AMD's Ryzen CPUs (Ryzen/TR/Epyc) & Vega/Polaris/Navi GPUs

Here is the thing, because Intel was losing on the traditional measure of transistor density with their 10nm versus TSMC 7nm, they designed a new way to calculate the density. Under it, their 10nm is 15% more dense than TSMC, but the person off the record from Intel said they were within 15% of TSMC, suggesting they meant behind. So, regardless of naming, we can still do math to estimate. But point taken.

Sent from my SM-G900P using Tapatalk

 

This might be of benefit:
https://seekingalpha.com/article/4151376-tsmc-intel-lead-semiconductor-processes

That's for TSMC 7nm process.
And according to that, it's 10nm process already achieved parity with Intel's 10nm if I'm not mistaken (although I could be reading it wrong).

But, something else we need to take into account... Intel only reported their own process density (possibly to make their process look better), which was never really properly measured if I'm not mistaken. Therefore, we only have Intel's word on their supposed 'process superiority' which has been brought into question easily enough.

Granted, Intel managed to get rid of a few gates according to latest reports (if I'm not mistaken), which resulted in better density... but I don't think even with this it stacks up to TSMC's 7nm.

Also, how does 7nm TSMC compare to 7nm from IBM (which GLofo will be using for Zen 2)?
Don't think Glofo will be losing a lot (if anything) even if the process is just comparable to Intel 10nm... because 40% higher performance over Ryzen 1 is still pretty good... so we're looking at 5GHZ base clocks for various Zen 2 CPU's which were in their Zen 1 state clocked at 3.6GhZ on the base (or 4.5 GhZ for Zen 2 vs Zen 1 chips initially clocked at 3.2GhZ).

This of course does not include boost... only base clocks (across all cores).
Either way, with 40% increase in performance from clocks alone, and of course IPC increases, we should see some great improvements.

That said... does AMD plan to integrate Vega into Zen 2 as standard or will they leave the CPU's just as CPU's?

 

that article supports my point beautifully! so according to this, intel´s "10 nm" node is basically on par with the "7 nm" nodes of the competition. so only if intel is pushing back 10 nm another 1-2 years will they actually be falling behind, and that only if the competition is able to push out their 7 nm products to the mass market.

not trying to defend intel at all cost here, its just that im kind of wary of these process node nomenclatures because they dont really have anything to do with the size of the actual nano-structures produced on the chips...

 

Truth be told everyone is saying they can do this, that and/or the other thing with their new process. In the end the proof lies in the pudding. We need the actual silicon to judge what the end results are.

AMD has a chance at getting towards real competition again. We still have bit of time to wait. I again reiterate 12nm has dampened my expectations as it did not deliver on the performance boost going from 14nm LLP to 12nm LP!

 

Because the 12nm LP turned out to also be designed for low clocks and mobile parts. It looks like AMD decided to change that at the last possible moment due to lower costs that particular design offered.
Plus, they never took advantage of 15% higher chip density on the node either (probably to save time and not have to redesign the whole thing - a feat likely reserved for 7nm - besides higher chip density doesn't necessarily always translate to better performance... usually more heat as you're packing more hardware closer together).
The specs for the 7nm LP still seem based on IBM's high performance process and the initial 40% performance boost claim over Ryzen 1 hadn't changed.
So there's room to be more optimistic in this particular instance.

 

The article also points out that Intel's 10nm superiority or 'parity' with other fabs lies predominantly in their claims... Intel's process never underwent proper measurements by third parties, so the industry has no way of verifying whether what Intel is saying is accurate or not.
And the article also points out that 7nm from TSMC is superior.

 

For whatever reason they did not capitalize on performance enhancement on 12nm they have dimmed my hopes on 7nm. I am not alone here either. But we will get a preview.

Since Epyc will get 7nm and Zen-2 first we may get a good preview of what to expect. I am sure the high end clocks will no be pushed but since they scale so well we should get a good idea or at least a great teaser.

 

same goes for the competitors and yes, based on the naked data the 7 nm nodes are superior in density by up to 6%, so yeah, lets call that basically equal to intel´s 10 nm node

 

Can't claim 'basically equal' when 7nm is obviously denser.
6% higher density is no small feat... Especially with Amd keeping some of the gates (unlike intel - so if Amd did the same, they would likely achieve eve better density). And besides, people love either making Intel look better or equalizing it with others even if its not.

