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

i was wondering where you went when you stopped posting in the gt83vr section. shouldnt u still need a laptop even in university?

 

I just got everything in, at least I hope so. Not sure if I have all incidental cables yet and what not. Once together I may decide on a need for more fans etc.. I have not touched a desktop since late 2,000's so all my old power splitters etc. are long gone.

 

An added post, my kids have misplaced every good Philips head screw driver. easilly over a dozen various good flat heads, but they have lost all the others. My laptop #0 is right in my desk but the others are no where to be found. I have to stop letting other people use my tools!

 

I've been running it for weeks on an AIO [email protected] on a taichi. My large order of water cooling parts comes in on Friday, small order on Tuesday next week, and still waiting for my XSPC pre-order to ship. Hoping it does before Maria. I'll be picking up a home water filter with 5 micron filtration (or smaller) and debating using my ddc over getting a submersible. If the particulate is all filtered out, it shouldn't matter. Then again, a pond submersible is $10-20. That is for flushing the 3 Nemesis Black Ice 480GTXs. What do you wanna know. I'm loving it!

 

I'm going to have 17x120mm fans, 1x200mm fan, 2x ram fans, 2x pumps (MCP50X DDC pumps), 3x fan hubs with sata power.

EK graphics block and koolance VRM blocks (and rigging up backplates). XSPC Raystorm Neo sTR4 CPU block.


Edit: in a stroke of inspiration (or just a stroke), we should start allowing benches of the TR and Intel 10-18 cores chips to be posted here. AFTER all, this is a comparison thread and people want real world results...

 

Yay for the 4xDMI3.0 bottleneck!

 

Myself I have no issue posting BM's to the thread as long as being used for comparison, otherwise they belong to the AMD specific thread. Normally within a laptop forum this would be a no no but since the tech is meant to migrate to mobile, both AMD and Intel, it is good to have a base knowledge of how it goes on the DT.

 

Let me explain why waiting for Intel 8-core will NOT be worth it:
"
Rome - Epyc's Zen2 processor, codenamed starship, will have 48 cores. This means they are either adding another 2 cores to each CCX or adding a third CCX or due to the manufacturing die shrink, plan on putting 6 dies on the chip. The first two mean that the mainstream could see a 12 core chip, which would be hilarious as it would arrive as Intel releases an 8-core mainstream chip. It also would mean up to 24 cores on the TR on Zen 2. Let's see Intel cannibalize their Xeon 24 cores for HEDT! HAHAHA! This is combined with improved IPC, which is easier to get after the first generation of a new design and the fact that logic density for AMD at 7nm will be equivalent with Intel's 10nm process.
Also, notice how Intel, except in mobile, seems to have killed the cannonlake architecture." Me in another place.

But the point remains valid. Rome will give 50% more cores. If they are adding a third CCX or adding two cores per CCX to achieve this, that means mainstream Ryzen will likely also get a 50% core boost. That means Intel's 8-core mainstream isn't going up against an 8-core where you have to worry if the IPC and speed fixes are enough, it means they are going up against a 12-core. You still need IPC improvements and speed, but with improved logic density, Intel doesn't look so hot.

Here is Intel's updated information on 10nm: https://www.overclock3d.net/news/cp...ustry_s_first_shipping_64-layer_tlc_3d-nand/1

Notice how they left out GF on the 20/22nm and the 14/16nm (even though GF is licensing Samsung's 14nm). Then notice IBM's process is missing. Now, GF is skipping 10nm fully, whereas TSMC does still have it on offer, so don't read into it there. Other calculations show a large increase for these manufacturers at 7nm, which is really what Intel is competing against. There is also a largel likelihood they killed cannonlake and updated to use 10nm+ numbers, which is fine, but it also should be mentioned along what the market will be using when that process is released.

The Z390 platform will be 14nm. The 10nm+ Ice lake is scheduled for late 2018 or Q1 2019. So, unless you plan to get the 6-core, you are buying a DOA platform that is worthless in about 6 months. Sorry, I cannot see waiting 9+ months for something that will be outdone in 6 from there by the same company. A refresh, fine. No problem. But, a 10nm+ chip, yeah, problem. Just giving you my view on this.

