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

Upcoming Ryzen 2.0 (Pinnacle Ridge) Will Get Soldered Heatspreader-Guru3d.com
If the prognosis remains true, AMD will release the optimized v2.0 of Ryzen (Pinnacle Ridge) processors in April alongside a new X470 chipset. It has now been confirmed that these 2nd gen processors will get a soldered heat spreader, and not one based on thermal paste.

 

Ever processor should have this done, I'm not sure why Intel/AMD think using thermal compound in place of solder is worth saving the couple pennies vs customer satisfaction, less warranty RMAs due to poor heat conduction, etc...

 

Der8auer say solder ain't best for small die (But he sell delidde tool you know ) https://overclocking.guide/the-truth-about-cpu-soldering/. But I'm sure they could use it (they have the best engineers in their stall). And yees, Intel should really get rid of the paste.

-------------------------------------------------------

The Intel SSD DC P4510 SSD Review Part 1: Virtual RAID On CPU (VROC) Scalability-Anandtech.com

 

@ajc9988 @hmscott @TANWare

https://wccftech.com/amd-ryzen-5-2600-pinnacle-ridge-cpu-performance-leak/

up to 15% 2600 vs 1600 DAMN SON. 4ghz vs 3.6ghz is about 11%, so extra 4% comes from else where maybe faster ram? i havent read the article yet so dont even know the frequency just yet. time to read now.

edit: nope 3.6 vs 3.8 assuming no other boost involved, about 5.55% from clock boost, so remaining 9.5% from else where, not bad.

 

That is the result on geekbench from 1/25. It has been making the circles.

 

Yet few paragraphs later...

and then...

This article is a well-paid job to convince people from known fact that solder is great TIM and provides the best longterm solution for best performance and stability..

 

I think these numbers might be considered a bit more 'sedate' :
https://www.forbes.com/sites/antony...-17-faster-in-leaked-benchmarks/#75c889d92169

 

the article contradict itself. for consumer base, solder is 10000% way better than any paste, even when overclocked 24/7 under a temp of say 80-85c.

 

The Author der8our sell as you know delidding tool

 

GTX 1050Ti | I3 3220 vs R3 2200G vs PG4560 | Comparison

 

R5 2400G vs I5 7600K vs R3 2200G


GTX 1080 | (OC) I7 7700K vs (OC) R3 2200G|

 

intel's 8 core is coming soon in a few months, at least rumour says. thats quite a strange thing for intel to do here, releasing 8c 14nm CFL just 8-9months after their 6c mainstream made to the market. only good thing here is that AMD will be out with ryzen+ first 2800x to further pressure intel into releasing these chips.

1 thing im amazed is that even though at 4.2ghz uses like 1.38v 8 core it still runs pretty cool i guess the current isnt needed that much unlike CFL line up, low voltage but high current.

 

If they mean 14nm is good enough for decent 8 core mainstream chips who will kill AMD's newest awful overclocker, why not? And they need a 8 core chips out now. People have waited long enough. If the results from AMD's new R 7 1700x won't be a lot better... I'm sure Intel's new 8 core chips will sell in loads.

 

amd need to have zen2 be very sucessful and needs to narrow the IPC down to that of intel, similar with clock speed at least 4.5ghz or higher.

honestly speaking, most consumer would be fine on AMD but issue here is these "most" consumers listen to tech junkies who review chips and a lot of these go for the best, even though these consumers will never have the best, its rather sad. if these sheeps got brains AMD would be in a far better situation right now.

 

curious to see if AMD is holding back a potential 2800x as an answer to intel's mainstream octacore chip. and if thats the case, what exactly that will look like. judging from current IPC and clock gaps between AMD and Intel, a 2800X would have to be a 10 core chip to hold its own against an intel 8 core.

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

Yeah, I think AMD hold back 2800x due Intel's coming 8 core chips. But not because it is much much better than what we already have seen. Pretty smart move. It would be a disaster if they launched it before Intel.

 

IDK why noone has mentioned this before, but it seems clear to me that AMD didn't offer a 2800x this time because there were too many sku's in the last release.

1700, 1700x, 1800, 1800x, and all of them could be OC'd about the same to the same effective maximum performance.

