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

Cannot love those TDP targets hard enough!!! And I do agree with the core count thing, but really want 8-channel memory next year. Just have 8-slots for the memory, instead of 16, but with 8-channel. They do that and I will be happy. If they include PCIe 5.0 support, plus the memory channels, I'd be ecstatic!

 

well, to be fair, the stated 250W TDP was defined as "very conservative" with typical loads far lower than that. so were probably looking at way higher efficiencies here and more wattage headroom for overclocking

 

Now if Ryzen master would allow a base and turbo boost based on a higher TDP, then we might be set. The one bas thing is the 2950x for 16 cores may be spread over the 4 CCX's now instead of 2. This might cause some issues with performance, then again maybe not.

 

i'd disagree. tdp is good only for a general guide line, for a very specific intel workload on how they test their chip, for their own tdp.

@jaybee83 generally yeah every core added is gonna bring the chip down, literally every core will have to be as binned as possible for us to see a 20+ core chip able to do 4.2+ ghz. with decent temp/voltage. since intel is going to start binning them and they do have the capability of doing this on 14nm and pushign it further, its a good bet if anyone were to do it, its intel.

 

Huh? TDP is the OEM's maximum heat output that is in need of dissipation running under designed limits. Since every chip runs differently actual TDP per chip is usually different. This is your silicon lottery.

My reason for 350watt is a 1950x is 180w, so doubling the cores gives 360w. 12nm should under the stock clocks of the 1950x not really use more power but because of the smaller die the TDP may have slightly increased.

I would have been more interested in not having a max OC set and better to have the new turbo boost keep up a higher IPC under low core usage. Again though we have to see what people start getting with the silicon.

 

Sent from my SM-G900P using Tapatalk

 

Nah, there are 2 CCXs per die and currently 4 CCXs on our chip. Where the issue comes from is 2 more dies. But, why would you add two good dies to the 8 or 16 core variants when you could save those usable cores for other chips? That is why I don't think it will happen, to be honest. No benefit, lots of drawback, and if you point to the 8-core variant, that was to get people on the HEDT platform and to have access to higher PCIe and quad channel memory. So there were benefits to be had. Here, there is all drawback to do it with four dies instead of the two die outside of the 24 and 32 core chips.

Tanware addressed TDP. It shows you should look into that more. Next, you do not bin per core like that, EVER. You are mistaking chip or die binning or recognizing which cores boost highest (a type of binning) for magically being able to select only the best cores ever made to be on the same die (or that is what you make it sound like). Instead, you need to look at the die surface area, yields, and whether enough chips make it to binning to make it worth it. The product won't be out for 6 months, will have low sales, and was meant as a reaction to the AMD news. It really is desperate and pathetic on Intel's part. Also, the blind "If Intel can't do it, no one can" cheer-leading makes no sense. Intel really hasn't even been on top since the start of this year and their descent started awhile ago.

So, Der8auer delidded a 2700X and found the die to be the same size. This means a hybrid node was used, where the base was the same as 14nm, but the transistors and process were 12nm, which can help explain some of the changes that occurred. So, potentially, the chip is less dense, which frees up the heat more, which allowed it to be more apparent that heat limits this design. But, that also allowed for reaching higher in OC.

As to the turbo and IPC, I posted the reading on the changes to the Ryzenmaster utility. Although I would prefer bios OC, some of the new features on there are worth a look and might allow for the exact things that you said you are looking for. Check it out if you have time.

 

i wouldnt think so. i hardly go by TDP value because i overclock, i only care about TDP value simply because notebook heastink are designed base off of it and affects my overall overclocking capability. 28 core 5ghz sounds nice, now we just need that mesh capable of running at around 4500mhz then maybe it'll somewhat match ring bus design. still better than ryzen infinity fabric latency issue tho.

theres also that hexa channel, 96GB ram woot

 

So long as with Ryzen you do not go crazy it is easy to stay below 200watt power draw on the 1950x. A reasonable overclock of 3.7 will get you there even under Prime95 and not all of that is actual TDP. The Enermax 360 is supposedly good for up to 500w total TDP. I could easily see a base of 3.7 or 3.8 GHz and turbo like it is now on the 2700x at 4.35 GHz. Even without further overclocking this would easily spank the 7980xe, and all of this without even touching 7nm.

