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

@tilleroftheearth
AMD CPU Sales Overtake Intel For The First Time In Over A Decade At Germany’s Largest E-Tailer-Wccftech.com

It was on time that Intel now got a nice kick in the ass(They have to up the game. Not only give better performance... Prices must down!!). Damn pity, nvidia isn't being in the same situation

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

BEWARE: Latest Gigabyte X370 Motherboard BIOS Can Permanently Damage AMD Ryzen CPUs, Sets Voltage Up To 1.7V-Wccftech.com

 

I'm not sure if this has been posted but its an interesting read comparing gaming performance of the R5 1700 vs i7 7800x.

https://www.techspot.com/review/1450-core-i7-vs-ryzen-5-hexa-core/

 

https://www.overclock3d.net/news/gp...s_emib_technology_-_multi-die_cpus_incoming/2

Don't know if this was seen here, but aside from the implication Intel is also planning multi-die chips comes them abandoning the talking point they have a true node jump, as some components will be on a larger node. As such, they can no longer continue to berate half nodes moving forward, also signaling problems in miniaturization, which is apparent to anyone with eyes.

 

We discussed this to death, but it is a good read. Some discussion got deleted, for good cause, but it has been discussed. It's worth reviewing what we said regarding it! Also search hardware unboxed, which is ran by that author!

 

Yooo mean Intel will use glued together dies package? So the Blue will follow AMD? If so... Talk about **** in their own nest

 

They've had this planned for awhile and filed the patent awhile ago (this was announced this spring). Most "blue" it off as them improving things, like with the mesh, but the multi-die part is new. I just have to laugh at people saying Intel is so far ahead or Intel abandoned multi-die because they were a dead end, then Intel blasts AMD for it, then does it again and tries saying their solution is better than AMD's, which was a ground up redesign of hypertransport. Lol...

 

Intel abandoned it as at the time it was no better than the BullDozer in performance. The tech just had not caught up to the format. It seems with Ryzen the multiple CCX's have gained new life.

 

To me, this gaming example is a really good way of showcasing that a lot of testing software reviewers use is pitiful when it comes to handling multi-core chips.
Most software that tests CPU performance for example in a variety of benchmarks (I'm looking at those programs that compared the intel newest 10 core to 16c Threadripper) showed that performance is consistently higher on CPU's with smaller number of cores.
The programs simply don't know what to do with the cores in question and performance ends up lowering on CPU's with higher cores (unless we're talking about software specifically coded to take advantage of multi-core chips like 3d studio max, blender, etc. - handy for me since I use that software, but not so much for Adobe programs etc.).
Most browser tests also aren't optimized for many cores - they can handle 1 or 2... possibly double that amount, but any more than that and things start choking.

Adding to that that majority of pro software still isn't optimized for Ryzen and Infinity Fabric.
This kind of optimization will likely change in AMD's favor when Ryzen is more widespread, but it will obviously take a bit of time.
To date, we've seen some game developers put out patches for Ryzen, but they were small updates to be honest... they likely aren't taking full advantage of its new instruction sets (unlike Intel CPU's) and that would probably be seen more in newest games, as opposed to older ones.

Software definitely needs to evolve. It's way too slow.
Throw AI into it capable of rewriting its own software... heck the algorithm could probably analyze the CPU architecture in detail, then rewrite the software (game or rendering program) to work most optimally on that specific hardware.

 

https://www.overclock3d.net/news/cp...ly_delids_intel_s_12-18_core_skylake-x_cpus/1
So, no solder on the 12-18 core chips ( @Papusan ). Seriously? LMAO!

 

Strange inded, because Amd's 16 core use solder. But don't forget that AMD's biggest cpu's are glued together I think one of the reasons for thermal paste instead of solder was that Intel tried very hard to decrease their cost due the price competition vs. AMD IN Norwegian... Men som du vet, det an å spare seg til fant (An old Norwegian adage sounds "to save to found")

"It requires competence. It requires courage. It requires the will for innovative thinking and innovative problem solving. It requires that you dare to do the basic analyzes and do nothing for granted."