I think AMD merged the 1700/1800 into the 2700 series this time to remove redundancy.

I wouldn't hope for AMD to deliver a 2800x, it's not needed.

And 10c/20t isn't CCX divisible and wouldn't be needed as Intel is gonna be behind the performance curve of the 14nm++ when moving to the 10nm(+) for the 8c/16t Coffeelake.

By then the 2700x will be price reduced to undersell the Intel 8 core, and AMD will quickly follow up with the 3700x 7nm.

Too far ahead to worry about now, it's better to build a nice 2700x system now, and enjoy it.

 

Agree somewhat, again reason to not now offer a 2800x but down the line if there becomes better available from the process who is to say they do not start offering a 2800x with 4.6 GHz capability. With nothing on the map for now AMD is not committed to getting anything better from 12nm, but if it shows up?

 

What do you mean 2800X is not needed? What did they collect High-ASIC dies for then?

 

could either be a 2800X or reserved for second gen TR

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

Intel Core i5-8400 [B360] vs. AMD Ryzen 5 1600 [B350]

Published on Apr 9, 2018

Spoiler: Details...

Check prices now:
AMD Ryzen 5 1600: https://amzn.to/2HhTVKD
AMD B350 Motherboards: http://amzn.to/2u3kJtc
AMD X370 Motherboards: http://amzn.to/2u2D5ZE
Intel Core i5-8400: https://amzn.to/2v1hqoy
Intel B360 Motherboards: https://amzn.to/2JtfacU
Intel Z370 Motherboards: http://amzn.to/2hNiMxg

Read the written article on TechSpot:
https://www.techspot.com/review/1608-...

Support us on Patreon
https://www.patreon.com/hardwareunboxed

Intel Core i5 8400 [B360] vs. AMD Ryzen 5 1600 [B350]

FOLLOW ME IN THESE PLACES FOR UPDATES
Twitter - http://twitter.com/hardwareunboxed
Facebook - http://facebook.com/hardwareunboxed
Google Plus - http://goo.gl/xx14Uj
Instagram - https://goo.gl/8lhprr

Music By: https://soundcloud.com/lakeyinspired

 

Ryzen 5:
https://hothardware.com/news/amd-zen-5-processor-architecture-in-development-potential-2021-debut

And some news on next-gen GPU's from AMD:
https://pcgamesn.com/best-cpu-for-gaming

 

https://videocardz.com/newz/intel-cascade-lake-x-hedt-coming-at-the-end-of-this-year

@TANWare @ajc9988 @tilleroftheearth @Papusan @hmscott @Mr. Fox

time to go desktop. 28 cores 5ghz binned of the best binned cpu? with 10nm delayed and intel pulling possibly another year of 14nm, this might be even better than 14nm++. although that 5ghz most likely on LN

 

First, Broadwell and Skylake were on 14nm. Kaby and Skylake-X were 14nm+. Coffee (Kaby refresh for the 6 core)/Whiskey (and true coffee) plus Cascade-X are all on 14nm++. So you are incorrect on process used. Next, 14nm++ has clear improvements over Kaby, so your comment on might be better is a truism. The question is how much this will act like the 8700K on the speeds obtained compared to Kaby.

Now, the 28 core seems to have the full hexa-channel memory (or at least one story I read suggested it might), while that story or another mentioned they may be changing the socket from 2066 to potentially the behemoth, as the 2011 and 2066 was barely enough to work above 12-core chips, but they stretched it to 18-core chips. For 28, that is doubtful, meaning they are likely using the 3467pin socket with Purley chipset, and are basically deprecating their 28 core Xeon to the desktop lineup. Now, yields are still good on 14nm, while yields on 10nm are ****. So, basically, without going multi-die on 10nm, Intel has nothing left for next year, which is pretty pathetic, in my opinion.

Now, as @Papusan said that AMD should be scared, I disagree 10,000 percent. There is no proof that the 5GHz was without LN2, afterall, did you see the vids on the closed LN2 with recapture? Even with the 14nm++ process, that would likely need exotic cooling, although likely also helped through the increased contact surface area. AMD, while maintaining compatibility, would only need to put four dies per chip instead of two (doable), disable two IMCs and the I/O for two die, and now you have Epyc for TR4 platform.