I said early on if AMD with the 1950x went to 24, or better yet 32 cores, it could devastate Intel offerings. Then again it was early in the game and I doubt their fabs could have handled it.

Edit with some math even a base of 3.6 CB R15 could be 6250 on a 32 core or even 4700 on a 24 core. all that at about 400 w power draw too. This also is based on the 105w of the 2700x at a base of 3.7 GHz. I should note the stock 2700x seems to scare 1691 and 4 times that is 6764 but I do not think it is a 100% scale and I am intentionally not leaving much room for the Intel camps to complain.

 

TDP still has purposes, even if you do not personally use it. But I can understand being dismissive of it generally.

As to the 28 core, review the video from GN where Steve literally blasts it because they were running the chip at -10C meaning this IS NOT a 5GHz 28 core. As he put it, he could, after delidding, LM, more extreme water cooling just short of chilled water, was necessary for an 18-core to reach 4.9, so this is deceptive at best and Intel lying to the press and consumers intentionally misleading them causing ******** headlines like 28-core 5GHz, which the chip CANNOT do under ANY normal measure. Intel pulled their displays of it to prevent the press from really digging into it, which shows this is a year old Skylake X chip unlocked because AMD was doing a TR2 with 32 core AND INTEL HAD NOTHING!

Also, Mesh is barely better on latency than IF. WTF are you talking about? Both Mesh and IF have shown that as they both stand without mod, they WILL NEVER approach Intel's old ringbus latency. Why? Because even though latency goes down with speed increases, there is a limit to how much that latency can be decreased. It is a simple math problem, and that math says what you want is impossible. Also, Intel's mesh transfer rates theoretically may be slightly inferior to IF, whereas the latency is superior. But that goes to discussions from a year ago I will not rehash ATM.

As to the ram, yank. As I and Steve said, this is old tech from a year ago that was offered to HEDT because INTEL HAS NOTHING. Meanwhile, if AMD does like I think they will with a controller chip AND to redefine with AM4+ and TR4+ next year, you may see that on AMD's side within months (top 8-months) before AMD answers with higher channel chips. Unless you have a need for the bandwidth, it may not help you. With all of that said, it is nice and the only quality I like about the product.

 

I knew it was time for me to come back to this thread.

 

Seems some x399 owners will be left out in the cold, at least when it comes to overclocking as this TDP increase caught some motherboard manufacturers off guard and their first gen boards can’t keep up with those power requirements especially when accounting for overclocking and going well above rated TDP.

“Because the original Threadripper included two dummy dies, AMD says the latest model should be compatible with current X399 motherboards. But AnandTech spoke with some motherboard vendors at Computex who warned that some X399 motherboards could have trouble with power delivery, and a new motherboard could be necessary for overclocking. A few new motherboards that could handle it were being demonstrated at the conference.”

https://www.google.com/amp/s/gizmodo.com/amds-status-competition-with-intel-ramps-up-1826599557/amp

 

That is a good and fair critique, but needs put into context. We all remember the Skylake-X VRM heat issue, right? And we all remember that roughly the same VRMs got used on X399. This means that it can overclock to a degree, but power delivery and heat could be an issue (my VRM sit around 38C under load at 4.05GHz, but I water cool them). Some X299 boards got revisions with, at minimum, better VRM sinks. But, just like you need a new board for those beefier Intel chips, you will potentially need a new board to power all four chips far. Because of that, even though I want OC numbers and heat at specified MHz numbers, a new board would be needed, which does defeat the allure of the same socket being used if you need more than the 16 cores currently offered. Depending on pricing, I'd consider the 24-core, myself, but would not need a 32-core. But my needs are not the discussion here, it is overclocking, needing new boards for the platform, etc.