 

wow looks like no bs.. though its Der8auer he might have gotten golden chip thats similar to top end of silicon lottery, it is still possible. http://cdn.wccftech.com/wp-content/...3DMark-Firestrike-Physics-Score-1480x1147.png

if cpuz shows correct, hopefully its under load of course and no bs, 5ghz 18c.. which should only be possible if die area is that big for dissipation, its literally the size of 3x 7700k's bare die surface area.

still though, too good to be true i gotta wait and see the review to believe it.

 

10% higher OC but +12% fewer cores Could be

 

Delid a $2000 chip, wow. Have to say though that is one powerful chip.

 

well its prob worth it as long as chip doesnt die. this is probably the best intel has given consumer so far with access to some half decent silicon.

 

I do distinctly remember der8auer saying that deliding Skylake-X chips is pretty dangerous since lots of small caps are located around the corners. I personally wouldn't try it. I mean, the risk-to-reward ration is kinda hard to quantify.

 

This is why people are waiting for a delid tool or being able to buy one delidded from SL.

 

Check the pics of the 18-core model. Some of the caps are literally covered in glue and basically "attached" to the IHS. I think delidding that one may be nigh impossible.

 

This is why paying someone else to do it is more efficient than risking it yourself.

 

I wonder how much people will ask for risking it themselves. 4000$ for an 18-core PSU delidded? xD

 

It makes a bit more sense now. AMD is claiming more head room with Ryzen2 a nd possibly better IPC. FS is much better attuned for Intel but I just do not see Ryzen pulling 5.0 GHz out from the next gen and the much needed IPC boost.

In hindsight now I see why TR did not get stepping 2 yet as it would have been that much less for the next CPU steps. I would still like to see the CB R15 for that 7980XE.

 

I mean, 5GHz maybe not but 4.7-4.8 may be possible. A friend of mine is currently toying with his 1700 (not sure if it was a 1700 or 1700X but yeah) and got 4.1 on 1.4V. Told him to try 4.2 on 1.45V but he won't push the voltage.

 

http://hwbot.org/hardware/processor/core_i9_7960x/
Finally have results from Intel's 16 cores 7960X ticked into the Boot. Aka The actual competitor for 16 cores Threadripper (1950X)

Sorry, forget it. I thought of 7940X

 

Thanks, but no interest in LN2 results.

 

Actually, Stepping 2 was meant for EPYC. According to the one article, TR was not originally planned. They needed all the dies possible for EPYC that they could get, as there are 4 dies per chip. So, out of the two lines, pulling the top 2-5% binned chips from Ryzen makes a lot more sense, as it doesn't effect your server line at all (which is where they want the most penetration).

As to speed, the GloFo 7nm document shows 5GHz target with 40% improvement (combined IPC and speed) on the transistors. Implementation may be less than optimal, but if you go up from 4-4.2 to say 4.7-4.8 (about 10-20%), plus 10% IPC, you see 75% of the optimal AND a damn competitive product to Intel. By Zen 3, everyone has EUV lithography, so the chip costs will be lower and better tweaking can be done.

 

We shall see. The original TR 1955 to 1998x I believe were all stepping 2. It appears though they walked away from that back to stepping 1. Be it for Epyc or again to save some for the next iteration of TR. That was one of the reasons I wanted a TR, stepping 2. Without that the 7980XE sounds more appealing TBH.

 

That's what I thought, ThreadRipper are Stepping 2 like Epyc.

 

honestly speaking, that screenshot is from korean forum and we can probably take it as fake for now. even if other people believe it, i wouldnt. given HEDT are failed xeons, their silicon quality wont be that much worse and quite frankly, the only thing that will make me believe 18 core at like 4.7ghz+ possible at a decent temp/voltage would have to be that intel used 14nm++ instead of 14nm+ in their HCC chips.

if intel specifically used 14nm++ with best of the silicon, ontop of that the chip is cherry picked, i'd believe it just might hit 5ghz on all cores but as always cherry chip is 1 or 2 CPUs out of the 20000 they sell, meaning even if we were to purchase the best directly from SL, it'll still be a level or 2 lower.

 

UV and OC Vega 56

 

Intel Chipset Roadmap Confirms Z370 in Q4 2017, Z390 in 2H 2018 – Mainstream H370, Value H310 and Corporate Q370, Q370, B360 Chipsets in Q1 2018 Damn, Intel have hurry

 

Wow, imagine just how "ready" that stuff is all going to be hurried to market and jammed into vendors production schedules to make it out in time to disrupt AMD.

