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

yea what surprises me is the temperature of these chips, too high. since broadwell-E can't overclock that well couldnt really compare unless one were to use broadwell E chip and skylake x chip at similar voltage at same clock then we could tell.

thats why we have silicon lottery, im willing to pay 200-300 more for a good chip rather than buying a 12c and hope for the best.

 

That isn't exactly true. The average overclock on air was 4.2GHz on the 6950K. The average overclock on water was 4.3GHz. So, you are looking at a couple hundred MHz difference.

Also, it is more appropriate to check heat to heat rather than voltage to voltage. The reason I say that is Kaby used much less voltage, but put out almost as much heat as Skylake (edit: at a set multiplier). You can go further on OC because of lower voltage, but if the heat is roughly the same, you still wind up limited (although it is safer, to a degree, with the lower voltage).

http://hwbot.org/hardware/processor/core_i7_6950x/

I would wait to see how many fall into each category first. Work out the math on likelihood of getting a dud versus paying the premium. If it is a tight OC grouping and few fall in the lowest one or two categories, then taking a shot can be better. But to each their own there.

 

incorrect on multiple points. it is important to check all aspect of things when comparing to architecture of chips. the chances of skylakex able to overclock higher than broadwell E is high because better silicon quality. there are times where your voltage is low and you try to overclock, and it just wouldnt allow it as that chip as hit it's dead end due to its internal physic reason.

 

"Interconnect"?

Do you mean network connectivity?

Mellanox is providing ethernet communications...

 

Wrong. You hit different physical limits. Heat, voltage, and silicon. What I am saying here is the silicon can handle it and doesn't plateau like Ryzen. Voltage isn't as high, which voltage can kill the transistors just like heat and voltage causes heat. Finally, once it hits a certain heat, even if the voltage is lower, it burns it up. Poof! I showed, specifically, what was gotten with BW-E on water and air. You get a couple 100MHz more here.

Edit: What I am saying is you will be heat limited, not voltage limited, on these chips. Also, read up on the fact it has Integrated Voltage Regulators on the chip. It just isn't a full FIVR. This is likely why the voltage runs more similar to Haswell and Broadwell than Skylake. That means what will fry it with voltage may vary and you may want to pay attention to what others wind up considering a "safe" voltage.

 

InfiniBand (abbreviated IB) is a computer-networking communications standard used in high-performance computing that features very high throughput and very low latency. It is used for data interconnect both among and within computers.
https://en.wikipedia.org/wiki/InfiniBand
http://www.mellanox.com/pdf/whitepapers/IB_Intro_WP_190.pdf
http://searchstorage.techtarget.com/definition/InfiniBand

Edit: Now it was not made clear if that will be used solely for network connection or in I/O in other uses, admittedly, especially since infinity fabric is used for comms between CPUs on the same MB. But, I do not know precisely where it will be used, nor did the article say, suggesting primarily as a network comms, which is the primary use to date.

 

I was trying to say to not use the word "interconnect", it's for networking, like ethernet. I didn't mention infiniband so as to not confuse the issue, but it's a network too.

" InfiniBand (abbreviated IB) is a computer networking communications standard used in high-performance computing that features very high throughput and very low latency. It is used for data interconnect both among and within computers."

Even the definition call's it an "interconnect", sheesh - why isn't network or networking not enough to communicate the thought?
https://en.wikipedia.org/wiki/InfiniBand

They seem to want to use it as "interconnect" for storage too. It's a network too, why do they want to use "interconnect"?

It's a network, dammit!

 

Read the white paper. I think you will see what they want to do. Check out the table starting on page 8.

Edit: They later say this: "Several of these new interconnects are actually key enablers for InfiniBand as they provide access to new levels of processor bandwidth and allow InfiniBand to extend this bandwidth outside the box. Technologies such as 3GIO, HyperTransport, and Rapid I/O are being developed and this will provide InfiniBand with a point of attachment to system logic that can support the 20 Gb/s required by 4X InfiniBand links and even the 60 Gb/sec needed by12X InfiniBand links. These technologies complement InfiniBand nicely."

Yet they also compare it to PCI-X. Meanwhile, they discuss using a switching mesh in between the systems. So, I definitely see what you are saying. But, they may also be vying to become an alternative.

Edit 2: They are just using interconnect in a VERY broad definition, while also wanting to be used on PCB, not just connecting systems externally by networking.