All this BS on extra core counts on AMD this year mean nothing and are fueled by ignorance and not understanding the changes coming to the platform with Zen2. Zen2 is when we get core count increases at earliest.

But, aside from this chip being just their Xeon line thrown to HEDT, the chip WILL NOT RELEASE UNTIL Q4 2018. That is literally close to six months from now. That is 3-4 months after TR release. This is why I see it more as a ploy, to see if they can force AMD to use more dies on the TR line and play the bigger penis core game. There really is no reason to play this year. This is Intel's last hoorah until 2021, so I say yank and am really surprised they would shoot themselves in the dick like this. Not only that, the Xeon equivalent is a $4K CPU. If you think you are getting that around $2K ( @jaybee83 - addressing your pricing estimate in the other thread), you are mistaken. They will be charging from $2400-3200 easily for the 28 core part (take, for example, this current listing on price https://www.newegg.com/Product/Product.aspx?Item=N82E16819117904 which is $10K for the 28-core Xeon 8180 Platinum). They've been walking the price up slowly and hate that AMD undercut them so hard. Meanwhile, AMD has shown they can sell a 16-core for $800-900. You do a 32-core variant, you could actually do it under $2000, you are looking at still winning on price/performance. They do not need the win this year, all they need is to wait. Funny, also the time of release, Q4 2018, coincides with the Zen 2 Epyc information. Now, why do they want to do that? Maybe because that is what this chip will really be competing with, not this year's release in August. It is, once again, a dick measuring contest. This is why they want to show 5GHz, as that is what Epyc may bring to the table, large core count (which may grow with Epyc to 48 cores), and to try to hold the line until TR3, praying for a maricle on 10nm, even though they are likely trying to find a way, in addition to going to 10nm, to use 10nm+ next year, which is lateral to the 14nm++ process. 10nm++ is when 10nm is supposed to FINALLY be better than 10nm+, but that is 2020 and going against 7nm EUV refined Epyc 2, which will be a tall order for Intel.

Personally, this is a nothing story meant to try to deter AMD sales in showing off their TR2 items at computex and sales in August. It changes nothing in the overall scheme of things and is not exciting. It is desperate. Good for Intel.

@Deks @hmscott

Edit: 5GHz on phase change. AMD is talking that level next year 7nm without (although, that depends on whether or not they were talking all core boost or single core boost).

 

7nm spec sheet indicates 40% performance improvement over 14nmLPP.
I would imagine this applies to baseline clocks across all cores, not single core.
Besides... 5Ghz on single core doesn't make sense for 7nm which will contain Zen 2 due to too many changes offered by the process.

 

most of what you said is based off what nm it should be on, since intel current line up of chips is all over the place, im not suprised if we get to see some more refined 14nm than 14nm++ thats used in CFL-S especially they are going to further delay 10nm.

also like i said, its most likely on LN or some kind of extreme cooling I dont need to see it to make assumption on things thats clearly not possible. it might even be 2 x 14 core CPU for all we know as 5ghz on that front sounds much more reasonable then again nothing of that sort is shown.

lastly, AMD's IPC and clock speed is a joke, pls fix it before talking about how they are better, besides im not even talking about AMD, i am talking about intel's chip kinda strange u'd bring AMD into every conversation.

 

You can see the second score below as well. 5912cb aka around 24% lower clock speed vs. 5.0GHz. AMD will only be bang for bucks for power enthusiasts if they don't up the core count for HEDT. If you go after most performance, then you all know this don't come without a cost. If AMD don't up their game, it will only means they will be sailing longer and longer behind. Bad enough with today's Threadripper vs. 7980Xe. Same vs. 7960X. Far too many paint Intel as the devil on the wall.

 

after we buy 8 core CFL for our laptop, we go desktop! 16 cores 4.8ghz im in

 

http://www.legitreviews.com/asus-ro...-core-demo-but-look-at-that-cpu-cooler_205889

Looks like this is what was used to cool that CPU at 5.0Ghz, so definitely not LN2 but a huge water cooling system hidden under that table for sure.