 

i'd like to disagree. since intel's current 4 cores already has higher IPC, intel 8 cores would be faster than current ryzen. if it stayed on 14nm++ then it'll be decent silicon to overclock, if its 10nm+ then there might be IPC increase and some power efficiency involved with the die shrink. it is directly made to compete with ryzen 2, which even if ryzen 2 were to get 10%+ IPC along with +500mhz overclock, it'll still fall somewhat behind intel.

 

No, Intel WILL NOT put it on 10nm+. That comes later and is the replacement. Also, transistor performance isn't set to outperform 14nm++ until 10nm+ or ++ (I'm betting the latter). So, the 14nm++ 8 core is competing against the 7nm Zen, made on IBM's process. Sorry to tell you I think you'll be surprised if you wait a year.

Either way, I don't care. I actually laid my bet. I'm typing on it right now. Do what you want, but I think your math is off.

 

doesnt really matter what they do, i explained if its 14nm++ then we can overclock well with the 8 cores, if its 10nm+ theres power efficiency and IPC increase. im simply listing out the benefit in either scenario regardless of what they choose to do.

7nm first gen ryzen will prob clock like junk, just like ryzen it is right now 4ghz with a heck alot of voltage and hits a 4.1 or 4.2ghz frequency cap.

 

Also, what is the actual IPC comparison of Broadwell to Kaby - about 10%, tops. What is Skylake-X chips getting on OC with and without delidding? 4.5-4.8, mostly with custom loop. The IBM process, at max utilization, is supposed to bring about 40% transistor performance if power is left the same. Your math suggests 22% improvement from overclock (12.5% on the 500MHz, using 4GHz as the frequency and 10% IPC). So you are saying that AMD cannot optimize better than 50% of what the process improvement is capable of? That literally seems like a lowball there. I'd guess 65%-80% of the process improvement is where it will wind up (higher than that and I'll be pleasantly surprised). So, please check the math first. Also, that means Zen 2 will have HIGHER logic density than Intel's 8-core. That alone is a hell of an achievement, one you fail to recognize.

Also, the 7nm IBM process, for server chips, is targeting 5GHz. This is with 48 core chips. So, once again, check the math. Plus, they said Kaby was so good. It uses less voltage, but was as hot as Skylake. So you are making assumptions without basis yet on the clocks. Good job.

Edit: The reason I know it isn't 10nm+ is Icelake is the generation that server chips get pushed to the front of the line. So releasing 10nm+ on mainstream already goes against their announcement. Then again, it is hard to trust anything they say recently with moving targets on releases.

 

- you simply cant use IBM and compare it like ryzen, who knows what IBM goes through with their CPU prob asked the silicon to be binned repeatly and select the top percentage.
- right now ryzen ipc is about haswell if not a tad slower, so that would make it just under 10% slower than skylake.
- AMD's achievement isnt solely on their design, i dont care if it has higher logic density just give me a faster cpu. what AMD's achievement is that despite having less than a 10th of intel's budget they still manage to do something amazing.

lets just say ryzen is on par with haswell for your argument sake, since you wont take it less than that rofl. skylake is around 7% faster than haswell. so if i have got my CPU running at 5ghz on 6 cores, that would be roughly 5.35ghz if its ryzen. lets not talk about 4.5ghz, average binned ryzen chips cant even hit above 4.2ghz.

 

Actually, you can. IBM's process is what is licensed by GloFo for 7nm. IBM has a record of producing 5.5GHz server chips. Did I say it would meet what IBM does? NO! I said it uses their process. The architecture utilizing the process is up to the chip designer, which is AMD. This is why I gave a range of utilization of the process, which is 65-80%, which seems fair. Last time, with Ryzen, they used Samsung's process. How well did they utilize that? The target for server chips on that process was 3GHz. What did they achieve on the highest chips in those slots with 32-cores? A 3.2GHz boost. I think the all core is 2.7GHz, 10% below that mark. So, they are actually in a specific range. If you get above a certain percent utilization on the process, that means you will get good results. Now, if on the speed target they are, lets say 20% off of target, the server chips hit 4GHz. That places the TR and mainstream at about 4.7, potentially. You then add in the IPC boost. This makes sense, not being able to utilize the IBM process to max potential, but still a gain above your faulty math.