 

I'd focus on the AMD Epyx build(s) info for IB usage, like here where it's said to be used for storage connectivity - instead of Fiber or Ethernet:

AMD Zen Based Naples Platform For High-Performance Computing Detailed – 128 PCI Express 3.0 Lanes
"The latest details show that AMD will be offering an optimized GPU server platform with Naples CPU. The AMD Naples platform will feature a maximum of 128 PCI Express Gen 3.0 lanes. The huge number of lanes will allow support for a vast amount of devices.

The detailed block diagram shows that a 1U rack will be able to support up to 32 NVMe devices and four discrete GPUs. The 1U rack will feature two InfiniBand EDR interconnects for data communication between storage and server systems."

Usually I set up fiber to storage, or ethernet although I've set up Infiniband it's usually too expensive. It's a big surprise that AMD would saddle their builds with Infiniband, where 100GB ethernet would be cheaper.

Now that there is 100Gbit Ethernet IDK why they would spec Infiniband (56Gbit), unless they have dropped their prices to be more competitive.

With the 100GBbit / 200Gbit IB HCA's I can't tell if it's per link one way or summed up 2 way, or summed up combined for all interfaces. AFAIK 56GBit is one way per link, but it could be newer stuff than I have seen. I kinda lost interest in it when 100Gbit ethernet was available.

For now lets drop it until we get better information on how IB / Ethernet is being used in these builds, and what AMD is endorsing.

Mellanox might be trying to promote Infiniband along side or on top of the actual 100GBit Ethernet being used in these motherboard implementations.

 

Wrong. but i'll let you guess where its gone wrong this time

 

Btw, voltage doesn't cause heat on it's own; amperes do.

Voltage is just potential energy - depending on the resistance of the circuit in question; while amperes is actual energy used...

But even amps don't cause heat on their own - i.e. X Amps doesn't give you Y heat in Celsius; just like ole!!! says and you hint at too; the actual details matter here.

 

Do what you want. Run it at 95 constantly and enjoy your thermal degradation.

Sent from my SM-G900P using Tapatalk

Edit: also, if you don't understand how watts is volts times amps, how resistance plays in, then how it correlates to heat in the circuit which must be cooled or it can vaporize or degrade the transistor, nor why I'm speaking more generally about heat degradation...

 

yes i shall do what i want!

 

BOYS @Papusan @hmscott @Mr. Fox @Prema @jaybee83 @Takaezo @Meaker@Sager @johnjames @jclausius @Blacky @Dominick_7 @TBoneSan @ajc9988 and everyone that i missed.
http://community.acer.com/t5/Acer-Ideas/i9-7920x-high-end-gaming-laptop/m-p/503436#M840

hope the guy isnt being sarcastic cause that would rather be an awkward attitude coming from Acer support, is it happening?

 

Wow, wouldn't that be nice...

Although, I was hoping Acer would put a cost effective Ryzen CPU into their next 21x, the price is already too high

 

rofl yeah 9k definitely isnt cheap and its probably the minimum configuration too.

 

I would have though too, the Ryzen would be TDP wise better suited as well.

 

Intel i7-7740X CPU Review vs. 7700K, 7900X, R7 1700, & More
http://www.gamersnexus.net/hwreviews/2965-intel-i7-7740x-cpu-review-vs-7700k-not-worth-it

 

Acer used the 21" JOKE to promote the brand. Nothing with socket hardware in notebooks will come from this company. Don't hold your breath bruh. They will follow the others and push out same welded on hardware!! In thin flimsy format. I hope you haven't forgotten Acer's engineers said if you OC'd the hardware above recommended max for this laptop, everything could go up in flames. And this statement Is confirmed!! By Acer themselves.

 

I agree. Acer is not trustworthy and I doubt they have the technical skill to properly execute on it and deliver a product that functions correctly. And, even if they did, only a idiot would pay $9,000 USD for a laptop. I've been called an idiot for paying $4,000 for a laptop. I cannot build the kind of desktop I would want to own for $4,000, but $9,000 for a compromised piece of trash from Acer is over-the-top insane. In fact, I'm not even going to burn any more calories talking about it or speculating on what might follow.

 

tbh i wouldnt mind paying for something thats one of a kind with what i really want 9k is definitely on much higher side compare to clevo. question come down to is that, if acer do make one with intel i9 cpu or TR, their bios would be junk but it could be the only machine that comes with that type of hardware. would 9k be justifiable for something unique in a laptop form.