 

This means +5555cb in Cinebench R15 if you put all cores equal up to stock max boost clocks (3.8GHz)

 

Wow, Intel is really needing to bring in huge power and cooling to run that 28 core behemoth - I doubt it's worth it. Intel really shot themselves in the foot, standing in the corner they've painted themselves into.

Go AMD ThreadRipper 2 or Epyc 2 and enjoy the calm.

Maybe AMD can help Intel by providing TR2 / EPYC product to allow Intel to resell to the Datacenter market, at least they can keep their sales and support staff busy...

 

https://pcgamesn.com/amd-threadripper-2-moar-cores

It seems AMD is also pumping up their power requirements for TR2. TDP is up 40% according to this.

 

It's all part of the power up game, but I think Intel might continue to heat up the power and thermal exposure category as long as they are stuck on 14nm+++.

Price, power, balance of both is important. It's not all about the straight-line performance, she's gotta handle well in the curves too.

 

yup yup, 32 cores confirmed, its official now: https://www.anandtech.com/show/12906/amd-reveals-threadripper-2-up-to-32-cores-250w-x399-refresh

let the core wars continue

now lets see: 1950X at stock gets around 3000 in CB15. overclocked on water u can get up to 3500-3600 @ajc9988 confirm or correct pls based on personal xp

so double the cores and we get around 6000 in CB15 with overclocks getting 7000-7200. Naturally, scalability is not perfectly linear but a bit less. however, this will again be offset by slightly higher clocks and IPC on 12 vs 14nm so im thinking these scores should be right on the money.

soooo....considering around 7200 points of TR2 overclocked, NOW its NO surprise whatsoever that intel chose to oc its 28 core behemoth to 5 ghz to BARELY beat the TR2 with +134 points = +1.8 %

 

So, 1950X at stock gets that, and you can run for an all core boost of 3700, depending on cooling. That score is achieved using 2400MHz ram, also, IIRC, as a friend achieved that score with stock and 2133, but pre-lauch and reviewers saw as low as in the 2800s.

The new Ryzen 2000 series achieves around 150-200MHz all core OC higher than the current gen. The current Epyc 32 core has a max boost of 3200MHz and a TDP of 180. So, considering the TDP is being raised by 70W (39%, approx.), we can expect more power draw and at least 3200-3600 on all core OC, tops. So, if a 16-core@4GHz gets around 3500, if you double the amount, then multiply by 0.8 or 0.9 (to give the max amount with the 10-20% drop in all core OC), you get 5600-6300. That may need further reduced due to the two extra dies not having IMCs enabled, which leaves more room for latency, which may remove any latency benefits of the updates to Ryzen 2000 series chips.

As such, and considering the 3.8GHz results on the Intel 28-core chip, we are looking at Intel sitting at the low end of what this chip may do, if the IMC's improvements are still in play and 3200+MHz ram is used for testing, of course. You have double the L3 cache with this chip, which is nice, but the latency will still exist as each memory access requires jumping to one of the two dies with the controllers (also why the use of a control chip being rumored on Zen 2 7nm is so important, as it means all must go to the same area, but can increase the amount of L3 even further due to what limited items need on the controller chip, which also cuts the space for L3 out of the actual dies, which could lead to adding an additional CCX, giving credence to the 48-core starship rumor. To do 16-cores, which is 4 CCXs, per die, you would likely want to wait for EUV. The other way to do it is tying more separate dies to the controller chip, but keeping the same number of CCXs per die, just having them much smaller in size. But that would be up to 8 dies on a 64 core Rome variant of Epyc. Not saying it couldn't happen, just that it is a lot. But, the dies would be like 80 sq. mm. each (more if more CCXs are present), so doing 4 is 320+ sq. mm. of die space, excluding the controller chip, or double that for an 8-die design (640+ sq. mm.). Now, before fully scoffing at 8-die solutions, think of the controller chip in the center, then arranging all other dies equidistant around it in a Brady Bunch configuration (the 9 squares at the beginning of the show). But, regardless of the countless ways to accomplish the goal, what matters is that it gets done.

Edit: turns out the samples ran at 3.4GHz for the 24 and 32 core variants and that 3.4 on 32 core on the sample was shakey. That places a better chance of around 6000, which is a clear win for AMD over stock values on Intel's chip. Further, this is with an air cooler.