By your estimate ryZen is 'basically the same' IPC wise even though Intel is ahead by 3-5%. Clearly, the industry always claims IPC superiority on Intel as if its a much larger one... But saying that for Amd is almost viewed as insulting to some.

I prefer accuracy. And this time, 7nm seems to be denser than Intel's 10nm.
But density alone doesn't guarantee which silicon will perform better. Higher density also brings higher temperatures as you're packing a lot more hardware closer together that increases heat production.

 

well, if you consider that just by the naming scheme, 7 nm should be 43% more dense than 10 nm, the actual 6 % in difference is negligible that was my point...

 

Ryzen 2700X | DDR4 4600Mhz vs DDR4 2400Mhz | Tested 13 Games |
For Gamers
Published on May 18, 2018


Corsair Vengeance LPX 16GB (2x8GB) DDR4 4600 (PC4-36800) C19 CMK16GX4M2F4600C19
https://www.amazon.com/gp/product/B076185YPC

The video details area has a lot of links to additional products, resources, videos, but be aware they have affiliate links, I stripped it out of the link above for the 4600mhz memory, but if you want to support the channel, please go to the link in their Details listing.

 

AMD Gains 46% CPU Market Share, 3 Out Of 5 Most Popular CPUs From AMD
By Talha AmjadMay 23, 2018
https://segmentnext.com/2018/05/23/amd-cpu-market-share-46/

"According to Amazon affiliate numbers, AMD has gained 46% CPU market share and there are 3 AMD CPUs on the list of top 5 most popular CPUs sold on Amazon. The Intel 8700K was the most popular CPU according to the numbers, which is not surprising keeping in mind the introduction of a 6 core 12 thread variant in the mainstream Intel series of CPU for the first time ever.

The previously most popular chip, the 7700K have been knocked back and have been replaced by the AMD Ryzen 5 1600 that takes the second place and in third place, we have the AMD Ryzen 2400G. In fourth place, we have the Ryzen 1600X. The 7700K came in at 5th place followed by the 8600K at 6th. All in all, out of the top 10 most popular chips, 5 were AMD chips and 5 were Intel.

That is a 50-50 percentage and very competitive indeed. If this is not a wake-up call for Intel then I do not know what is. The increase in the number of cores and threads has been the major selling point for AMD and while we have been hearing about an upcoming 8 core chip from Intel, you can argue that Intel is too late to the game. You can check out further details in the video included below.

Q1 2018 has been a good year for AMD looking at the CPU market share according to the video. It is impressive to see how far AMD has come ever since the AMD Ryzen series of CPUs came out. It is safe to say that AMD saw an opening in the market and was successful to take advantage of that. We have seen the refresh of the AMD Ryzen series and the top of the line AMD Ryzen 2700X.

Let us know what you think about this increase in AMD CPU market share and whether or not you are considering shifting to AMD Ryzen this year."

 

what the crap? how in heck did they manage to set 4600 mhz on the RAM with a 2700X? dayum

 

More 4600mhz DDR4 test results:

4600Mhz DDR4 | I7 8700K vs R7 2700X | Tested 13 Games |
For Gamers
Published on May 18, 2018


Corsair Vengeance LPX 16GB (2x8GB) DDR4 4600 (PC4-36800) C19 CMK16GX4M2F4600C19
https://www.amazon.com/gp/product/B076185YPC

 

So Ryzen beats the 8700k?

 

Sure thing... if you add 500$ memory module kit.

 

Depends on the RAM speed for games, for productivity the 2700x / 1800x / 1700x / etc win.

If the 8700K has 2400mhz memory and the 2700x has 4600x memory, then the 2700x is very close in gaming.

Even at 4600mhz on both the Ryzen is close in gaming, really not enough difference to notice if you tune the FPS to 120/144 in most games.

That same youtube channel has test video's with Intel CPU's @ 4600mhz vs slower ram comparisons, check them out too.

 

Closer to $600 ($590.89)

Well worth it to not run Intel, if that little bit of extra performance is important to you.

Otherwise build a Ryzen / ThreadRipper system and search for the "fastest" ram - at a discount.

 

How ryzen 2 is compatible with Windows 7 on practice, anybody knows?

 

I think AMD has a bit of a problem on their hands. This is with their own skews. Right now you can build a 2700x so there is less of an incentive to build a 1900x system. Especially looking at in a year or so having access to a 3700x Zen-2 or possibly further core counts?

