Very Excited About Upcoming 7nm Ryzen, but...

Why claim "AMD hasn't delivered anything to consumers on 7nm" except to be misleading?
it's not misleading if you read it in context. This is not a gpu thread, and whether they released a 7nm gpu or not has no bearing on my argument.

iow, I made a mistake, admitted it (well, I am now, okay?), and now you know what I meant. Please continue discussion on the topic.
 
Last edited:
Zarathustra[H] You make good points that can't be ignored but there is a reason Intel is having issues shrinking the die that I'm sure we aren't aware of at this time. Even if they come roaring back I intend to support Zen 2 becasue I love supporting the underdog.
In any case I'm sure the image below will be a nice blast from the past for you.

View attachment 159382

I used to own some DFI boards. I regret getting rid of them. They were some nice boards for sure.
 
..

There is an elephant in the room no one is talking about

Why is it AMD needs a 7nm process in order to be competitive with Intel's 14nm chips? By all accounts all else being equal, a 7nm chip should be crushing a 14nm chip. Smaller process size means less heat and power, which means you can crank things up more. The fact that AMD is only catching up to Intel's 14nm chips on per core performance on a 7nm process strongly suggests that all things are not equal.

AMD's current architecture may be better than the highly flawed Bulldozer architecture, but the above means that the architecture itself is still WELL behind Intel's, and that they are competitive solely because they currently have the process advantage.

I don't think Ryzen architecture is WELL behind Intels. Clock for clock they trade blows in some applications, so isolated IPC seems similar. It's in more latency sensitive applications that Intel pulls ahead, and a some of latency is process related. GF 12nm (really 14nm+) helped close some of that latency gap. Some amount may be related to layout CCX vs Ringbus. It remains to be seen how the architecture and process change impacts Ryzen 3000 latency.

Also the days of new process crushing the old process are not as clear cut either. I am doubting anyone's 7nm is going to hit 5GHz, so for brute clock-speed the old process remains on top. Remember that in pushing clocks speed in it's mature 14nm, Intel actually increased some feature sizes, not shrunk them, probably to aid in power handling at higher clocks.

We may have reached a clock speed wall on process shrinks, where going smaller makes it harder to hit 5GHz, this is part of the reason I expect Intel will stay on 14nm for it's top desktop parts even when 10nm has yield issues solved, or at least wait for Sunny Cove for desktop parts to mitigate clock speed loss with a native IPC boost.

IMO neither side is going to have a large advantage. They will both have competitive parts, which will allow AMD to make continued inroads, because the current market share still reflects a hangover from the Dozer years.
 
Well that kind of goes with any board to be honest. The boards I used to have the absolute nightmare issues with was always Abit.

I always had good luck with Abit. I just pulled out my first motherboard. An Abit IT7 Max2 v2. Still works fine ;). I had a number of DFI boards. I should have kept them. Probably worth something now ;).
 
Those were great times. I got all that stuff and enjoyed the crap out of it. I wouldn't call myself an AMD fanboy though.

I'm pretty similar.

I like AMD for many reasons. Part of it is the joy in rooting for the underdog, part of it is the good memories of college and how exciting the lead up to the K7 launch and the following performance race was.

I don't consider myself a "fanboy" though. There is a reason I haven't built an AMD system as my desktop in close to ten years. They haven't brought the performance.

I have been reluctantly buying Intel. I'm a huge believer in the concept that whenever you spend money you are voting on what kind of world you want. I find Intel's litigious nature and their manipulative an illegal business practices over the years contemptible, and I hate giving them my money.

This is why I am excited to be able to spend money with a company that better reflects my values, contributes to open standards and supports backwards compatibility.

I felt the need to state something in the beginning of my post to clarify my background- however - because people are often quick to accuse anyone they disagree with as biased, a shill or a troll.
 
I was an AMD fanboy. Phenom I drove that out of me. I still have the core2quad linux system that I purchased on the profits of selling shares of AMD + Microsoft live.com eBay / cashback..

With that said there is a Ryzen 7 2700 to replace that. I just need the time to make the switch.
 