Edit 2: http://www.hwbot.org/submission/3819103_gerkin_cinebench___r15_ryzen_threadripper_1950x_2996_cb

 

why so conservative though on the TR2 clocks? if we argue that since TR2 will use the same dies as Zen+, then we might as well say that the same overclocks can be achieved, so around 4.2-4.3 Ghz on good chips

 

So, to counter that, I offer up the GF docs on the 14nm process. On it, they said they were targeting 3.0GHz on the HPC/Server side. You had all core boosts in the 2.8-3.0GHz range and 3.2GHz max boost. Here, they said they were targeting 5GHz for HPC/Server, placing that as the single or all core boost speed, most likely. The consumer chips, both mainstream and HEDT, reached about 700+MHz on all core boost over the server chips. You add all of this up, it means that all core boost will likely be targeting 5GHz+ on consumer products, at minimum. With the additional cores, it may be a push on the server side. This is why I worded it the way that I did and predict the speeds that I predict.

What process node is being used is published. I will not argue over published information from Intel on this topic.

Also, before the mesh, on the same die, for high core count and extreme core count, Intel used to breakup the cores into two groups in straight lines and used two rings. Likelihood of multi-die using EMIB is possible to bring down costs, as that is a $1400 product, times two, plus assembly, brings the cost to around $3K, which is likely the best deal you might find on the 8180 server chip (not purchasing retail).

Then, on IPC, you are a joke for not seeing them being at Skylake levels this year, almost Kaby levels (still lower than Coffee) on IPC. You are being blinded by bias. On clock speed, that was already discussed as happening next year, which coincides with the year Intel has no answer to AMD. So, that being the case, read up on public information and inform yourself.

Why do I bring up AMD? THIS IS THE AMD/INTEL COMPARISON THREAD. Maybe that is why we are comparing offerings from the two companies. Imagine that.

Once again, chest thumping and dick measuring. I back worked this score of 7334, which suggests a score of 5867.2, or about there, for an all core OC of 4GHz. Considering the 32-core I estimated to hit around that same level using the 3.4GHz all core boost expected, we are seeing equal performance on 15% less speed at stock, with 4 extra cores, which is about 14.3% extra cores. Looks like a push, unless using extreme cooling solutions. We'd need to see average OC values with what cooling to really get an idea between them. That means we are looking at more I/O on TR2, but potentially hexachannel instead of quad channel memory on Intel, so Intel may take back the bandwidth wars in a large way once released in around December.

 

Great question. It comes down to die numbers. Instead of only two dies (with a very large surface area to cool), you have 4 dies, so halving the effectiveness, approximately, of the cooling. You increase TDP as well. So, because of the two extra dies being operational, even though you do not have the additional IMC and I/O heat on those two dies, you still have the primary source for heat still there. As such, and pointing out the Anandtech article you posted said explicitly 3.4 all core boost on the samples, it follows perfectly.

Even with the 4.2GHz on Ryzen 2, and seeing similar clocks as ryzen 1000 on TR 1000, you then must reduce based on the two extra dies putting out around the same amount as the original two dies (no I/O and mem controller means needing some less, although IF temp effect is there). As such, reducing the max clocks by around 20%, along with the efficiency of the Zen uarch, suggest this gets the heat within the additional 70W (instead of putting out a 320-360W chip pulling over 400W when overclocked, which may not be that far off the OC wattage on these chips anyways). Remember, every number is a clue! That is how I make logical guesses on products.

 

yes of course, as long as we talk staying inside the stock TDP envelope of 250W and being temperature limited, those clocks make sense. i was rather talking about the actual silicon limits though. with two 360 rads on water and a strong mobo with sufficient VRM phases and chokes and a strong PSU this shouldnt be a problem though after all, as the mythbusters used to say: if its worth doing, its worth OVERdoing!

edit: btw, when i posted the anandtech article it wasnt updated yet with spec details on clocks theyre still updating the article as we speak.