Even if TR goes 12nm how far is Zen-2 based TRs? With Epyc going to Zen-2 later this year would it not make more sense to release all Zen-2, Ryzen and TR, in 2019 at the same time?

Not a perfect comparison but; https://www.cpu-monkey.com/en/compare_cpu-amd_ryzen_threadripper_1900x-767-vs-amd_ryzen_7_2700x-876

 

That isn't a problem at all for AMD. Intel has all the problems.

What AMD is doing is putting Epyc on 7nm by the end of the year FOR SAMPLING, not for mass sales. Volume will be available, IIRC, in Q2 2019 for Epyc 2.

Zen 1 and Zen+ (2000 series) is meant to get people on the platform. AMD even said they were not expecting to get a lot of upgrades from 1000 to 2000, but that the 2000 chips are even more potent and competitive.

As to the core count rumor, take it with a grain of salt at the moment. Yes, on 7nm it is possible that they may have a 12 or 16 core mainstream chip in the chute. If that came with at least 40 PCIe lanes and quad channel memory, I'd consider making the switch, if being honest. But, I do not know if that is what they are doing, and without those, it is a no go.

Now, TR 2 (Zen+) will be the 12nm variant, and all signs point to Zen 2 TR being released at the same time as recently, which is July/August of 2019. Meanwhile, we know that TR is made of binned dies from the Ryzen line and there are two revisions in play, one in Epyc and the other in Ryzen and TR. But there is a very good reason they pulled Epyc in so much this year versus the first release: Intel said they were transitioning to the HPC market getting their chips first, before mobile and desktop. With that statement and threat, skipping the 12nm refresh of Epyc and going directly for 7nm, while pulling in the release date by nearly 6 months (both for testing and for volume release dates), puts it at about 18 months, another prediction I made, although I was wrong in that prediction in regards to TR (I said that TR would also be pulled in and released at the same time as EPYC when skipping 12nm refinement, but I think that was made before it was known that TR and Zen shared a revision and Epyc had its own revision. Either way, I was wrong on TR skipping 12nm). Now, we know that TR is binned chips of Ryzen. They could, this time, use the chips from Epyc, but I doubt it.

As to 2700X vs 1900X, the new 2900X will spank the 2700X, plus it has quad channel memory and four times as many PCIe lanes. If you need either of those features, it becomes a no brainer which chip to get. It is the 2900X that is meant to go against Intel's 8-core mainstream chip and against Cascade's replacement of the 7820X, which has cut down PCIe lanes, neutering that chip. Considering expected prices, that is most likely what will go on. And the 2700X taking away HEDT sales for a couple months, if a person doesn't need the PCIe lanes or the quad channel memory, isn't that bad of a deal considering the new chip will be out in a couple months and the old chip is 8 months old.

 

My point is today the 1900x makes almost no sense except where PCIe lanes are needed over the x470. there is memory too but you can get a lot faster memory on the x470 too. To this point a 2900x makes little sense as well with ZEN-2 coming so close. This is why I think the should get the releases aligned. Get them all too Q2/Q3 or so of 2019.

 

I disagree. Zen 2 release for HEDT (the 3000 series) TR is about 14-15 months out and the 29xxX chips are out in about 2 months. So there is plenty of reason to release it, rather than leaving Cascade-X completely unanswered, which is also released soon (possibly even Computex). And, as to aligning releases, I also disagree. Intel RARELY released mainstream with HEDT. In fact, they often released mobile by itself a month or two ahead of desktop mainstream, then dropped server and HEDT together. You occasionally got mobile at the same time as desktop, but in the past 5 years, it was usually dropped in Nov/Dec for mobile chips, then Jan/Feb for mainstream.

Now, AMD has many different lines with a more diverse release window, if being honest. The primary ones focused on are HPC/Server, HEDT (TR), and Mainstream (Ryzen). They also have the Ryzen Pro series, the Ryzen mobile series, and APUs. It seemed they grouped the mobile and APU releases to roughly coincide, which makes sense. Ryzen Pro is kind of out there, but....

From this, we have to examine timelines. Companies that don't own their fab have to purchase wafer and time on those machines. Intel it is part of their business. Intel also has been known to do massive inventory buildups in the 3 months before release, which makes sense. So, with less competition for fab time (GloFo is number 3, TSMC is #2 but is trading blows with Samsung, which is #1, and Intel is #4) for Intel, they do not likely need the spread on release that AMD needs.