\Why is it AMD needs a 7nm process in order to be competitive with Intel's 14nm chips? By all accounts all else being equal, a 7nm chip should be crushing a 14nm chip. Smaller process size means less heat and power, which means you can crank things up more. The fact that AMD is only catching up to Intel's 14nm chips on per core performance on a 7nm process strongly suggests that all things are not equal.

AMD's current architecture may be better than the highly flawed Bulldozer architecture, but the above means that the architecture itself is still WELL behind Intel's, and that they are competitive solely because they currently have the process advantage.

That's pretty far off. First off, Intel's 14nm process is better than GloFo's 12nm. Juanrga was an Intel shill most of the time, but he was right about one important thing: these are marketing labels now. Intel's 14nm is amazingly good.

And Zen, even on inferior nodes, was kinda-sorta competitive with Intel's offerings. Yes, lower per-core performance due to IPC and clock speed disadvantage, but the architecture allowed for cheap core scaling to compensate. And that was on an inferior process. Give and take. Where Bulldozer used a similar strategy, it did so with a major (> 50%) IPC gap. That was too much. Zen had a more modest 12% gap or so (and Zen+ at 9%). Much better.

Now, AMD has a process advantage, yes. But it isn't as great as the marketing labels would lead you to believe. Intel's 10nm is superior to TSMC's 7nm. The node advantage over Intel's best 14nm process is there, but not huge.

Also, AMD has supplied architectural updates which SUPPOSEDLY bring the IPC to rough parity, or perhaps a hair better. Now, take that with a grain of salt. A 15% IPC bump is very large by today's standards, and I will wait to see benchmarks before truly believing this. BUT if true, this is huge. How much of an IPC gain will we see with Intel switching to 10nm? My gut feeling on this one - and also a grain of salt, because gut - is that Zen had a lot more low-hanging fruit to pluck off the tree than Skylake/CFL/etc..., which were very mature. If Intel supplies another 3% IPC gain or something, then the tables won't be turned in any major way. You still have a competitive AMD offering.

We haven't seen final clocks. But it's starting to look good-ish in the rumor mill. Supposedly a 16 core Zen 2 early ES was spotted with 3.3 base/4.2 boost. This is near Zen+ Threadripper levels already, and presumably at a much lower TDP. If final clocks are TR-ish for 16 core, we can probably expect a higher-clocked 8 core product. My guesses: +200 over ES. So 3.5/4.4 for 16 core. And +300 for 8 core fast product. So 3.8/4.7. But that's pure guesstimating, so *shrug*.

In other words near-Intel clocks. Similar-to-slightly-better IPC. Much greater core density at given price bands. Intel 10nm will have a lot to deal with to beat that - at least at a value level. Intel will probably retain max single core performance due to minor clockspeed advantage.


The fact that AMD is only competitive with Intel's 14nm chips because they are at 7nm means that once Intel fixes their process (and they will eventually, probably skipping 10nm and moving on to the next smaller one at this point, I'm guessing in 2021) Intel will come roaring back and crush AMD.

AMD really needs to not rest on their laurels with 7nm Ryzen, and not spread themselves too thin with acquisitions and adventures in other tech and really focus on improving the core Zen architecture to the point where it is competitive with Intel AT THE SAME PROCESS NODE. They have ~2 years to get there. If they don't, they are toast.

They won't be toast. Zen 3 is supposedly going on a "5nm" marketing labeled process. If true, we can figure that TSMC may retain a slight process lead over Intel anyway. But figure that at any given marketing number, Intel will be BY FAR the best.

Edit: something to realize is that cheap core scaling is IMPORTANT. This is an imperfect analogy, but if you have V8 vs. a Turbo 4 banger, neither one is explicitly better based on that information alone. Cost matters. Power under the curve matters. Torque. Etc... Focusing too much on one variable is a mistake.

AMD back when Zen was being developed didn't have the R&D budget and market clout of Intel, nor access to a process that was as good. They probably knew creating a Skylake competitor in single core output when they were bleeding money was too hard. But they could create a close-ish 85% competitive architecture and make it easy to scale core count instead. Then, when the coffers started filling up, push it to near-competitive single core performance whenever they could get out from under the wafer agreement with GloFo (or if GloFo could deliver a competitive process - but we all know how that turned out).