So, back to the point. Only recently were boards introduced or floated designed FOR the TR4 platform anyways. As I mentioned, many elements of X299 boards were borrowed by MB mfrs. to create the X399 platform boards. That was obvious to anyone that has been on X399 for awhile during the past year. So, if getting this chip, which most people on the platform likely won't upgrade unless they NEED the higher core counts, will just wait for next year's offerings, which may include more memory channels, which would require a new board again also. It also may bring PCIe 4.0 or 5.0 with it, which further makes the upgrade look less favorable. And if buying TR2 new to the platform, you usually opt for the newer board.

Nonetheless, it is a completely fair critique and needs given weight.

 

Why would you think that? Why would AMD go out of their way to make a more expensive die arrangement just to make it run worse?

There has been no specific mention by AMD of TR2 16c / 12c / 8c CPU's, and in this video Jim Anderson said in that context that AMD will continue to sell the TR1 series of 16c / 12c (/ 8c?) CPU's.
http://forum.notebookreview.com/thr...ga-polaris-gpus.799348/page-470#post-10741065

Perhaps AMD won't make TR2 versions of the lower core count CPU's and only produce 24c / 32c TR2 CPU's?

 

cool thing is: with the newer versions of ryzen master, as previously discussed, well be able to fine tune each and every individual core on a TR2. so if it was me, i would start with tuning every single of those 32 cores to its best capability in terms of VCore and clocks. finding the best one, set that as the one core turbo. then moving on to the best pair of cores, tuning in VCore and clocks, setting those as the 2 core turbo. and so on.

that way you can find both the highest clocks, lowest wattages/voltages AND highest performance for any given scenario (games less cores, high clocks, rendering/transcoding more cores lower clocks) and get the most out of your chip

sure, itll all take like....FOREVER to tune everything. but hey, thats part of the game. u want the most performance, u gotta put the elbow grease into it! one button OC wont get u far in this scenario

 

Hmmm, doesn't TR2 have PB2 and XFR2 like Ryzen 2? If so that means if you let the CPU control things, keep the power coming and the cooling going, the CPU will get really close to that optimal performance on it's own, at least if TR2 does as well as Ryzen 2.

 

the software looks good indeed but i believe it doesnt work as what i wanted, which is TB3, unless it work just like TB3. however i dont own either of x299 and x399 so i cant really compare. if intel can have a few cores running at super high frequency we can expect 1-2 cores at like 5.5ghz should be simple for them cause CFL is doing just that on rare chip and these are highest end xeon should be ez for it, if that works then we dont need to have 28 cores at 5ghz, just 4ghz would be enough imho, which will prob be what it'll run at with just normal water cooling.

amd is good at 1 thing tho, bringing more cores/competition so I can get faster and more cores cpu from Intel.

 

yea i guess thats always a possibility. like with 2700x, manually overclocking giving u less performance than at stock *lol*

 

I'll respond to this in the morning, but some good points and I don't think anyone has compared the new features on the 2000 series with Intel turbo boost, so I'll do it in the morning. Disagree with the phrasing on waiting in Intel. But, good stuff is coming from this talk now, so thank you.

Correction, manual overclock for simplistic all core overclock did less, but I can't remember if a differential OC through ryzen master was done or not (can't remember, but know the all core was worse in GN reviews). But, even if the case, it only removes overclocker's fun, but is great for the average user. That is a benefit of Intel for those that want to overclock, even though Intel's superior performance is sliding....

Sent from my SM-G900P using Tapatalk

 

I can see where my air cooled VRM's could be an issue. But I agree I want to see the extreme overclocking disappear and outright give all the performance to all users.

My only other wish is for them to allow user to set a plus on the base and turbo to up the TDP. Lets just say to the clocks of the 2700x and about 400w TDP. Even a 3.5 base and 4.1 turbo keeping TDP to hopefully 350w. I know this might require better VRM cooling but so be it.

My wild guestimates based on TR1 for stock speed and 2700x for TDP;

 

Looks like they had it running with a standard air water cooling setup as well. 6100~CB at whatever clocks they had it running. This thing looks like it's going to be a overclocking monster and require no less than personal nuclear power plant to do it. I imagine you'd be sweating just sitting next to that thing under load.