- Actually, the IPC is Broadwell, not Haswell, look again. Seriously, it makes it look as if you have paid attention to nothing in this thread or the world. Everyone has said it is Broadwell level, except for people like you in forums that are not using numbers.

Faster isn't everything, but not rehashing. Rest of it I agree with on this line.

Edit: I may be low-balling here myself, the point is that literally 50% utilization is too little when you saw closer to 80-90% on Zen. IPC is more architecture dependent than the speed,even though the speed is too.

 

yes, IBM is specifically asking for binned after binned silicon, no point saying ryzen will be similar as to how much money that would cost just to bring it to consumer level.

ipc is not broadwell lol. ST benchmark proves it, nothing else to talk about here its black and white, no room for wiggle.

 

http://hwbot.org/submission/3649677_meankeys_geekbench3___single_core_core_i9_7920x_4677_points
This shows it taking single stage phase change to hit 4.6GHz on a 12 core.

This is why Intel isn't worth waiting for.

 

Actually, it seems you are using the old ST scores from release, not the new ones after platform updates, if I'm correct. So, you are actually quite wrong.

 

@ole!!! Can you read charts? Something in this chart suggests closer to BW-E performance, but I just cannot place my finger on it. Can you help me out?

 

i want to ask you instead, can you read charts? do you see anywhere in the graph tells you what frequency they are running at? cant be that blind.

besides, i am suggesting IPC, you are suggesting performance, at w/e clock it runs at in that chart, just no.

 

Then provide your evidence or hold your peace. You provide nothing to even match your assertions, but spout anyways. GIVE THE PROOF. This is at least better than your talk.

Edit: BTW, few have done any true same speed IPC tests.

 

nice ipc u got there bro, i cant wait for mainstream 6c or 8c from intel in my next laptops
https://www.pcper.com/reviews/Proce...d-Zen/Clock-Clock-Ryzen-Broadwell-E-Kaby-Lake

honestly, even without looking at the link i provided, if you go by graph anandtech provided you can roughly calculate that ryzen is behind broadwell which is only 2% faster than haswell. that should be ryzen by haswell if not slower.

 

There is always something better around the corner. This could be said of either camp. You could easily spend forever waiting for the next thing, then the next and the one after that.

 

First, you admit to not looking at the link. Second, it does not include Haswell-E in the list, meaning you cannot compare it from another company without the same adequate controls and settings.

Third, There are many other factors in this, plus, do we need to rehash that IPC varies by task and compiler?

 

I just thought of how to put this into perspective.

Intel's 6 core still falls short of Ryzen 8 core in some tasks. It will be 9 months and a new chipset in order to use the 8-core chip, without an ability to use IceLake on the socket as it is FIVR, meaning no chance for compatibility. The Icelake chip will likely be released about 6 months later. So, while people could be enjoying their Ryzen 8-core over the period, you will be waiting for the mythical 8-core for 9 months, followed by quick obsolescence, requiring a new motherboard, while people that went with Ryzen can just update to Zen 2 7nm in that time frame between the 8-core and IceLake. And this wait is worth while? I'm very confused at that statement. I didn't wait around for Zen. I'm not waiting for Intel. I had my 6700K rig. I just went to a TR build. I won't wait for the decent answer from Intel because this generation, I don't see any. Next gen, well we are already starting to bicker at the 9-15 month wait for those chips. See the ludicrous nature of this argument unless you already got something now? Promise of better performance later is there by all parties. Put on your pants and decide if you need it today. But wait on the speculation on speed and performance until the 6-core drops. It is a month away. So hold your skivvies for a moment before the recommendation for people to wait 9 months for an answer.

Edit: In other news, I didn't realize the used Graphics block was coming with the EK backplate. So, for both the block and backplate, $85 delivered isn't bad. No one was selling them anymore, and so I have no choice on black vs chrome.