In addition, AMD bins the chips for Ryzen for HEDT TR. That means they are taking chips during the couple months before release of Ryzen, along with any subsequent runs between release and the assembly of the TR package runs. With the unmitigated success of TR last year, it would be silly to let Cascade go unanswered, or to push up release this year when demand may be higher this year and next year for this product, meaning needing more binned chips to meet release demand this year.

With those considerations, I think they are doing it right. On graphics cards, I'm incorporating what the info on the radeon "zen" news from last month means, as I initially brushed it off as nothing. What I need to do is dig on Plummer's role on the Zen team. With that, I may get a better idea on what is going on with the low end for the fall release, potentially, of a 600 series.

 

Not the TR Zen-x, the say 3700x or possible 12 core variant in the Ryzen format. In other words the 2900x will be ib the same spot against Zen-2 Ryzen as where the 1900x is today against the 2700x.

 

(4.3Ghz) R7 2700X vs R5 2600X vs R5 2400G | Tested 14 Games
For Gamers
Published on May 25, 2018

 

AMD EPYC Secure Encrypted Virtualization Not So Secure: Researchers
by btarunr Today, 07:13
Secure Encrypted Virtualization (SEV) was touted as one of the killer features of AMD EPYC and Ryzen Pro series processors.

 

https://www.theregister.co.uk/2018/05/25/amd_epyc_sev_vm_encryption_bypass/

https://securityaffairs.co/wordpress/72929/hacking/amd-secure-encrypted-virtualization-hack.html

 

[Next@Acer 2018] Acer Predator Helios 500, Helios 300 Special Edition gaming laptops announced
Asif Iqbal Shaik, May 27, 2018
https://www.mysmartprice.com/gear/2...laptops-announced-specs-price-next-acer-2018/

"Acer Predator Helios 500, is a completely new high-end gaming laptop. It features a 17.3-inch G-Sync display, up to Intel Core i9, Nvidia GeForce GTX 1070 GPU, upgradeable storage, 16GB of DDR4 RAM, 2.1 speakers, Killer gigabit ethernet and Wi-Fi, and Waves NX 360° audio through headphones.

There will also be a variant with AMD Ryzen 7 2700 GPU, AMD Radeon RX Vega 56 GPU, and FreeSync display."

AMD RX Vega 56 & Ryzen 7 2700 featured in a future variant of Acer Predator Helios 500 laptop
https://www.reddit.com/r/Amd/comments/8mhvw9/amd_rx_vega_56_ryzen_7_2700_featured_in_a_future/

FURIOUS POWER - Helios 500 - Ryzen 2700 + Vega GPU!!
https://www.acer.com/ac/en/US/content/predator-series-features/predatorhelios500

 

That's new old news

As far back as 2016 it was estimated that the SEV wouldn't be 100% secure, and the same hacks were posited back then based on previous hacks:

Researchers Point Out "Theoretical" Security Flaws in AMD's Upcoming Zen CPU
By Catalin Cimpanu, December 9, 2016
https://www.bleepingcomputer.com/ne...ical-security-flaws-in-amds-upcoming-zen-cpu/

And, already demonstrated back in Dec 2017:

Secure Encrypted Virtualization is Unsecure
Zhao-Hui Du, Zhiwei Ying, Zhenke Ma, Yufei Mai, Phoebe Wang, Jesse Liu, Jesse Fang
(Submitted on 14 Dec 2017)
https://arxiv.org/abs/1712.05090
https://arxiv.org/ftp/arxiv/papers/1712/1712.05090.pdf

Intel's SGX:
https://www.google.com/search?q=intel+sgx+vulnerability

 

Ryzen 5 2600X Overclocking & Tweaking Methods Explored, Spend $$$ on Cooling or RAM?
Hardware Unboxed
Published on May 28, 2018

 

Same timeframes. The 3700X, if it has more cores (still rumors), would go up against an 8 month old processor from Aug. of 2018. Couple months later, in July or August, you get the 3900X. This argument makes no sense to me. It LITERALLY is arguing they should not release such a product for canibalizing sales of a product 8 months old, when Intel did the same thing with the 6-core mainstream chip that was 2 months old! (8700K and the 7800X). Do you see how Intel's decision is more questionable from 2017 versus now? Or how about their release of the 8-core chip at the same time as releasing an 8-core chip on HEDT with Cascade this summer. Yet you don't say anything about those. Why? Because, in a way, Intel is playing catch up on cores, not AMD.