Each company is playing to its strengths with their designs.
 
Last edited:
I don't believe AMD will be able to complete with Intel in the single core, but I think they will be close & I think their pricing will be enough to justify the purchase.
 
I don't believe AMD will be able to complete with Intel in the single core, but I think they will be close & I think their pricing will be enough to justify the purchase.

By definition, close is competing. :) Currently, AMD is competing across the board with Ryzen/Threadripper/EPYC; they may not be beating Intel in the same core segments (but assuredly do on price), they may be a bit slower than Intel for most games, but they're certainly competing.
 
They won't be toast. Zen 3 is supposedly going on a "5nm" marketing labeled process. If true, we can figure that TSMC may retain a slight process lead over Intel anyway. But figure that at any given marketing number, Intel will be BY FAR the best.

Supposedly still 7+nm for Zen 3. there is an "interesting" video on Zen 3 by moore law is dead on youtube however it is video where he lists possibilities for what Zen 3 might have :).

There is a problem for me when people start yelling marketing numbers for nano meter process manufacturing. You basically can not check what yields the process is having you never hear anything about 3rd party using Intel foundries on the numbers that makes you feel that the information that you have at your disposal is something you can take to the bank .....

There is one thing that you can take to the bank that AMD profit will fund their R&D and not be used to fund all kinds of activities Intel has been known for that are crippling the competition by other means ....
 
I don't believe AMD will be able to complete with Intel in the single core, but I think they will be close & I think their pricing will be enough to justify the purchase.

Intel will not remain on top forever. This is a fundamental law of the universe. Entropy determines this in all systems and it cant be violated.
 
...I am worried.

First off, let me be excruciatingly clear.

I have been a long time AMD fan. Some of my favorite times in this computer hobby for me came when I started college, which perfectly coincided with the 1999 Athlon launch. Having been a long time PC builder (since I was 11 and had a 286) I finally had part time work and a budget to build a real system, and I had a long string of AMD systems.

Ever since 2006, I've been excited about the prospect that AMD would come back and once again be competitive per core with Intel.

- Phenom was a disappointment with its TLB bug
- Phenom II couldn't quite keep up.
- Bulldozer was an utter disaster, and none of the follow-on designs helped much.
- The first two Ryzen releases almost got us there.
- 7nm may finally seal the deal.

Depending on which rumor you believe, 7nm Ryzen will either catch up, just barely miss, or just barely beat beat Intel's current offerings on a per core performance basis. This is great! it's finally happening! But...


There is an elephant in the room no one is talking about

Why is it AMD needs a 7nm process in order to be competitive with Intel's 14nm chips? By all accounts all else being equal, a 7nm chip should be crushing a 14nm chip. Smaller process size means less heat and power, which means you can crank things up more. The fact that AMD is only catching up to Intel's 14nm chips on per core performance on a 7nm process strongly suggests that all things are not equal.

AMD's current architecture may be better than the highly flawed Bulldozer architecture, but the above means that the architecture itself is still WELL behind Intel's, and that they are competitive solely because they currently have the process advantage.



Lets take a historical perspective

Last time AMD was successful with the Athlon launch it was through a combination of great efforts on AMD's part, some key acquisitions (NexGen), Technology Licenses (DEC Alpha EV6 bus) and strategic hires (layoffs from DEC), but also in HUGE part due to Intel's spectacular failure with Netburst and the Pentium 4.

This gave AMD a limited opportunity to break in, and try to cement themselves in the industry, before Intel came roaring back from the Netburst mistake. Through a combination of mismanagement on AMD's part, different priorities and a ton of unfair (and illegal) business practices on Intel's part including exclusivity bribes to OEM's and the Intel compiler intentionally sabotaging AMD performance (later resulting in a $1B settlement paid by Intel to AMD) they were not successful in this regard. It was a desperate and damned near insolvent AMD that finally accepted a lowball $1B settlement from Intel because they didn't have any other choice. When they received it, acquiring ATi and entering the GPU market took money away from R&D resulting in a string of disappointing CPU releases over the next decade.



History Repeats Itself

Similarly, this time around a combination of an internal effort (Zen development under Jim Keller) and Intel screwup (the spectacular failure of Intel's 10nm process) have given AMD an opportunity to break in and cement themselves once again. Unless something drastic changes - however - they look set to fail this time as well.

