. Anything unconfirmed is a slippery slope in the discussion.

 

TANWare, where is the TOS? Can't seem to find it with search...

 

That is my philosophy. Now, AFTER Zen runs its course, in 2020-21 (Intel at 7nm unless they jump and AMD at 5nm), then we'll have pcie 5.0, which at 4 times the bandwidth, I bet it will be hard to saturate for awhile if you had 64 lanes (especially with modern cards only needing about 4x PCIe 5.0 lanes to run)!

Sent from my SM-G900P using Tapatalk

 

The only sticking point could be open sourcing / sharing / licensing the next generation IO interface for the Intel/Micron "Optane-like" high speed storage which could take advantage of 8(?) lanes.

Then those 64 lanes don't seem like "too much" when a 4x RAID = 32 PCIE lanes of throughput.

Until then...

Due To Interface Limitations, 3D XPoint SSDs Will Only Be Able To Deliver 6GB/s Of Bandwidth
http://wccftech.com/interface-limitations-deliver-actual-3d-xpoint-ssd-speeds/

"...The only thing standing in way of high read and write speeds will be the interface limitations, which at this point, will only be able to deliver up to 6GB/s of bandwidth. We are confident that when newer and improved interfaces are introduced, we will finally be able to witness the true potential of 3D XPoint based SSDs."

 

I have a gut feeling that with PCIe 5.0 (and even with 4.0, coming soon...) that RAID will effectively fall off the radar of any truly real world performance seeking individual. Unless you have another RAID0 array that you could show off by copy files back and forth; that kind of performance (mostly sequential, btw...) will be only 'potential***' in most systems. Even with NAS' equipped with 10GbE connections or better; even a single PCIe SSD today can saturate the pipes used...

Maybe Intel knows what they're doing saving die space/design complexity/etc. by not having PCIe lanes 'needlessly'? Idk... but it is something to think about (i.e.; their 'plan' going forward).

Something like their much laughed at igpu's... that I've been using since they've debuted to huge benefits (heat, battery life, power consumption, $$$ savings) in my workflows. When they do increase the performance of their igpu's; it is real and tangible in everyday 'flows...

Sometimes; what we think we 'need' is not what we actually 'want' - I'm confident enough in my powers to test for real gains that I'll leave the cpu building part to someone like Intel and even AMD (if/when they grow up a little).

Alright AMD; I've said it. Show me I'm wrong and I'll put my $$$$ where my mouth is.


***Like having an 8 second QM car and living within the city core... either you'll pay more for fines, tires, tuneups, etc. (for redlining, speeding and stunting) than fuel, but you almost certainly won't use that 8 sec car to it's maximum at all.

 

For forum rule you may start here. There are other things to loook at in the general area as well.
http://forum.notebookreview.com/threads/forum-rules.109941/

 

thats correct with 18 cores and a 3 fans closed loop cooler can probably only do 4-4.1ghz on all 18 cores with delid and binned chip. i mean the reason having so many cores is that trade off with high frequency and can still be plenty fast, where as xeon cpu are stuck at default clock like 3.2ghz lmao.

not to mention turbo boost max 3.0 is there for me when i need single threaded software, hope to jack that thing way up in the 5ghz+ area.

yea ssd improving fast, imho once we all hit optane we'll stop notice any difference, from that point on its all how software take advantage of multi threaded storage workload ontop of that using multi process on cpu.

and i agree in raid, i'd rather than 1 SSD with everything fast than having 2 in raid 0 for just a tiny bit better performance. lots of headache when comes to raid.

 

There are many other RAID's than RAID0, and many good reasons and architectural reasons for combining storage devices, cascading them, and failing over to other storage.

What ever it's called in the future, being able to use grouped or singly, 4 x 8 PCIE lanes is still 32 out of 64, which has me beginning to wonder if we need more than 64 PCIE lanes sooner than later.

 

So, my point is that the SL-X OC will be in line, roughly, not-delidded (most professionals don't delid as it voids the warranty) with the 16-core chips from AMD which don't need delidded as they don't use toothpaste!

Also, unless you only have a single task going, boost 3.0 means nothing when everything is used max (or even as more than two cores are needed). I think it is nice only if you set the affinity.

And I'm not talking raid 0. As hmscott pointed out, there are many more types of raid. Hell, with raid 5 you get way more speed plus data security in redundancy, thereby achieving much faster transfers. Now, if you throw 1-2 40Gbps 4xPCIe cards in the build, you are able to use that server in a manner befitting a data king! This is assuming you have fiber network to utilize those speeds, of course. You don't dream big enough.

Exactly. Now, the fact a pcie 5.0 4 lane could handle a 160Gbps fiber connection, I think the state of storage will change drastically. Then comes the question of increasing capacity significantly with the speed.



Sent from my SM-G900P using Tapatalk

 

Your assumptions are so off...

Even with 2x 40GbE connections between two compatible devices (let's say a workstation and a NAS...) that is still only a theoretical 10GB/s (at perfect scaling with the right type of file and compatible network hardware, end to end).

And note: with a single transfer happening; it will only happen at 40GbE speeds anyway...

A single PCIe card can barely do ~85% of its rated spec's over LAN (and not even on all aspects of those spec's either...).


And yeah; RAID0 isn't the only RAID. But considering RAID 'anything' with today's SSD's or Optane SSD's is likely a waste of time and $$$$ for a consumer and even most 'professionals'. RAID5 and RAID10 do increase performance over a single drive, even if it's mostly sequential (how many TB's do you shuffle each day between your systems?). But hardly worth the $$$$ invested (you need a real RAID card with real SSD's - not TLC garbage) to the actual real world gain you'll see doing real transfers of mixed file types.