 

http://www.anandtech.com/show/11558...nm-plans-three-generations-700-mm-hvm-in-2018

So, a bit on processes and timelines, focused on GloFo, but has a table for all, including Intel, although I believe the use of 10nm+ from Intel has been accelerated and may be in the H2 2018.

It also seems the 40% increase is focused on frequency, which places 3.6, if fully increased by 40%, at 5GHz, while 3.9 would be 5.46GHz. That is on process developments, whereas IPC would be improved by architecture refinements.

Also, they are looking at 0.65-1V operation.

But wanted to show what the foundry for AMD is doing for chips next year.

Sent from my SM-G900P using Tapatalk

 

No, that's beyond the reasonableness in cost no matter how special it could turn out to be. If I were going to spend $9000 USD on a computer it would ONLY be a custom-built desktop and it would utterly destroy the notebook in terms of performance.

 

There would have to be a new category for HardwarePorn before I drop $9,000 on a laptop.

 

dual socket epyc + 6way vega crossfire in a laptop form factor? add that transformer as a psu

Sent from my HUAWEI NXT-AL10 using Tapatalk

 

at that point its gotta be all optane SSDs in there to be worth it, in terabytes.

 

hey, the least you could do is be positive. though i wouldnt hold my breathe on it and reason is simple, intel being the best in the industry having issue on that front, glofo and TSMC got ways to go in silicon quality, theres also amount of money involved, intels got way more cash.

if ryzen can hit 4.7 ghz with a decent voltage on say 8-12 cores, i'd be extremely happy already along with a nice 5-10% ipc boost. AMD claim 10% boost with ryzen 2 right? thats good enough to catch up to intel's skylake and be on par or even better than skylake.

 

Actually, before announcement of Zen 2, they claimed 20% IPC for Zen+. Now, it is uncertain whether they meant the 14nm+ design or Zen 2, but that means they are chugging along.

As to the other companies, they all mostly skipped 10nm for a reason. Samsung, GloFo, and IBM are in a research venture together and GloFo went with the process developed by IBM to make a 7nm chip from 2015. TSMCs being a low power variant and already doing risk this summer, while announcing 5nm already and being the second largest foundry, IIRC, I'd actually say you may be wrong.

Intel has the most dense logic on a given node, just everyone jumped over them. So, density wise, Intel now has to beat smaller nodes.

BTW, because of IBM giving the foundry, IP, and $1B to GloFo awhile back, that made them the third largest, with Intel now the fourth largest foundry.

But I expect good speeds and IPC.

Sent from my SM-G900P using Tapatalk

 

...and, it would need to function like something that costs $9,000. I can almost guarantee that will never happen. The people that manufacture laptops seems to go to great lengths to ensure that their products have performance that sucks out-of-the-box. So, paying $9,000 for some crippled piece of dog dukey would most likely never make a lick of sense. We've seen example after example of laptops with magnificent hardware with their gonads cut off by cancer firmware. I doubt that will ever change for the better.

 

To give you some information on TSMC's 7nm, which is who GloFo and Samsung are really competing with at the moment for clients with, and which will be used for the Volta successor which will fight with Navi:

March -
"It’s 7nm’s turn for the spotlight today. TSMC claims a lot of improvements made on its plans for the process. The manufacturer reports 76% yields on 256Mbit SRAM modules manufactured on 7nm. The manufacturer also claims that an ARM Cortex A72 processor in the node exceeded 4GHz in frequency.

Following this progress, TSMC will commence risk production for first generation 7nm process next month. The fab will generate a total of 20 tapeouts for this year. The Cortex A72 processor which exceeded 4GHz, utilized the company’s 7nm HPC platform. This is an upgraded form of 7nm and features high performance transistors that guarantee a 5% speed gain over basic 7nm. 7nm in general will deliver 3.3x increased router gate density and a trade-off between 60% less power or 35% processing speed."
"7nm+ with EUV will enter mass production by 2019. ... 7nm+ will feature 1.2x times logic density and a trade-off between 10% speed increase or 15% power efficiency over its predecessor, 7nm."
http://wccftech.com/tsmc-7nm-proces-sram-details/

April-
"TSMC has already showcased a 7nm SRAM chip, a key milestone on the road to more complicated SoC circuits, and states that it is seeing “healthy” yields from its process. The company has also recently been reportedly testing a 7nm, 12 CPU core processor developed in conjunction with MediaTek.