The fact that AMD is only competitive with Intel's 14nm chips because they are at 7nm means that once Intel fixes their process (and they will eventually, probably skipping 10nm and moving on to the next smaller one at this point, I'm guessing in 2021) Intel will come roaring back and crush AMD.

AMD really needs to not rest on their laurels with 7nm Ryzen, and not spread themselves too thin with acquisitions and adventures in other tech and really focus on improving the core Zen architecture to the point where it is competitive with Intel AT THE SAME PROCESS NODE. They have ~2 years to get there. If they don't, they are toast.


I'd like to hear your thoughts.
I sincerely hope they knock ot out of the ballpark. I'm huge AMD fan but.. I have to agree with a lot of your points. If AMD doesen't manage to establish a killer lead on Intel with the Zen 2 ... Intel will likely start kicking ther asses again. I think a lot of weren't hoping for them to establish parity with previous Intel Generations. Im sure we were hoping for AMD to establish a massive lead over them.

While I can understand why AMD went with the 12nm node on their mobile refresh, allowing them to use parts readily available
They really needed to skip it and go straight to 7nm before Intel gets their 10nm node put at the end of the year. Their gains in mobile have been impressive only because of intels lack of chips.

For now, I will simply stick with my intel processors. The only AMD stuff I'm running now are the 1700 for my server, down clocked and under volted for efficiency and a G200 on my bedroom micro pc. (Asrock A300). What i want is a chip that just owns everything. Worried we just wont see that. Hoping AMD gets their frequencies up to the 5Ghz mark.

I guess we just have to wait and hope for the best.
 
There is a problem for me when people start yelling marketing numbers for nano meter process manufacturing. You basically can not check what yields the process is having you never hear anything about 3rd party using Intel foundries on the numbers that makes you feel that the information that you have at your disposal is something you can take to the bank ....

Yields are important, yes. However, that does not change the fact that when Intel claims 14nm, or 10nm, it's different than when TSMC or Global Foundries claims the same number. They are marketing labels - and this is partially because node size is really more complex than can be easily expressed in one number. Intel is much more conservative about this. This image has been posted here before, but it should help:
 

Attachments

  • Dde0RFIVQAAcZfO.jpg
    Dde0RFIVQAAcZfO.jpg
    102.6 KB · Views: 0
I sincerely hope they knock ot out of the ballpark. I'm huge AMD fan but.. I have to agree with a lot of your points. If AMD doesen't manage to establish a killer lead on Intel with the Zen 2 ... Intel will likely start kicking ther asses again. I think a lot of weren't hoping for them to establish parity with previous Intel Generations. Im sure we were hoping for AMD to establish a massive lead over them.

While I can understand why AMD went with the 12nm node on their mobile refresh, allowing them to use parts readily available
They really needed to skip it and go straight to 7nm before Intel gets their 10nm node put at the end of the year. Their gains in mobile have been impressive only because of intels lack of chips.

For now, I will simply stick with my intel processors. The only AMD stuff I'm running now are the 1700 for my server, down clocked and under volted for efficiency and a G200 on my bedroom micro pc. (Asrock A300). What i want is a chip that just owns everything. Worried we just wont see that. Hoping AMD gets their frequencies up to the 5Ghz mark.

I guess we just have to wait and hope for the best.


AMD is not going to have a killer lead (except maybe on core count) on Intel, and Intel is not going to start kicking their ass again in the foreseeable future. Intel was only kicking AMDs ass because Dozer was a blunder of epic proportions.

Core count hits diminishing returns very quickly, unless you have extreme parallelism, as in pure rendering/encoding task (Amdahls Law), and even in many rendering tasks you will start to see falloff due to thread management overhead.

Though I agree mobile will be a short term exposure for AMD, if Ice Lake ships on time (shipping in June likely means to partners, likely for a big fall launch). Most of the home PC market is actually a laptop market, and 10nm Ice Lake looks like a very good laptop chip:
2019-05-08-32-100795893-orig.jpg
 
AMD is not going to have a killer lead (except maybe on core count) on Intel, and Intel is not going to start kicking their ass again in the foreseeable future. Intel was only kicking AMDs ass because Dozer was a blunder of epic proportions.