 

rofl i already know where haswell would sit on that chart and boy oh boy look at where ryzen at 3.5ghz is. so yeah.. nice ipc u got there.

for people who wanting to get new computer i'd ask them to wait till late 2018 for intel's 8 cores because it'll prob be of a tremendous value and performance. given intel 6 core capable of clocking to 5ghz, it'll be close match to ryzen 8 cores at 4ghz and beat it hands down when it comes to legacy software, which mind you is majority of software out there.

yep, and this will prob be my last wait. again all dependent on how the laptop is though. with a restricted bios, decent cpu heatsink allow high enough overclock my options are very limited, rather theres only clevo to go, or the F7 in late 2018.

if it's not good enough, then just dont buy it lol. if i really really REALLY need it i will get it.

 

https://www.pcper.com/news/Graphics...rds-Will-Support-External-Graphics-Card-eGPU-

wow that a single PCIE slot GPU? so nice. decent for gaming and win in terms of other GPU type workload.

 

That makes no sense. If you are waiting 9 months for the 8 core, then waiting for zen 2 to compare makes sense. If waiting for Zen2 makes sense, then waiting for ice lake makes sense. Considering some reviews needed a small custom loop to hit 4.7 on the 6800, you are using delid numbers and binned chips. So, instead of the crap, let's just wait for coffee before making recommendations. Then we'll have actual data to extrapolate. Also, remember, since Intel released data on TR release day, what do you want to bet on fair turn around with ES silicon for Zen2 on that 8-core release! ;-) so hold your horses on recommendations on chips 9 months out. I've already said zen2 vs ice is the main event and to wait for it if possible. I formerly said get the six core and sit on it, but after this project and my build is complete, I'll have to revisit that statement (with six core numbers hopefully out early next month). Until those numbers come out, giving a 9 month wait recommendation doesn't make sense, especially since we'll have real data points in two to three weeks.

Then saying not good enough, then don't buy it. So wait 9 months on dreams, then IF it doesn't do good, wait again. Sounds like you'll never upgrade. What a joke!

 

Do we have any news on which software developers (Gaming and non-gaming wise) started optimizing for Ryzen architecture?
I mean, we got a simple patch from AOTS and Tomb Raider devs raising performance on both of the games after a simple patch by up to 30%... I was wondering if we have any details on other software devs such as Adope, Autodesk (for 3d Studio Max), open source devs, etc. have decided or are planning to release similar patches to take advantage of Ryzen's goodies?

If not, then we're still comparing apples and oranges.
Right now the industry is heavily leaning towards Intel in terms of software optimizations.
I can't find anything on them supporting Ryzen.

 

dota2 got optimization as well. as for software optimization will take awhile, the more cpu amd sells the more pressure and power they have and can put onto developer to adopt it.

well in general, there are bunch of software uses 1 core, they need to take advantage of multi core first before even talking about intel/amd lol.

 

That's accurate.
I do my work in 3d Studio max, which is pretty good at taking advantage of multi-core cPU's.. however, we know that this performance can be radically increased if the devs optimize for Ryzen features.

On an related note: Vega will need same developer support for games.

 

no, what you said waiting 9 months for zen 2 doesnt make sense.

ryzen2, 8 cores, on the optimistic scenario we are looking at CPU 8c @ 4.5ghz with 10% IPC over current ryzen which would be priced around $370 for a binned chip.

continued from above scenario,
intel gen 9 if 14nm++ we got 8c 5ghz at similar IPC with ryzen 2 which has 500mhz boost over ryzen2, along with a built in iGPU and wifi on chipset, mainstream possibly at $370 to $400

iintel gen 9 if 10nm+ we got 8c maybe at 4.7ghz, 2-3% IPC boost from CFL plus built in iGPU and built in wifi on chipset. pricing to be similar.

on the other hand since i am getting it in laptop, mobo cost is less relevant because laptop OEM will charge premium either way. from the looks of things intel still comes out on top a bit and thats assuming BEST CASE scenario from ryzen 2.

 

https://videocardz.com/72680/intel-core-i9-7980xe-early-benchmarks

a more realistic temp, youtube video shows 4.2ghz 18c CB15 multi threaded test 84c temp peak on 1 core. however doesnt mention the voltage nor cooling used though.

 

I didn't say wait for Zen 2, I said if you wait for the 8 core, you might as well wait that little time extra for Zen 2, which then can push it until waiting for Ice lake. Read what I wrote. Sometimes you are as bad as tiller!

Once again, you are estimating about 50% of execution of a process, which is low balling. Don't make me repeat myself. You have nothing to justify your calculation on.