 

I'm playing within the knowns. Ryzen is efficient, but HOT (not as hot as Intel's recent chips at times, though, but close). Once you pass the curve on increasing voltage, heat goes up drastically. Even with good cooling, you are now dumping twice the heat energy into the same surface contact area of a cold plate. Double heat, but more evenly distributed, which helps to a degree. Either way, the water can only transfer so much energy relative to the contact surface, meaning it will still be a large amount of heat on the product. Rest you know, like flow rate, cooling capacity, ambient temps for relation to heat dissipation, etc.

As such, expect 3.8GHz with good cooling, tops, is my guess. That for doubling the core count from 16 to 32 is nothing to sneeze at! Also, adding 200-300 over the all core OC is what we saw on TR 1000 series, so, going with 3.6-3.8 is not unreasonable for the product. 3.5-3.65 will likely be what most see, though, without exotic cooling solutions, is my bet.

No matter how much you overdo it, the laws of physics are a ***** at times!

 

Yeah, they seem confused on the L3 on the expected specs as well. What likely is happening is having two different production lines, one with only two active dies, which is the 8-16 core variants, then another line with the 24 and 32 core variants with cores disabled on the 24 core variant. Even though they added the CCX block disable feature, my thought is not enough overhauling occurred to not use 4 dies for the 24 core variant. Considering that the L3 cache on 6 core variants accesses the amounts of L3 of the disabled cores, it means the full amount of the cache, 64MB of L3, should be available unless they are doing CCX level disabling rather than the mirrored core disabling. Also, with the addition of binning cores (I showed in the RyzenMaster utility the new versions tell you what the fastest single core is per CCX and the second fastest), they likely used that to disable not just damaged cores, but, if not damaged, turning off either the lowest performing or second lowest performing cores of the relative CCXs to provide better performance on the 24-core variant than they would have achieved otherwise.

 

so lets see:

- TR1 3000 points on 16 cores and 3.7 Ghz all core clocks. so thats 50.68 points per core and Ghz
- TR2 32 core at 3.8 Ghz OC would be 50.68 * 32 * 3.8 = 6162 points in CB15
- Intel 28 core monster at 5 Ghz scores 7334 points, so thats +19 % higher score
- In comparison, we have around 3699 points on water for 1950x vs. 4786 points for 7980XE on water (averaging top scores on water on HWBot), so thats a + 29% score for Intel

Looks like AMD is catching up, and fast

that is a quite interesting thought actually! so we could see even better overclocks on the 24 core models, even taking aside the inherent advantages of having 8 less cores to deal with vs. the 32 SKU.

 

naw those 250w TDP thing dont matter anymore. if you're grabbing one of these 24-32 cores cpu and overclock the hell out of them, only the highest end 360mm water cooling or phase change cooling case will help. getting like 24 cores cpu only running at 4ghz is actually very disappointing to say the least.

 

well im staying optimistic in terms of sufficient thermal headroom for reaching 4.2-4.3 Ghz on both the 24 and 32 core TR2 models, albeit with like 2 360 rads or maybe a water chiller . if users are able to reach 4.4 Ghz on a regular AIO with an 18 core 7980XE (see silicon lottery binning, they test with run of the mill configs, not super high end to better represent typical usage scenarios. plus their testing is very rigorous, involving Prime95 so less stable overclocks should be higher than that))

and

we assume that we lose about one multi max OC for every 2 cores added to the mix (see silicon lottery range again)

we could be looking at something like this:

7980XE 18 cores = max OC 4.4 Ghz
Intel 20 cores = 4.3 Ghz
Intel 22 cores = 4.2 Ghz
Intel 24 cores = 4.1 Ghz
Intel 26 cores = 4.0 Ghz
Intel 28 cores = 3.9 Ghz (--> this makes it clear how a water chiller was absolutely necessary to reach those 5 Ghz at computex. also consider that out of all the CPUs they tested they had this ONE SINGLE sample that was able to do 5 Ghz in a CB15 run....)

as for TR2 the matters would get a bit more complicated because so far Zen seems to reach the silicon limit before it hits the thermal one. however, since the chips offer a much larger cooling area and the heat is dispersed over four dies instead of just one monolithic die with Intel, id say at same core count and identical clocks they would generate less heat due to the overall larger package. so following that logic and continuing the above-hand calculation, with Intel at 32 cores we would end up at around 3.7 Ghz max OC.