Core count hits diminishing returns very quickly, unless you have extreme parallelism, as in pure rendering/encoding task (Amdahls Law), and even in many rendering tasks you will start to see falloff due to thread management overhead.

Though I agree mobile will be a short term exposure for AMD, if Ice Lake ships on time (shipping in June likely means to partners, likely for a big fall launch). Most of the home PC market is actually a laptop market, and 10nm Ice Lake looks like a very good laptop chip:
View attachment 160087
Intel already launched their early 10nm on a Lenovo laptop. Overall it was underwhelming but its AVX2 performance was beastly. So, I have no doubt Ice Lake will be on time. Intel should have had the necessary time to refine that node.

I think intel is banking on the fact AMD won't do enough to close the gap and thats why we're going to see more 14nm++++++++++++ . my guess, Intel releases something very fast around 2022 on their own 7nm process. If AMD's Zen 3 isn't amazing by then (on 7nm+ or 6nm as TSMC expects they will move to) Intel will really have something amazing. All the talent they've assembled is going to yield something amazing.
 
I think the Ryzen 3000 CES demo of the (likely) 65 W chip which matched the 9900K in CB15 shows that 7nm gave AMD plenty of headroom to work with. Higher clocks, more cores, or a combination of those will fit in the 105 W AM4 power budget.

About how confident Intel is in their 10nm process, moving fabs back from 10nm to 14nm says all I think. The only products that came out of 10nm so far were Cannon Lake (decidedly underwhelming) and now Ice Lake.
Though I agree mobile will be a short term exposure for AMD, if Ice Lake ships on time (shipping in June likely means to partners, likely for a big fall launch). Most of the home PC market is actually a laptop market, and 10nm Ice Lake looks like a very good laptop chip:
I too think that Ryzen Mobile will not do well against Ice Lake at least until 7nm APUs ship sometime in 2020. Too bad that AMD failed to iterate more quickly on the APUs, a situation which reminds me of the RS780.
 
Too bad that AMD failed to iterate more quickly on the APUs, a situation which reminds me of the RS780.

AMD strategy is sound. They do the New CPU and New GPU separately first before they combine them. That makes sense, debug them seperately before combining them.

Also a 6 month to 1 year lead/lag matters a lot more to internet commentators, than it does to actual companies involved.
 
Adored TV is still standing by their 6 core APU leak. Won't be here until Q3 though.
 
I hope Zen 2 supports AV512 somehow. Not sure if it is proprietary licenses tech or not by Intel, but I could really use it to video encode with at the processor level. My threadripper is fast but AVX512 would be some damn nice icing on the cake.
 
I thought a big part of AMDs problems where that in the Athlon days, after having just sold the fab off to GF they were in an exclusivity agreement, which severely constrained the production they were able to do.

I don’t disagree about ATI price... I think where they were headed with it made sense, it took a lot longer than they thought but today it’s doing ok for them. APU style chips are in a good spot for AMD; trying to compete for the high end with nVidia less so.
 
I know AMD Had some misses and maybe will never be able to beat Intel. But shouldn't we all be excited that we can get 90% of the performance at 85% of the price of Intel while not supporting evil INTEL!!
 
I know AMD Had some misses and maybe will never be able to beat Intel. But shouldn't we all be excited that we can get 90% of the performance at 85% of the price of Intel while not supporting evil INTEL!!
Closer to 90% of the performance for about 45% of the price.

8700k = $390 vs 2600 = $170
9900k = $495 vs 2700 = $220
 
  • Like
Reactions: N4CR
like this
^ If you go only by single thread, it's 80% of the performance. Still a win for most people who don't need bleeding edge single threaded performance.
 
If you're going to be Fuddy McFudderton about AMD, you should also turn the same lens on Intel's last 10 years of basically doing nothing but tedious incrementization and stagnation.

Quick question: 4 core i7 at microcenter for $199 is a great deal: what year is it? (followed by several years of 4 core i5 for $199ish)

Sandy Bridge was the last real desktop leap from them, with a decade of refinement on top. Also a lot of why the Core leap (rescue that came from mobile team) before was "so great" was because Pentium 4 was so bad.
 