 

Not stock cooling. Notice the left lower side of the water it turned to low away from the 480 radiator to the other side of the case. It looks like the liquid is flowed to an external cooler before returning to the case. The wording is most expect this to have been a minus 10c evaporator cooling and why the pulled the demo out right away before the bigger houses caught on.

 

gonna need minimum 360mm water cooling for this for just entry level overclock at 4ghz. 28 cores gonna be hot. phase change for say 4.6 should be doable if phase change is power enough but those cases are hella expensive.

 

It's highly unlikely they were using below ambient cooling in this situation or for the floor demo as that would cause condensation and you can clearly see there is none and nothing wrapped to preventative measures. So again that is a stock closed loop water cooler with a huge rad.

 

on the floor model, no, on the 5GHz, it shows it is clearly wrapped. It isn't with the foam type, it is a tape wrap insulator on it. Asked and answered that the 5GHz was with a phase change, potentially just single stage phase change, cooler, with the temp at -10C. This has been covered to death.

Now, rumors have the all core at 3.8. A 4GHz clock on the floor model would be around, or at least in the ballpark, of that score. As such, I'm inclined to think the chip was running around that on good cooling, not sub-zero cooling, for the LTT score. Stop with trying to muddy the waters.

Edit:


This can hit and hold 3C. With mods it can go sub-zero. But notice the insulated hose, etc.
https://www.tomshardware.com/news/intel-28-core-processor-5ghz-motherboard,37213.html
http://www.hailea.com/e-hailea/product1/HC-1000B.htm
https://www.aquatuning.us/water-coo...tra-2000-hc1000-1650w-cooling-capacity?c=6569

And here is one step up from the HC-1000:
https://www.aquatuning.us/water-coo...-hc2000-3300w-cooling-capacity-uk-plug?c=6569



Edit 2: Here is the picture of the board again. Who wants to count the phases?

 

Muddy the waters? I posted a video showing the chip running closer to stock or stock and I'm muddying the waters? There has just been a lot of talk about this thing REQUIRING a water chiller just to run, and its obvious from the video that isn't the case. That isn't muddying any water but rather showing it will run just fine on standard water at stock settings.

 

I think the elephant in the room is how Intel is now causing a divide in their HEDT platform. I've invested in both AMD's X399 and Intel's X299, with the 1950X and 7980XE, respectfully. With AMD, I can likely still use my TR4 mobo to use the 24 or maybe even the 32c TR 2nd Gen processor. On X299, I feel like Intel is going to abandon the platform just to get more cores over AMD. Sure, Intel will likely release new CPU's for X299 but the limitation will be 18 cores. I will reserve my complaints for my Intel rep if Intel is truly splitting up their HEDT platform.

 

Now, there is ambiguity in the first quote which I though you were separating into the 5GHz run, which we now know with certainty WAS sub-ambient, although the sub-zero reports were wrong, and the lower run that LTT did. If I misunderstood the meaning and both were referring to samples in the LTT video, not referring to the 5GHz demo, then I do owe an apology and am sorry. Misunderstanding. If anything other than that was meant, then not really sorry. Also, on the LTT case, we do not know if there was only a 480, or if they also had a 240 with the 480 in the bottom. But, I'm not going to quibble on that.

You are correct. This requires a new socket for Intel, requiring the same socket as with the Xeon platinums. It is mandatory with power requirements. AMD can push that power, but the VRM may be a problem on extreme overclocking. This is why MB mfrs are going from 13 to 19 phases on many boards, which the 19 phase number is coincidentally the same phase number that Intel board LTT showed has.

Let me paint a picture: After Intel showed off a misleading 5GHz score, then AMD announced their 32 core design, followed by it coming out that the 5GHz was sub-ambiant with a VERY powerful water chiller (one of the strongest on the market with one of the lowest temp setting regulators). Intel pulled display samples for awhile, then one appeared at the Asus booth the next day. No one is allowed to see multiplier on it. Even though embedded is a new socket and board are needed, breaking the two gen cadence, or that Intel will split HEDT like you mention between HCC and XCC chips), we see a problem of either not being faithful with the two gen on HEDT (sounds like Intel), or that they just don't have a clear path and seem more reactionary than even last year. So whether it is split or not, they are screwing up.