"The most interesting thing about the roadmap is that it is the first public confirmation of the Z390 chipset. We have already heard a lot about the Z390 chipset but it looks like it will be replacing the Z370 chipset or will be a more better tuned design for high-performance 6 core chips in the Coffee Lake family.

For users that are worried if their Z370 investment will go to waste need not to worry a lot as this chipset won’t be available until second half of 2018."

So 2H 2018 for Z390 optimized for high-performance 6 core Coffee Lake CPU's, so what is releasing now? "Chopped Liver" to hold everyone over till the good stuff comes out in 2H 2018? That's when Coffee Lake was supposed to arrive 2H 2018, so what's coming out now is probably relabeled / re-jiggered chipset support cobbled together to get 6 core out the door.

I think sticking with AMD for the long term makes much more sense, as Intel is going to have to make up for lost time in development by using the customers to take it all through the final stages of testing...

 

The 300 chipset seems more for those willing to wait for 3/2018 or beyond. So this is a battle for next year. Since Zen 2 is supposedly good on current boards we should be good with AMD till the end of 2018 at least.

Remeber too that the 6 core Intel is a Ryzen 7 competition not a chip killer.

 

well if intel keep pushing it back, im simply going to keep pushign it back lol. having waited few yrs theres no real issue waiting some more. intel wont get my money unless they give some half decent silicon this time, im sure there are people who is as fed up as i am waiting forever for a 4c to improve to mere 6c. for 8c and higher will have to go for gimped pcie lanes on x299 with higher latency mesh design (of course if 18 cores can hit 5ghz i dont really care about the extra latency).

so is it confirmed that 8700k and z370 will be delayed to 1H2018?

 

Where did you hear that? I thought Intel was trying to get the cobbled together 6 core motherboard chip set and Coffeelake out before November.

Here it looks like September, but that's already going to be past soon, so maybe October with November availability? (rumor, as is the leaked image)

 

In all honesty, it would be great if Intel delayed until 1H 2018.
That way, when they do release their Coffee Lake, they can go up against Ryzen refresh on 14nm+ (which probably means higher clock speeds, possibly past 4.0 GhZ and better efficiency - for Ryzen of course - that way, we can more directly compare AMD and Intel on a same, or very similar manuf. process).

 

Why should they? Mind yoo, 2 rev of MB aka Z370/390 need to be pushed out And I can smell 8 core coming with the release of Z390 2H 2018 Milk to the last drop

 

AMD does need to get 14nm+ out. If clock speed and IPC were less of an issue Ryzen could have a clear cut dominance in the market place. This though is the learning curve needed to get the chips on the market. It is too bad Vega was such a failure because of delays, they could have been essentially pitching a no hitter for game one of the season.

 

Ryzen Threadripper Core X in the test: 1900X and i7-7820X with eight cores in the duel 5/5 - Computerbase.de (9.7.2017)

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

Edit. AMD’s Threadripper Was Designed By Engineers In Their Spare Time
"What if I told you that the world’s fastest desktop processor available today was never originally planned? That’s right, in a revealing interview with Forbes, AMD’ers Sarah Youngbauer and James Prior revealed that the company’s pride and joy, Ryzen Threadripper, was developed by engineers in their spare time."

 

Here to me is the problem. If TR x399 is upgrade even in the future where does it get us? This would maybe be at 7nm and if so what then? Well looking at CB r15 @ 4.0 GHz we get about 3,400 points. If 7nm gets 5.0 GHz and a 10% true IPC advantage that under 100% scale is only 4675 points.

I think the 7980XE as it is now could get us there or more under extreme overclocking. In the end the upgrade path over the x299 just may not truly exist. Also by the time we see 7nm the 7980XE may be cheaper and a better chip than it is at first release overall.

The recent video's, touting more from Ryzen is to come, smells of Vega type hype to me.

 

All the top CPU's from Intel I've had never dropped enough in price to make them attractive to anyone over what was new from Intel. Intel keeps their top CPU's prices high all throughout their shipping lifetime.