Closer to 90% of the performance for about 45% of the price.

8700k = $390 vs 2600 = $170
9900k = $495 vs 2700 = $220

I was just trying to be conservative. Don’t understand why people think that getting the extra 10% performance is worth supporting a company that has been juicing us up for the last 10 years. If it wasn’t for AMD, intel will still be selling 4core i7s and telling us it’s amazing. Think twice before giving intel your money; even though they finally are giving you something that is worth the money, they don’t deserve it.
 
Closer to 90% of the performance for about 45% of the price.

8700k = $390 vs 2600 = $170
9900k = $495 vs 2700 = $220
Intel has a price and performance answer for the 2600, with the i5-9400. But that's now. This wasn't the case, when Ryzen launched and even with their Zen+ refresh.

I've always been an AMD fan and continue to be a fan of Zen. But, currently, the i5 9400 is a VERY good deal and for gaming, its tough to argue a 2600 instead. A 2600 makes tons of sense, if you are into content creation and other workloads which scale well with the extra threads.

*heck, you could even go for a 9400f and skip the integrated video. Save $20. Personally, I like quicksync and all that. But you could do that, if you were pinching.
 
Last edited:
Intel has a price and performance answer for the 2600, with the i5-9400. But that's now. This wasn't the case, when Ryzen launched and even with their Zen+ refresh.

I've always been an AMD fan and continue to be a fan of Zen. But, currently, the i5 9400 is a VERY good deal and for gaming, its tough to argue a 2600 instead. A 2600 makes tons of sense, if you are into content creation and other workloads which scale well with the extra threads.

*heck, you could even go for a 9400f and skip the integrated video. Save $20. Personally, I like quicksync and all that. But you could do that, if you were pinching.

And you should still give money to AMD just for making intel finally releasing a 6 core i5 cpu ;)
 
Intel has a price and performance answer for the 2600, with the i5-9400. But that's now. This wasn't the case, when Ryzen launched and even with their Zen+ refresh.

I've always been an AMD fan and continue to be a fan of Zen. But, currently, the i5 9400 is a VERY good deal and for gaming, its tough to argue a 2600 instead. A 2600 makes tons of sense, if you are into content creation and other workloads which scale well with the extra threads.

*heck, you could even go for a 9400f and skip the integrated video. Save $20. Personally, I like quicksync and all that. But you could do that, if you were pinching.

The 9400 needed to be released, but it doesn't hands down beat a 2600. Each requires extra cost to reach near peak output. 9400 needs an higher end mb to use faster memory and the 2600 needs a better air cooler to hold a higher all core clock. But even with all of that, the margins between the two are almost always with 5%, and not always with the 9400 winning. Overall, it matches a 2600 for gaming and definitively loses for content creation or streaming.
 
The 9400 needed to be released, but it doesn't hands down beat a 2600. Each requires extra cost to reach near peak output. 9400 needs an higher end mb to use faster memory and the 2600 needs a better air cooler to hold a higher all core clock. But even with all of that, the margins between the two are almost always with 5%, and not always with the 9400 winning. Overall, it matches a 2600 for gaming and definitively loses for content creation or streaming.

Intel needs to stop Hording SMT/HT for the top end models only and do what AMD does, enable SMT/HT everywhere except disable it on the bottom model. That way users would get better CPUs in the midrange and 9400 would be more competitive at content creation.
 
Wow - and this gets released just as AMD is about to announce their new hotness. Double-ouch!
 
Intel needs to stop Hording SMT/HT for the top end models only and do what AMD does, enable SMT/HT everywhere except disable it on the bottom model. That way users would get better CPUs in the midrange and 9400 would be more competitive at content creation.

Yeah, about that...

Intel just lost hyperthreading. Woops!

The server world is about to get INSANE:

AMD is going to be shipping 64C128T per socket, while Intel will be shipping 28C28T per socket. (if you pay a lot of money for a bizzaro HPC prebuilt barebones-only with a special BGA-soldered "single" CPU it will be 56C56T, but it truly is two CPUs stuck together)

AMD: 100 more threads than Intel.

LOL
 
Back
Top