Let's say as an example that you are transferring TB's worth of data a day (and you have a budget to replace those worn out SSD's the moment they're 'gone'). Without RAID; ~3GB/s is just under 6 minutes per TB (theoretically - if we're just including huge zipped files...). With three of those drives on a RAID card worth owning (cost x 4...) you'll get 3x the performance and save a whopping ~4 minutes per TB, in ideal 'test' situations.

That sounds great except for the part where you'll be replacing SSD drives every ~10 hours or so (of use). Yeah; we need lots of PCIe lanes now. Lol...

I've stopped using RAID 'anything' except in my NAS' for over a decade now. Haven't missed anything with regards to real world day-to-day performance (i.e. 'productivity') or reliability (actually more reliable today than any previous/older platform(s) I've ever used).

Once again; your interpretation of the 'theory' and number crunching leaves much to be desired. At least for any real world applicability.

I don't want to you or anyone to stop 'dreaming big'.

But after waking up, try to make those dreams applicable in the world we are in now.

 

Intel Core i9-7900X 10 Core Processor Overclocked to 5 GHz on AIO Liquid Cooler, 4.5 GHz on Air With 1.15V Across All Cores-Wccftech.com

"Intel’s latest Core-X series processors might feature some impressive overclocking capabilities as demonstrated by two overclockers. The previews posted by Der8auer and Lucky_n00b show that Intel’s 10 core, Core i9-7900X can reach 5 GHz on AIO cooling."

Intel Core i9-7900X No Delid, AIO Cooled Hits 4.8 GHz and 5 GHz With Delid Across All 10 Cores
"The demonstration posted by Der8auer shows that he was able to hit 4.8 GHz on the stock CPU interface across all 10 cores and 20 threads. A Corsair AIO 280 liquid cooler was used but nothing as aggressive as a custom loop. Liquid cooling is the go to option of many enthusiasts who purchase HEDT platforms so it makes sense previewing the overclocking capabilities with such solutions."

"The chip was later delidded and the stock thermal paste was swapped with Liquid Metal. This had the CPU running 200 MHz higher at 5 GHz across all 10 active cores. Temperatures were in the 88 to 91 degrees Celsius but it’s mentioned that you can gain even further headroom of 5.1 GHz if you are running a custom loop water cooling solution. It’s stated that the average Core i9-7900X can hit 4.7-4.8 GHz without delid which is impressive"

Btw.
Intel Core i9-7980XE, Core i9-7960X, Core i9-7940X Processors Arriving in October – Core i9-7920X 12 Core Launches in August

 

i saw that and think threadripper might be in trouble.

at 5ghz each of those 10 intel core is clocked at 1.5ghz over or 40% over 16C/32T ryzen core, so approximately the same performance ;/

 

That's a lot of extrapolation from one "cherry-picked" CPU directly from the trembling hands of Intel's chief binner into the hands of a professional delidder / overclocker.

Lets see how the production line CPU's perform for Joe Schmo upon release... 4.3ghz turbo on 2 cores, 4.5ghz turbo on 1 core, 3.3ghz base with stock air-cooling.

Of course AMD will be slower, but that's the charm, buying AMD even though it's slightly slower than Intel, because no one is forcing you to pay Intel yet another time.

Don't let that logic bomb go off in your head, blowing your cash into Intel's bank account.

"But, it's faster!!, I *must* buy it, right??"

 

My best guess at 5.0 is apr. 2750 in R15, whereas a 16 core TR hopefully will be better than 3,000 plus in R15. So close but no cigar. Also that was a binned CPU etc they are figuring on average 4.5 to 4.6 which again is fine by me.

 

You forgot one crucial detail.
It hit those speeds with liquid nitrogen.
So, this ain't water cooling or air cooling.

I wonder how many people use liquid nitrogen for their cooling.

EDIT: At the time of writing this post, I remembered reading another article on i9 being Overclocked with liquid nitrogen to higher speeds.
That's why I posted that.
I did not notice this article was newer and meant air cooling as I skimped through it mostly.

 

Maybe re-read?

 

AIO cooling is jut standard liquid cooling. Not just a Heatsink with AIR cooling.

 

Ah. My mistake. Didn't know about implied air cooling

 

http://wccftech.com/amd-ryzen-threadripper-1950x-specs-performance-leak/
New benchmark comparison of the TR 1950X versus a BW-E 16 core Xeon. Note how slow of ram was used. Intel had 2400MHz ram (normal, and doesn't benefit as much from fast ram) compared to AMD having 2133 (and greatly benefits from faster ram, with around 10%+ going to 3200MHz because of Infinity Fabric, which would be running around 50% faster).

Edit:
https://www.kitguru.net/components/cpu/matthew-wilson/source-amd-to-launch-threadripper-in-august/
TR release date rumor


Edit 2: https://browser.primatelabs.com/v4/cpu/2217479
Here is an 1800X overclocked to 4.0. So take the WCCFtech bench with salt. The link they use doesn't work and the naming scheme does not match at all with any other rumors.

 

I responded on the other posting, those numbers are horrid.

 

I answered you there as well. I do agree!

Sent from my SM-G900P using Tapatalk

 

I love it... 'one cherry picked CPU directly from the trembling hands of Intel's chief binner'. Like poetry for geeks.

I don't buy anything for charm or because it's slower. Nobody is forcing anyone to pay Intel a dime... Time is $$$$ and performance/productivity isn't something you fool around with to have a one time saving of even $500 over a few years of use from a platform.

Of course; don't let logic interfere with making a rational decision. Once the money is gone; I don't care who has it. I just care what I benefitted from giving it.

Those leaked TR spec's are pretty bad right now. This isn't a little slower; it's kindergarten vs. college when $$$$/hr are in play...

 

You could correct your original post, it's misleading to people, explicitly stating things that aren't true... it's that "Edit" button.