So i would expect the max doable OC with TR2 to be above that.... 4 Ghz 32 cores, here we come!

 

This is absurd! First, TDP, although calculated differently from one company to the next, is a measure that matters because it still gives some guidance on cooling requirements, etc.

As to the CPU cooling requirements, 360mm was already known to be needed with the first chips with 16 and 18 core chips last year for consumer. But you are misleading that people should run out to pay for phase change, adding significant costs to the rig and running the rig. THAT is absurdity! These are not gaming chips. Get over yourself. It is as bad as your arguments a year ago on Optane. What did Optane turn into? 1) a way to increase memory capacity on server side memory forthcoming, 2) as a cache drive for a large hard drive, and 3) for professionals with large enough page files that they would be significantly slower than if not present (such as 8K video renders).

Third, NO REGULAR CONSUMER SHOULD EVEN LOOK AT 24-32 CORE CHIPS! Sure, they seem fun, but these are made for heavy multi-threaded software to run things. Most consumer products are not heavily multithreaded and will not be able to use them effectively, except for benching. Very few people multi-task to that level either to max them out. Now, if doing blender renders, V-ray renders on CPU, etc., have at it as these CPUs will be awesome for that (if using CPU rather than GPU to do the task). I love the concept better than I love the chips themselves on this high of a core count, which is why I keep on bringing up the dick measuring. They are just throwing more cores out there, but software hasn't caught up. But, your comment on "getting like 24 cores cpu only running at 4ghz is actually very disappointing" is just ignorant ********. Seriously! Already, we have seen that Ryzen with the same core count at lower multipliers able to outperform slightly higher clocked Intel chips on heavily multithreaded workloads. Most server software that is made for multithreading doesn't need speed, it usually is limited by cache or memory bandwidth and is made to take advantage of extra cores, even at slower multipliers. The fact you still think that speed matters at this level to the degree you do shows your lack of understanding of where the industry is. Once software vendors create products that can perform n-core scaling nearly perfectly, then we can talk raw IPC and speeds. Instead, you are talking like these offerings from Intel or AMD are mainstream offerings where single core performance means something. It shows a lack of understanding of the products, the product space, and how to use the product. If you did, you'd understand using phase change cooling to reach 5GHz while the CPU is around 30 C is really pushing the limits of the phase change if a single stage and is pushing voltages high as hell to accomplish it is nothing more than a PR stunt. Good job at missing the obvious. Wait for the huge price tag coming down the chute. My guess is that you would be the mark/rube to buy the 28-core and not actually use it the way it is supposed to be used, even if the costs were $3000-4000 for the chip.

You can be optimistic all you want, the cynic in me (plus the samples having all core clocks of 3.4) is what leads me around to what range I give. The 7nm talk is where I have high hopes for single core performance and heavy multi-tasking. But, we need to tell these software vendors get off their asses and find better ways to utilize parallel processing because the core race is only starting and they are what are preventing full use of our hardware. I thought I would need to lodge that complaint again later this year to next year, but I have been proven wrong and the time is already again upon us.

Your calc on overclock seems about right, although the use of the new, larger, socket may offset some of those calcs. Remember, TR had 40% surface area, approximately, over the 2011 and 2066 sockets. Applying that for cooling capacity to their new chips, which should be like their server chips which are about the size of the TR chips, you could see a slightly different scaling. Also, all Intel fanboys should be thanking AMD on the cooling side, as TR is what designed the layout for the cooling plates that are to be used on these new Intel CPUs for the consumer side of the equation.

As to silicon limits, that is BS, especially for second gen ryzen. AMD for first gen hit a voltage wall. To scale it put out too much heat without exotic cooling. That means the limit was heat, not silicon, or a blend. With second gen, it becomes more obviously heat as the voltage walls changed and more people saw higher clocks and are hitting throttle temps rather than having to stop right at the voltage increase due to damage caused by that increase. Listen to buildzoid talk about this a bit on his youtube channel.

Now, even though you are correct on die size being monolithic on Intel, the actual package sizes will be about the same between the two, meaning the surface contact areas, cold plates, etc., will be roughly the same this year on. So there is less advantage there other than the cores being more spread out rather than concentrated on the Intel side. This means Intel's max OC is subject to slight variation, although the math should be roughly correct on max core count multiplier, especially since sometimes it is 200MHz between instead of 100MHz. But, point is still taken.