It isn't "if", it is 14nm++. I explained that to you. Tape in started, not tape out. Tape out takes 9 months to market. If you add it up, that means Ice lake will not be ready for volume by then, most likely. Also, no one gives a flying ... about the iGPU!!!!!!

You literally now sound like an Intel sales rep. Not even proper analysis or addressing my statements.

And features, only slightly, while being charged way more than what you should be. But whatevs. Don't respond without hard proof. We can pick this up NEXT MONTH, like I said, AFTER THE 6 CORE IS RELEASED! Only then can we actually have this talk, considering in the real world, without delid, the 6 core and 8 core overclocks on Skylake-X were abysmal!

 

dude you are not making any sense. i listed out all the pros intel has and the pros amd has, and intel looks better. stop be in denial rofl.

 

Not all, I too was lured by the 7980XE but one concern. Power, the thing eats it. The TR is bad but this thing will be ferocious. I am even thinking of, other than benches or when the power really is needed, down playing the overclock here. I had used a laptop for one main reason, it sips power, these HEDT's do exactly the opposite.

 

yea thats a power hog for sure. need 360 rad + delid to have any decent clock at 24/7 usage. tbh though people clocking that cpu like theres no tomorrow, where as AMD its limited by silicon rather than temp but of course at a 20% lower frequency, really makes me wonder if AMD is really more efficient.

if AMD is capable of say 4.5ghz with the same silicon quality we'll prob see voltage hit 1.6 or 1.7v ez and power consumption no doubt shoot way up.

 

So you even research? 7nm AMD is designed for 0.6V to 1V. I've covered this before. Stop talking unless you bother to research.

Source: https://www.anandtech.com/show/1155...nm-plans-three-generations-700-mm-hvm-in-2018

"VDD: 0.75V nominal or 0.85V overdrive"
https://www.globalfoundries.com/sites/default/files/product-briefs/7lp-product-brief.pdf

Edit: But see, contra:
"VDD: 0.8V nominal or 0.945V overdrive"
https://www.globalfoundries.com/sites/default/files/product-briefs/product-brief-14lpp.pdf

 

man all look good on papers. intel has a history some long term decent cpu, of course given the fact that they purposely not making new ones ot milk us which kinda sucked.

remember how often both intel and amd claims thing, such as we will be hitting 10ghz by year 2010 or 2005 something like that, or how good bulldozer was prior to it's release etc.

 

Where the wall is and efficiency are often different things. It is hard to talk where 7nm and 10nm, let alone + or ++ , will leave us without the silicon. so for now please all leave the hypothetical out of the thread. I had let the argument continue to now but there is no end in sight, so I am stopping it here.

 

I know this is financial news but shows market share.
https://www.streetinsider.com/Analyst+Comments/Advanced+Micro+Devices+(AMD)+Should+Benefit+From+Vega+Release+-+Susquehanna/13308294.html

 

this b interesting, http://wccftech.com/amd-announces-2nd-gen-ryzen-vega-launching-12nm-2018/

no more refresh on 14nm ty glofo/amd. we better see even more energy efficiency and performance improvement from this shrinked silicon. i hope they can push intel even further that mainstream cpu will just be HEDT rofl

 

According to the press release, 15% increase in logic density, 10% in transistor performance. But that is on the process side and doesn't include architectural tweaks for the AMD optimization side. So take that as you will. More in depth post about it in the AMD Ryzen thread.

 

yea, thats silicon quality determines the efficiency and how well it can clock which ultimately affects the final cpu product as a whole. look at intel SKL and KBL as an example, KBL is bit better binned so on average they can do 200mhz at similar voltage and power consumption than SKL, which is likely the real benefit of rebinned/optimized silicon within 14nm generation.

i dare say this will probably get AMD to around 4.4 to 4.6ghz ryzen 2, ontop the shrink makes it just a bit more power efficient, this will be nice for laptop with 8 cores. now all comes down to is if ryzen 2 has any design change or optimization from ryzen 1, that would generally improve latency which increase overall IPC.

 

Also, this is Ryzen+, the refresh node. Ryzen 2 will be Zen 2, 7nm.