There are no bargain's when it comes to Intel CPU's

AMD on the other hand has dropped to about 50% over time, so turning that whole concept around, you can get the AMD parts that match the Intel part's for 1/4 the same price

Besides, when X299 dies, the TR4/AM4 sockets will still be going, so you could actually buy a new AMD motherboard with new goodies, and run the old AMD CPU until you saw fit to buy a new AMD CPU, for the same socket.

Intel's the reverse, you gotta buy the new motherboard to get the goodies and a new CPU to fit the new socket / motherboard chipset.

Intel's CPU /socket is a loser moving forward, you may get a lucky binned CPU OC'er, but that CPU's locked into the motherboard and features for today, forever.

 

You are 100% sure 7980XE is the last maxed Core i9 we can see be used in X299?

 

It's going to be tough to put more power through those x299 motherboards without combustion.

Do you really think Intel is going to fund new CPU releases for x299 after leaving that socket behind?

 

My point is when we thought 7980XE was going to only clock to 4.0 or 4.2 GHz there was a chance for a blow for blow fight out between AMD and Intel on x399 vs. x299 platforms. At 4.7 to 5.0 GHz TR will just keep on trying to match the 7980XE with little to no hope of catching up even with AMD at 7nm. So while down the line we may get an upgrade to x299 and the 7980XE and this will become old tech the TR stands little to no chance to even best this.

If AMD has anything they need to come up with those FACTS, not just hype. This or the market will soon see AMD is running scared and will loose confidence in Ryzen itself as a platform (get shady on info on one aspect and all aspects will look shady).

 

AMD 1950X = $1000
Intel 7980XE = $2000

 

naw, get silicon lottery 7980xe, $2600

 

Was googling for that link, and found this estimate of power draw for the 7980XE

http://www.overclock.net/t/1631319/skylake-x-binning/590#post_26329639

"I reached similar numbers for non-AVX code when extrapolating the 7900X. Throw in the AVX512 and we're upwards of 1000 - 1200W.

From my 7900X: cache 3.0 GHz @ ~1.1v?
Non-AVX: 280W - 4.5 GHz @ 1.24v
AVX: 250W - 4.0 GHz @ 1.05v
AVX512: 320W - 4.0 GHz @ 1.05v

Extrapolating to 7980X @ 5.0 GHz: cache 3.0 GHz @ 1.1v?
Non-AVX: 280W * (5.0/4.5) * (1.35/1.24)^2 * (18/10) = 664W
AVX: 250W * (5.0/4.0) * (1.37/1.05)^2 * (18/10) = 957W
AVX512: 320W * (5.0/4.0) * (1.40/1.05)^2 * (18/10) = 1280W

I'm speculating vcores of 1.35/1.37/1.40 for non-AVX/AVX/AVX512 respectively.

If 1200W could be sustained (which I highly doubt on an X299 socket regardless of cooling), it is 10 amps at the wall on 120V. Most circuit breakers trip at 15 or 20 amps."

SiliconLottery.com doesn't have it yet, but I wouldn't be surprised to find the top bin go for $2600

 

Well a 1950x system = $4,000 USD
a 7980xe system = $5,000 USD

Now that is 25% higher cost but CB R15 3400 on a 1950x @ 4.0 GHz as compared to 7980XE 4645 is 1265 higher a score or a better than 33% boost. If the 7980XE were not so much faster it would not justify, but it is. Now if those numbers do not justify things may be different but at this point it does not look good.

 

It's more like $6000 - 7980XE vs $4000 - 1950x, given the SiliconLottery tax + extra cooling + power supply costs, and it's yet to be seen how practical in day to day operation that performance will be on existing x299 MB's.

It'll be a while before we know for sure how much the bump up for that extra performance costs and how useful it is for every day use.

Give it a few months before committing to a build, so improved motherboard + power solutions + cooling can be released to support that over the top power required.

 

I am not counting SL, until I see we need to do that. Again cooling and PSU maybe, but we shall see.

 

None should jump on it before new tech is finally out and tested!! Especially expensive tech. I would regret and being upset/angry with my purchase of the next best, if I really wanted the best. And we haven't seen max oc'd 7960x on other than LN2. Maybe a killer vs. low clocked oc'd Threadripper with decent cooling.

 

i am expecting SL 7980xe chip to be around $2600 to $2800 for top end binned, expENSIVE.

only 1k watt, bring it