AMD, though, already had new redesigned boards coming. You have rumors that MB mfrs were not ready for 32 cores and AMD sprung it on them again. I call ******** since a couple of those new X399 boards have the 19 phases, meaning they knew and are giving cover to Intel, which is just dumb. If I knew which ones were saying that, if they have a 19 phase board at computex, I would take them to the ****ing ringer.

Either way, with the changes coming, it seems like a new board for TR would be needed next year or the year after, depending on the core count opted for in all this. With Intel adding 6-channel memory, I bet next year AMD does six or eight. Notice how the LTT MB only has 6 slots, whereas my current board has 8. Tells me that I could see AMD doing similar, keeping the same number of slots, just changing it to have extra channels. But, that requires new board, etc.

I also still owe a comparison on turbo boost and the new Ryzen 2000 features (forthcoming).

 

https://linustechtips.com/main/topic/934304-msi-tease-new-x399-refresh-meg-creation-motherboard/
Here is proof at least MSI knew about it. I would need to go over every mfr. to check (will eventually), but this shows that the MB mfrs knew that the 32 core WAS coming. Look at the phase lineup. They are basically borrowing designs from server hardware for these boards.




Edit: But imagine monoblocks for a board like this. You could do a Thor's hammer or pickaxe or some shape like that (from a design aspect).

 

wow, TR2 and the 28 core intel sku are gonna be SO close to each other in terms of CB15 raw performance...

- intel: 7334 at 28 cores / 5.0 ghz and 6109 & 6095 in LTT video, so that would be around 4.1 Ghz clocks

- AMD: 1950x with 3700 pts in CB15 at 4.15 Ghz (hwbot AIO top 3 avg.) and 2950 pts at 3.4 Ghz (stock), that equates to ~55 pts. per core and Ghz. so TR2 at 32 cores and 3.4 ghz all cores would equate to 6000 pts. so even with scaling losses, the 2950x can beat the intel 28 core on ambient water with a "modest" 3.5-3.6 ghz on all cores. however, reaching those 7334 pts. would then require...4.15 Ghz! same as we saw on TR1 abovehand, would u look at that!

@ajc9988 just goes to show how good der8auer really is. after delidding and analyzing the TR1 dies he predicted a 32 core TR as being quite likely...

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

 

This is what happens when they are being secretive. Now this time Intel will not show core count or multiplier info and we are all left in the dark .We supposedly will not see these till the end of year if that. A good 4 months plus for AMD to be the top HEDT game in town.

Edit; Most 7980xe video's at 5.0 GHz show the CB R15 score way below;
5.0 GHz CBR15 of 5039


Stock gonna get spanked


5.7 GHz cbr15 5635

 

curious to see if intel is man enough to let AMD take the crown in terms of core count...

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

 

Hey man I get it, you probably figured I was talking about 5ghz, which as well all know was a crazy PR stunt with a water chiller and isn't a likely overclock for a stock water setup. My intention was to show a realistic water loop setup. I even poked fun at the ridiculous power requirement to run that beast.

 

Intel doesn't have a choice. I noticed that neither of my predictions came true on Intel releases, but I know why it didn't happen either.

I predicted at minimum Cascade-X would drop. WRONG. That was the server replacement. I also predicted the mainstream release would happen (8-core). WRONG. I figured that Cascade would coincide with the HEDT release. That may actually be correct IN Q4!

So, remember, Intel does NOT have 10nm yields yet. They may not have that until late in 2019. So, Intel moved the release dates later to have more room to play. They know that 10nm will not compete well even against their 14nm+, needless to say the 14nm++.