Small volume risk production of a more advanced 7nm+ process is expected to appear by June 2018. This suggests that TSMC is quite far ahead of Samsung, which could lead to the company securing a number of contracts from its mobile chip rival. Apple already moved over to TSMC last year, and Qualcomm may being eying up space on the foundry’s 7nm line if Samsung’s technology remains a year behind."

"[Intel] has been later than its competitors with 10nm and Intel is looking increasingly behind the curve in the pursuit of 7nm too. It’s production line is not expected to start rolling off volume production until later in 2020."

http://www.androidauthority.com/who-will-be-the-first-manufacturer-to-7nm-757916/ (article comparing all 4 major fabs and the race to 7nm)

"In last week’s investors meeting, TSMC Co-CEO Mark Liu added to Chang’s report, stating that the company will likely move its 7nm process technology to volume production during the first half of 2018. He said that over twenty customers are currently “engaged” with the company regarding the new process technology, and that fifteen customer tape-outs are scheduled to take place in 2017."

“N7 is a further extension of N10 technology, with more than 60 percent in logic density gain and 30-percent to 40-percent reduction in power consumption [compared to 10nm].”

"Although Liu did not reveal during the meeting what companies have already jumped on the 7nm process bandwagon, ARM announced a multi-year agreement last month to collaborate with TSMC on 7nm FinFET process technology to create high-performance, low-power processors."
https://www.digitaltrends.com/computing/tsmc-7nm-2017/

May-
"TSMC’s first-generation CLN7FF will enter risk production in Q2 2017 and will be used for over a dozen of tape outs this year. It is expected that high-volume manufacturing (HVM) using the CLN7FF will commence in ~Q2 2018, so, the first “7-nm” ICs will show up in commercial products in the second half of next year. When compared to the CLN16FF+, the CLN7FF will enable chip developers to shrink their die sizes by 70% (at the same transistor count), drop power consumption by 60% or increase frequency by 30% (at the same complexity)."
http://www.anandtech.com/show/11337/samsung-and-tsmc-roadmaps-12-nm-8-nm-and-6-nm-added/2

"The Samsung-TSMC match-up is an interesting one. TSMC appears to have more capacity on-tap for 10nm and is planning a quick transition from 10nm in 2017 to 7nm in 2018. Samsung, in contrast, is planning to hold at 10nm for multiple product generations. As nodes have become more difficult to hit, each foundry has adopted various methods of defining what constitutes a node shrink. This is most apparent with Intel; its 10nm node is expected to be equivalent to 7nm from TSMC."
https://www.extremetech.com/computi...samsung-globalfoundries-intel-gear-7nm-beyond

"The Nikkei is reporting that TSMC is currently ' testing cutting-edge 7-nanometer chips for 12 customers, and the company will start to make the product in large quantities in the first half of 2018.' ... With Apple being TSMC's number one customer, it's likely that Apple will be able to secure 7nm processors in sufficient quantity for their 2018 iPhone." (emphasis original)
http://www.patentlyapple.com/patent...omers-will-ramp-up-production-in-h2-2018.html

So, considering TSMC is on a roll, has tapeouts in from some customers and are working toward HVM, they will be first, most likely, to 7nm, with densities equivalent to Intel at 10nm. Then, due to taking the IBM fab and the research agreement between Samsung, IBM, and GloFo, the licensing of the 7nm process from IBM, and the IP agreement between AMD and GloFo, we are looking at AMD being one of the first customers on GloFo's 7nm process, with tapeouts coming in Q4, and Su suggesting the first chips on that process may be in Q3, although Q4 2018-Q1 2019 seems more reasonable. Considering TSMC has already produced a 4GHz ARM chip for HPC and IBM produced a chip in 2015, and the engineers from IBM are currently working at GloFo (that is what the $1B payment was for, a guarantee the employees would work there for at minimum a couple years), I'm not seeing why you suggest TSMC and GloFo have a ways to go. If the logic density is there, it doesn't matter if it is at a lower node. They are going to have equivalent densities. Then you get the energy efficiency of 7nm vs. 10nm, any speed increases, etc.

Just trying to give insight into what will be in play for the next couple of years between these players. TSMC is relevant because AMD has an agreement in the Wafer Supply Agreement amendment that if GloFo is unable to perform, it can go to other Fabs, making TSMC the logical alternative.

 

I have to say thanks to AMD, because of them I have lost almost all interest in HEDT. Intel from day one with paste has turned me off of everything they have. AMD with their lack of info has now done the same, just nothing other than vague promises to sustain any interest. TBH Vega at 375w TDP has me turned off as well.