Meanwhile, I will get what I said on this in a chat:

"that is going to be the last decent chip from Intel until 2020 or beyond!
Basically, their 10nm is ****ed
and they will be owned by AMD for a couple years
Also, I have a problem with their limits on PCIe lanes on mainstream
They really don't need to, and could be more appealing if they increased that
same is true of AMD

hate the limit on lanes for either of their mainstream platforms

But, I really wonder if they will put out an AM4+ board with 7nm that supports quad channel and like 32-40 lanes on mainstream, but with backward compatibility, so that the older chips run the 16 lanes and dual channel
then increase the TR to six or eight channel memory

these are my desires, not rumors nor what I have divined they are doing
But, two settings, not three+ like Intel tried to do with the KabyX line on both lanes and ram channels, the cut down 6 and 8 core on lines, then the full Skylake-X
So, doing an AM4+ with dual modes depending on the CPU inserted is not out of the realm of possibilities, but they really have to want to screw Intel if they did such a thing
That would practically make Intel's HEDT the same as AMD mainstream, all while leaving even better lane count and ram channels on AMD HEDT, which you would need to move closer to server lines, so think maybe going to 80-96 lanes on HEDT and hexa-channel ram, while servers get the full 128 lanes and 8-channel ram

That is the easiest way to **** Intel in their 10nm down years.
remember, in 2021 or so, DDR5 will be coming with even more bandwidth as well
And redefining mainstream while Intel cannot compete with your 7nm lines is the easiest way to make Intel have to not only catch up, but to redesign to meet your new mainstream offering, kind of how they tried to downplay the 64 lanes on TR.
We were told by media and reviewers "we don't know how you could use all those lanes"
That is due to artificial limits on graphics cards to dual sli and two way xfire
but, for storage, INCREDIBLE!
Also, if you do not need SLI or xfire, like doing movie editing, you can do 4 graphics cards easy on a workstation
but, the lanes were limited by MB companies not doing full 16x traces on all slots, etc.

AMD did a crappy job educating the press about [some of] their products, including the benefits of the extra lanes on HEDT. Luckily, some of them [reviewers] figured out their own ways to use TR efficiently and gave better reviews of the product after they figured it out."

I left out the comments from the person I was speaking to for privacy and because, generally, this is to speak about what my views are, and it speaks to my hopes, not what I think will happen. There are two die revisions. Even if they include the controller chip rumored for EPYC onto the AM4 mainstream chips and TR chips (which the controller chip would contain all IMC, I/O, and L3 cache, and would build on the high bandwidth cache controller from the Radeon group, which we've seen it's ability to manage cache, HBM2, and NVRAM on the card, while theoretically building on the heterogeneous memory tech from years ago, and could be in a position to do that, cutting out the cross-CCX talk to get to the L3 of the other CCX, making only one trip off chip to a much more massive L3, then out to memory if needed rather than jumping to another die to get the other IMC or L3 cache, which can be mitigated depending on what they do with IF and its gearing), they could, due to having two revisions, have a completely different CCX and Core count for the Ryzen and TR dies, which the increased core counts like that come from rumors of the EPYC Zen 2 7nm core count rumors. Yes, the language is a little rough, but it was in chat.

 

My point is to not release the products but align the release so they may have a longer viable marketing life span. I do understand though in doing so it will be longer overall before we see Zen-2 so in the same vein keep Intel on top for a few extra months, not overall what I am looking for but a side effect.

Maybe they are awaiting to get on top of Intel and then align releases. I might be a more comfortable time for them then as well. Much easier to play around when you are number one than playing catchup at the same time.

 

Align the releases? Like all new AM4 parts, all new TR4 parts, and all Epyc parts all at once?

That would be logistical nightmare. They don't have enough resources to do them all at once. It wouldn't be good trying to stretch release of so many different sku's across different product lines just to align new silicon process releases.

Not to mention the foundries trying to release product all at once. Or stock up each run and hold it till all of them are done.

That's not something I would expect as possible, even for Intel.

 

8GB+ 2400G vs 16GB+ 2200G | Tested 17 Games
For Gamers
Published on May 23, 2018

 

AMD Radeon RX Vega 56 available with Acer Predator Helios 500 notebook
Published: 28th May 2018, 10:57 GMT
https://videocardz.com/76295/amd-radeon-rx-vega-56-available-with-acer-predator-helios-500-notebook

"Eleven months after Vega 10 introduction, five months after Vega 10 mobile announcement, it’s finally here: Radeon RX Vega 56 in Acer Predator notebook.