Because of this, they needed a PR stunt to combat the 24 or 32 core they expected might drop. Fine, but they basically used an unlocked Skylake-SP chip, which is year old tech. I doubt that chip was on 14nm++, but could be wrong, just it seems closer to the 7980XE versus based on Coffeelake OC and clocks, but it is an ES anyways, so things are not final on the chip.

Now, that means AMD wins on core count and takes the lead for 4 months. They drop a finished product, debuting when AMD is sampling their 7nm chips, which is to put Cascade-X/SP in the best light going up against AMD's sample of 7nm Epyc chips. Considering Intel won't have another interesting chip until possibly H2 2020, it makes sense to put it against a sample to try to take the wind out of AMD 7nm server sales. That is, after all, where the real money is. So good strategy, but it is really starting to feel like Intel stood around with its dick in its hands for too long and is now frantically waving it around like a mentally ill person in the street, if being honest.

Meanwhile, they are also likely trying to get back on the yearly release cadence for the mainstream chips, so I'm expecting October or so now on the new 8-core mainstream chips.

Also, AMD, GIVE ME MOAR MEM CHANNELS!!!! Like, 8 of them should be good for next year! I want BANDWITZ!!!! Also, get better designs for multi GPUs and start working on 4 card xfire or SLI or pushing for the multi-GPU support in games for DX12, etc.!

I hear ya. Sorry about that. But, if you recognize the image for my avatar, it is Mugen from Samurai Champloo! LOL! But, yes, it is necessary for people to see what the setup is capable of and what common clocks may be. I want to know the temps on the VRM with 19 there!!! I bet that sweet buttery smooth power delivery helps a LOT! Bet it also helps on the X399 platform as well (since my VRM, under heavy sustained load only reach just under 40C, imagine what a custom water cooled VRM block for these beasts could do if not pushing it to extreme voltage levels).

 

Mathematically if we go by 100% then 5.0 on a 7980xe = 5036 the 28 cores would then be at 3.9 GHz or maybe 4.0 GHz to do CB R15 of 6100 or so.

 

We've calculated it a couple different ways, but since they all are coming up about 4GHz, give or take a couple hundred MHz, we've got a good idea of what it can do now, which is nice for discussion and comparison purposes. Now we just need to see the 32-core TR2 stock score....

 

The Intel 28 cores CPU achieved that score through heavy overclocking and ridiculous cooling that wouldn't be sold as standard (not by a long shot).
Essentially, Intel released benchmarks that are not representative of STOCK performance (which would be much lower).

It might be possible that the TR2 (32 cores and 64 threads) could achieve boost of about 3.6 to 3.8 GhZ across all cores, while Intel might be able to do 4 GhZ across all cores.
Scores should still be in TR2 favor in multicore on stock.

Yes, you could overclock Intel part to higher levels, but the added cost of ridiculous cooling (not to mention the Intel CPU itself) would be too high.

The 12nmLP process gave AMD an ability to drop overall costs... so, TR2 would likely cost in the area of about $1500 (or possibly less when you take into account how much less 2700x retails vs 1700x and 1800x).

 

When I spoke about 5GhZ across all cores for baseline (or extremely close to it), I was referring to consumer products, not TR/EPYC parts.

 

For the 32 core I would take right where TR1 is now stock, 3.4GHz base and XFR2 at 4.0GHz, a little faster would be nice too. Having 100% more cores and XFR2 would be a worth while upgrade. Not like I need it but it would be nice.

 

Given the reduced voltages, on 12nmLP, AMD should be able to get at least 3.4GhZ, possibly 3.6GhZ on base across all cores, and boost 200MhZ higher across all 32 cores.

 

Nice thoughts, but 3.0 is base and they are working on trying to get 3.4 for the all core boost on both the 24 and 32 core chips. Whether or not a person can achieve higher is still out to lunch, but they are also trying to get this achievable on air cooling, which is impressive. That is still 200mhz over the prior year's single core boost, which is a good achievement. So I believe the voltages are baked in already. That is from the anand article.