I was thinking 7900x, but again the paste, omg why paste? Is it they could not think of a better way to screw it all up, oh yeah at least it is not BGA.

 

AMD is what is keeping my interest alive. Intel turned me off with the toothpaste and incrementalism. Meanwhile, after looking further into GloFo's 3 steps of 7nm processes, I am left wondering if Zen 3 well be on the second or third process and whether or not EUV will be ready. Zen 2 is where the fun starts anyways, as is ice lake for Intel HEDT.

Sent from my SM-G900P using Tapatalk

 

Yeah, same, it saddens me that the TDP is so absurdly high on these high-end technologies. I mean dear god, the power draw is going to cost a percentage of the total build price... christ

 

They are going to have to start rating these Multi GPU box's in BTU's.

 

when u say intel's 4th, looking atm intel is still destroying them. until zen2 comes and 7nm comes i wont even consider other cpu lmao. IPC is hella important! snappiness in computing thats what i need, high ghz with good architecture.

 

MSI Leaks Intel Coffee Lake 6 Core Mainstream CPU Performance – Compared to AMD Ryzen 5 1600X, Single and Multi Core Performance-Wccftech.com

 

This is by foundry size. Facts are facts. Intel was third and GloFo was fourth until the IBM deal. Do what you want, but don't ignore facts on the ground.


Sent from my SM-G900P using Tapatalk

 

I told you coffee looked really good!

Sent from my SM-G900P using Tapatalk

 

Coffee lake 6 Core chips will be a big disappointment if you want go for high end laptops with socket hardware. And I haven't talked about coming 6 Core 45W BGA chips. Almost BACKWARDS into the future

 

The Coffeelake benchmarks the same as the Ryzen 1600x, but is 45w instead of 65w, so it will throttle under sustained load, unlike the Ryzen

Also the Ryzen 1700 is shipping in laptops, so it will outperform the Coffeelake - Coffeelake only matches the lower end sku 1600x laptop CPU, and only for 8-28 seconds.

No mention of Coffeelake release date, except it's been said to be 1H 2018. The article misleads the inattentive reader by stating the HEDT CPU's release in October...which isn't when Coffeelake is supposed to release - so why mention that date unrelated to Coffeelake?

Intel is dooooomed

 

Coffee Lake i7 6 core in next *high end* Clevo that will be eaten by a mainstream Ryzen... No thanks. I will pass. There will not be a new laptop for me as it looks now. Asus + Ryzen 1700... Same, No thanks (even a worse option than Clevo's next P870/970 series). It's seems a new Laptop for me is at Dead End!!! R.I.P

 

Wow, Intel just can't win lately...heads up, if you have a Kabylake / Skylake CPU:

In short, disable hyperthreading in the BIOS until you can get a BIOS / microcode update to fix the errata.

Skylake / Kaby Lake Hyper-threading bug

 

Intel has completely lost their sh**.

They are releasing SO much, that they are essentially cannibalizing on their own low- and previous generation releases. You buy something now, they release something new within the minute, it is like the Titan XP XP XP XP XP nTOILETIA thing all over again.

 

My P770ZM will last me awhile. I've changed it to be my work laptop, and desktop for personal. I'm going to just work on a house server using TR, most likely. If it wasn't a 45W part, a 90-95W part would have done wonders.

I posted that yesterday. Now I have to keep an eye on the Asus website for the update.

 

I am on my 3940xm and have been happy for a while. 16 core sounded nice but I just can't play the waiting game, too stressful.

 

Then don't. You have enough info on why Intel's offerings are not great and need time to smooth out the release. So just don't look as if buying, wait until August and see information, not with a view to buy. If the information changes your mind, then no problem. Otherwise, you wait for one more generation (AMD; Intel, you may want to wait for Ice Lake-X).

 

No, With W7 on the outs what I have this year will most likely be it for a very long time. I will not go W10 and I needn't go further for Linux for quite some time.

 

I use win 10 enterprise heavily modified. I also need to brush up on Linux. But, as you've seen, everyone that has seen and knows more of Epyc really sing it's praises.

Sent from my SM-G900P using Tapatalk

 

I'd love to see Clevo bring back the big non-intel-core workstations, but IIRC they didn't see much actual demand for them.

 

That may change with AMD now having processors in the race. Also, did you see Epyc able to run VMs both encrypted and not side by side?

https://semiaccurate.com/2017/06/22/amds-epyc-major-advance-security/

This is a little about the security on the chips.

Sent from my SM-G900P using Tapatalk