AMD Radeon RX Vega 56 in a notebook
Is this a full-fat desktop variant of Radeon RX Vega 56? Well, it seems unlikely to expect 210W graphics cards in a mobile form factor. It could be downclocked or optimized for mobile use, or be a proper mobile variant (which AMD confirmed a few months ago).

Acer did not reveal any details about the graphics card itself, but Vega 56 clearly stands for 56 Compute Units. It is offered as an option to GTX 1070 variant, which itself is 115-120W graphics solution.

Acer Predator Helios 500 notebook will be offered in few main variants including Intel i9-8950HK + NVIDIA GTX 1070 and AMD Ryzen 7 2700 + AMD Radeon RX Vega 56. This, ladies and gentlemen, is the first true high-performance AMD-only laptop in years.

The Intel+NVIDIA option will support G-Sync technology, while AMD solution will give you FreeSync support. All Helios 500 laptops are equipped with FullHD 144Hz panels.

Prices of Predator Helios 500 notebooks start from 2000 USD (i7-8750H variant). Unfortunately, pricing of i7-8950HK / Ryzen 2700 models has not yet been confirmed.

Source: Acer "

https://www.reddit.com/r/Amd/comments/8ms3l3/amd_radeon_rx_vega_56_available_with_acer/

 

I am not saying a same day release but within say 60-90 days then wait apr. 18 months for another cycle. GPU, low voltage and APU's could be off that specific cycle.

 

I love my laptop and the idea of continued upgrades with it but I must admit I am tempted by the Acer.

 

Now isn't the time to do a longer release cadence. They just got it down to a 12 month release cadence. Plus, the HEDT is released approximately 120 days after the mainstream. The last TR release was their first foray into HEDT. Right now, they need the 12 month release and pulling off 7nm to basically take massive amounts of market share from Intel. It sounds more like you want HEDT, which uses the mainstream binned chips, immediately on release, including saying sit for 18 months, an extra 6 months on their release cadence, or even the 4 months to wait for TR, to match Intel's release head on, which then costs money to store the packaging and inventory of chips, which can be very large costs to a company, to accomplish what?

In fact, the only one they pushed up, and skipped a node, is EPYC, for clear competitive reasons. They had partners trying test beds for release from around Dec. or January before release, with more signed on by march or april. They did a soft release, as there was low availability of EPYC on release. It took until OEM partners integrated the chips into systems closer to December to pick up sales, with the majority of sales on the product to come this year. Now, they are sampling in November or December, most likely, and doing volume around the time that they announced the release next time, skipping the iterative 12nm node. All three of these chips are shipping within a third of a year to each other, with the order of importance to sales being EPYC, then Ryzen (and Ryzen Pro, if being honest), then TR for HEDT. The HEDT is higher margin, which helps make up for the lower volume and does make it an important contributor to their revenue. But, the goal was to get the names out and get back in the game known as a challenger. The next couple steps was to build on that good will with competitive products while closing the performance gap. They were not sure when they layed out their roadmap that Intel would mess up 10nm. That is just a happy coincidence. Now, you are suggesting taking the foot off the gas or altering the plan all to benefit Intel, which any delay in 7nm mainstream or EPYC chips would signal struggles on 7nm, thereby giving breathing room as analysts declare anything below 12nm is plagued. Does that make sense to do, which can burn good will with the market and with investors?