But this is why I do not have hopes for higher clocks than Intel yet. Nothing I've seen suggests that, although notice the narrowing of the gap on frequency at this high a core count. That is the monolith and densely packed cores versus the cores being spread among four dies which disperses heat more evenly. That alone needs heralded, especially with the fact that SMT works better than HT on heavily multithreaded workloads, which practically erase the gap.

 

So the 32 core chip is the Epyc 7601 cut down with Ryzen 2 improvements? That chip has an all core boost of 2.7Ghz. So at most were looking at 200-300mhz improvement over first gen giving you 2.9-3.0Ghz all core.

https://www.newegg.com/Product/Product.aspx?Item=N82E16819113471

https://www.anandtech.com/show/12906/amd-reveals-threadripper-2-up-to-32-cores-250w-x399-refresh

These chips have lower base and boost than first gen Ryzen TR. It shows boost frequency of 3.4Ghz and that isn't even an all core boost. Where the hell are you guys getting 4.0Ghz+ all core boost pipe dreams from? I'm willing to bet this thing does 3.0Ghz all core at most.

Also we didn't see AMD roll out Cinebench this time around.. hmmfff lol.

 

Did you even read the articles you posted? I agree 4GHz all core is a pipe dream (READ MY ESTIMATES AGAIN). But you are sounding real ignorant on the other end of bias. Do you see it yet? Such as the 32 core Epyc having octochannel memory and I/O on every die and a package TDP of 180W, whereas this has a TDP of 250W (70W more) with 4 fewer memory controllers and half the I/O of Epyc. Now, if you want to have a reasoned, logical discussion, lets have it. If you are going to spew biased ignorance, I will treat you as such. So, go reread your own sources before commenting and reread the thread.

 

LOL Ok buddy calm down and take your happy pills. No reason to get so angry or upset or something so trivial.

 

My point is I've seen you come to the AMD thread to thread crap, then you are coming here and spewing that bias and literally contradicting your own sources on information. I do not suffer fools well, which is obvious to anyone in this forum who knows me well.

Now, there are some advantages to the Intel chip, some to the AMD chip. I really want to dig into those, but the news is so new that no one is ready for that convo yet. It will happen though eventually.

I have even stated or hinted at each of Intel's and AMDs benefits and drawbacks. Another aspect is that AMD will have their product on shelves in August. Intel will have the 28-core on shelves in November or December. That gives AMD a couple months lead on the market. Intel's Phase change cooler running at 30C would damn near burn it out, considering phase change is made for -50 - -30C, and most target 0C as the driver. I need more info on that cooler to be honest, but could see either company run a phase change to the point of burning it out just for a short demo.

On Cinebench, it is telling, but also could be to prevent fanboys from comparing exotic cooling scores with the air scores on the TR2. Now, I did mention there was a potential latency on the two new dies. What I didn't explore is that AMD has a 90% scaling over a single die (similar to how we used to talk about scaling on multiple GPUs with SLI or Xfire). We do not know if the numbers I gave may need scaled down further, or if the doubling of the two die solution is enough, which already built in the scaling into the score. We also do not know what having to always go off die to fill the cache will do on latencies and how that will effect the compute. Those are things I really want to know.

But I think you get the point. I'm asking for thoughtful discussion, not just throwing out crap for the hell of it. Let's keep the discussion at a higher level and the company bias to a minimum. When we do that, you get to hear more of my solid critiques on AMD and hear my praise on what Intel does right more. Otherwise, I wind up sounding closer to an AMD fanboy and I do not like that as I am company agnostic (yes, I root for AMD to do well considering they are the underdog, but that isn't in my analysis usually).

 

TBH I wish they would have targeted a TDP of 320w or even 350w. I know it would be hard and still stay on air but few here do anyway. The more important point is being missed here.

This means TR 7nm will be at least 32 core and maybe even 48 core.

Edit; as far as CB-r15 at stock 2950 for 3.4 on a 1950x is about right, so 5200 or better should be stock on a 12nm.