Further interesting news, the coffeelake 6-core is actually Kaby refresh. It does require a new chipset because of electrical elements, but the Z390 and 8-core is the real coffee. The person couldn't comment on the 8-core compatibility with Z370 nor on Icelake when I mentioned it is rumored to be FIVR. So, cool cool.

 

those are too far out to confirm anything atm, even for ryzen imho. we already knew 6 cores CFL is kaby theres literally no change except for name change, except that its on 14nm++. as for 8 cores, since 10nm is delayed, it can only be on 14nm++ at this point plus i wouldnt want it to be first gen 10nm cause it'll just clock like crap.

 

Repeat of 12nm AMD refreshes:

https://www.pcgamesn.com/amd-12nm-technology-refresh

We were supposed to get Ryzen+ on 14nm+ next year... however, as we now see, Ryzen+ is slated for 12nm instead.

What I find interesting from that little diagram in the article is 'lower supply voltage'.
Now, if 10% performance increase is accurate, coupled with lower voltages... then a Ryzen 1700 + (refresh) should be able to boost to say 4.0 GhZ at lower voltage than a regular Ryzen 1700.

That should boost it's scores impressively - by effectively moving it up a tier (to better than 1800X on stock) and preserving same efficiency (or possibly improving upon it).

I don't know how much of an architectural improvement we should expect to see from a refresh, if any.
Did AMD mention anything about architectural optimizations for a refresh?

I think AMD working on clocking up Ryzen, improving efficiency, and pushing the industry to actually optimize for it (and Vega of course) would be the best we could expect for 12nm.

 

https://twitter.com/IanCutress/stat...io/iframe/twitter.min.html#910583388704329728

looks like the 10nm delay was incorrect so 8core mainstream from intel 10nm+ still slated for 2H2018, or so to speak end of 2018.

 

No, the 8 core isn't 10nm+. That is ice lake, which there are ZERO reports of a tape out. It takes 9-15 months from tape out.

Meanwhile, coffee is an announced 14nm++ process. Canonlake is a 10nm process. Canon will be used in mobile parts, which need low power products. Considering coffee seems to be a full line, it wouldn't make sense to compete against the more powerful coffee other than low power options.

Finally, the Anandtech editor is using data from the end of last quarter during the earnings conference call. The article that said delays did not disclose source. So, yes, it is rumor, but the source may be newer than that call, hence saying additional delays. You will have to wait until the conference call after Q3 earnings to see if they changed or an official 8-K press release before Anand reports it.

 

i want 8c to be icelake on 10nm+, if intel can do that totally worth buying.

 

And they will, it just will be in 2019 most likely. That is why I was going to recommend coffee and wait for Ice or Zen2, but the 6 core is a refresh Kaby that requires a new chipset. So there may be a shift in my recommendation. It also makes me question if they'll use the 14nm+ or 14nm++ process on the six core chip (if they use 14nm++, the recommendation will likely stand, but only if confirmed compatibility with next year's 8 core chip). That would carry almost everyone over to the main fight: Zen2 vs Ice.

Now, 7nm designs from AMD are taping out in Q4, meaning Q3 2018 would be the earliest to see Zen chips, most likely going to Q4 or Q1 2019. Ice lake is taped in, a new way of making chips. "A tape in is a relatively newer term that is reserved for complicated SOCs that require more than one IPs and designs to be integrated. You can think of it like this: the various components of the SOC have been finalized by their low-level design teams and handed over (taped-in) to the top level design team which will basically finalize the top level layout and design of the soc before sending it to the fabrication facility (taping out)."

http://wccftech.com/intel-ice-lake-processors-taped-in/

Here, they say coffee is 14nm+, but many authors this summer and supporting bungled keeping the processes straight, thereby confusing consumers. Also, considering we now now the first coffee chips are Kaby refresh, it may be correct, even though Intel has said coffee will have 14nm++. So getting that information straight is still in the air, although the 8core coffee will for sure be 14nm++.

So, considering almost 3 months ago ice was taped in, we could see a tape out next quarter to Q1. That means ice and Zen 2 will likely be released close to each other. That is why I say that is the main event.

Now, we don't know how the 12nm refresh will go yet, or how it will sandwich or skew timelines. But from my understanding, that is the best I've seen out there....