Sent from my SM-G900P using Tapatalk

 

curious to see what the beefy "stock" air cooler wraithripper will be able to do with good airflow/fans and thermal paste

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

 

Intel 28-Core Fantasy vs AMD 32-Core Reality
Hardware Unboxed
Published on Jun 8, 2018

 

Good point.
With such a high number of cores, core frequencies would need to be even lower.
Between 8 core TR and 16 core TR, there was a base clock differential of 400MhZ... so this seems to track with that (3.0GhZ base for 24 and 32 core versions).

Though, usually Ryzen manages 200MhZ boost above baseline... so if AMD can get 400 MhZ boost across all cores when they are stressed (thanks to the improved voltages on 12nmLP), that would be good.

 

I know the extreme overclocking community won't be thrilled but at that point it would vastly be the most powerful desktop available even overclocked unless LN2 is used..

 

You have to remember with both the Ryzen and TR when we were speculating based on leaks we were seeing speeds a couple of bins slower than the release just 2 months before. Their demo was 3.0 base and 250w TDP, does not mean where the skews will end up though in the next few months.

 

this. ill be expecting a bit higher clocks of the finished retail product

 

I am not saying but sure as anything I am hoping, they even said it was an engineering sample!

 

So, although I hold out hope, and you guys are correct on what happened first gen, there are a couple factors I'm bearing in mind.

First, they told anandtech that 3.4 was a work in progress. That suggests to me that they are trying to finalize voltage to qualify the chips within the TDP. If they raise the TDP to, say, 280W, I do not think it would be an issue hitting 3.4, if not higher. But, AMD doesn't want to put out what would be called the first 300W chip, even at 280W. It gives a lot of fodder to the haters and some journalists that would bash the product on energy consumption or performance per watt, while saying, potentially, that the energy efficiency is worse, which it isn't. That is the first reason I think.

The second is related, but is solely on the voltage. I believe AMD, this round, is trying to more tightly control the voltage levels for qualifying dies. By being more stringent, you only use more efficient dies at certain bins for products. This greatly increased efficiency of the product by keeping it before the large up tick in voltage consumption and heat output. So this plays into the first one.

Third, for the flagship server chip, they used the best binned dies. The rest of the server line had lower single core and all core boost, mostly. I believe they are trying to accomplish the same here, which is why we are not being told the 24 core chips are targeting a higher all core boost. You could use the voltage binning where they are not enough for the 32 core inside that envelope, but they can do it with 8 cores disabled. This is why, I believe, we are setting the same TDP and boost targets.

Of course, I could be wrong, but this is why I have the view I do on the chips' clocks. Also, AMD has a solid score on their 24 and 32 core chips. It is when people hype or pump up expectations that AMD's wins turn to a loss. Next year comes an even bigger win, and with Intel's demo being exposed, AMD really pulls ahead with getting some people on their HEDT platform, especially with a view to next year's chips, 19 phase boards, etc. But, of expectations don't meet reality, this can effect sales and performance.

Now, I'd love to get a 32-core chip to play with and overclock! I have zero need for one, as the 16 core alleviated the issues I had with the Intel mainstream chips for my own uses (in fact, it is more than I need, but Intel's mainstream was less than I needed and Intel's HEDT priced itself out of consideration at price per performance). So, if anyone wants to lend me a chip... *bats eyes at community*

Sent from my SM-G900P using Tapatalk

 

honestly speaking we can speculate but kinda hard to do estimate. given both intel and amd have so many cores on the cpu its guaranteed overclocking in the 4ghz will be limit to most in terms of heat. intel MIGHT switch back to soldering, which was stated in the hardforum article, that might take place starting from ICL though, they did mention intel going forward is back to soldering and target binning rather than IPC improvement, guess core architecture is already so optimized not much can come out of it anymore.

AMD with 32 cores is around 13-14% more cores than intel's 28. given they are both going to clock the same due to temp, it MIGHT be much better to go for 32 cores for a much better pricing. IPC wise intel isnt much ahead consider this will be an optimized ryzen vs intel's mesh. for other benefits going with intel at this point would be hexa channel memory and better storage performance, and higher overclocking if people choose to go exotic methods like phase/chiller.