I think messing with the CPU side right now would be a folly. Instead, using this time to create more powerful GPUs, trying to get those onto regular cadences for release, and clearly stratifying the products is what needs to happen. When watching the live stream of Gamers Nexus OCing the Intel NUC with Vega graphics, he gave scores related to AMD and Nvidia as comparison. That integrated Vega (which is almost part polaris, part vega), was performing at between a 1050 Ti and 1060 on the Nvidia side, then between the 560 and 570 on the 500 series side. Nvidia's lineup goes -> 1050 -> Ti -> 1060 -> 1070 -> Ti (newer) -> 1080 -> Ti -> Titan. AMD's goes -> 560 -> 570 -> 580 -> Vega 56 -> Vega 64 -> Vega 64 liquid edition. That is 8 cards for Nvidia and 6 for AMD. In fact, the Vega 64 and LE cards are practically the same card with different cooling. You could include the frontier edition to go against the Titan, which give 7 (or 6 if you count both variants of Vega 64 as the same), but the stratification is a bit off. You could also point out that the 1070 Ti was an abomination created as a mid-point without reason, all so that they could calm their fears related to the Vega 56 taking market share from those at the 1070 level. There was nothing really Ti about it. It was a moderately pared down 1080 with slow ram. So, the moving of members of Zen over to help with the graphics card designs to try to get a 600 card that isn't just a rebrand out this year (now rumored to be the Vega 12nm cards), while working on Navi for next year (which could be an April/May release, a June release (E3), a July/August release, or a September-November release (these are just historic points that make sense for release of the product). I'm not really addressing mobile releases, which have happened in Dec-Feb before. I'm also not addressing the lower end of cards right now. What I am wondering is how they will flip the script to get Nvidia playing their game, instead of them going up against Nvidia's stack. People have been saying Nvidia refuses to release a consumer card because of lack of competition. I call ********. Nvidia immediately responded to losing between the 1070 and Vega 56 to putting out a new card in that slot (thereby creating that slot with an abomination of a card). Meanwhile, we've seen the Titan V and how little extra it gave in performance compared to Pascal, in regards to gaming. We know they are making the new 1180 with the same amount of "cuda cores," which are shader processors, as the 1080 Ti. In effect, they are having to slide the stack back closer to what it was a decade ago, instead of using the mid-cards as high end, because they hit another design patch of crap. They have been riding high since around Kepler and Maxwell timeframe, even with the heat problems of the 400 and 500 series, then the abandoned 800 series due to heat, followed by the Maxwell 900 series cards, and the apparent stretching between releases, which no one has talked about, but loves to hit on AMD doing rebadges with minor tweaks to their cards each year which at least gives something of value, even if de minimis. Then, with the discovery of the math rounding errors on the Titan V (which might be a problem with Volta in general), the lack of uplift in performance relative to cost, the delay in supplies, which were memory related, in part, but may also have been yield related, considering when AMD went from no products to limited products during last fall, but it took until Dec. for Nvidia to get out the Titan V with way larger costs (nearly 3X that of the Titan X), followed a month or two later with the quadro lines, which arrived after the rounding problem with Titan V was known, I think people are giving way too much leeway to Nvidia and need to turn a critical eye. This is why they announced they are going to a multi-die solution, because of yields on the monolithic dies right now. No one says that second part, about the yields, but it makes a lot more sense. People act like Nvidia has destroyed AMD when the graphics market has more competition than Intel's monopoly did, with about 99% of the market before Ryzen. Nvidia isn't a miracle maker, they hit one good design with Maxwell, then did iterative improvements and die shrinks since then. Nothing revolutionary has come from them, except for tensor cores, during that time period since. Reminds me of AMD riding the wave with Phenom, which the folly of acquisitions stretching R&D too thin nearly led to the demise of the company.

But, back to AMD. With the inclusion of Super-SIMD in with Navi, if done correctly, that could theoretically close the design gap on how you get from their current tech to a lower end chip giving 1080 performance, which can then be used in a multi-die configuration moving forward on 7nm. That and memory bandwidth, that is. But this is why I want to know more about Plummer's role in bringing Zen to market, including which elements of the design she worked on (the redesign or improvements of the HBCC, the interconnects of IF, etc.). If AMD, at least short term, plans on deprecating Vega to the 600 series and low end GPUs, including the redesign to include a GDDR5 (and GDDR5X) memory controller, followed by Navi hitting the mid to high end, depending on the die count, maybe with variants based on specific common defects to reduce the number lost to yields, that sounds like a solid business strategy. They need to push back on the Nvidia facade. If they do, you could see bad reviews of Nvidia soon, considering what is known of their next lineup makes me sad for the company.

Edit: to be clear, I am not saying AMD is ahead on graphics, I am saying Nvidia seems to have stagnated to a degree, which leaves an opening for AMD.

 

In my mind the whole Intel/AMD discussion is really simple, if you don't want more than 6 hyper-threaded cores then buy Intel, otherwise get AMD for the more cores. I think this because Intel excels at performance per core, but AMD excels at cost per core. Intel's 8 core plus models are really expensive & I'd see AMD's offerings as more competitive in that arena currently.

 

There is a little more to it than that, but basically, yes.

With the 8700K being able to clock to 5GHz, it actually goes head to head with the 8-core Ryzen 2700X. So it really comes down to use case. If just gaming, viewing internet sites, etc. Intel is the one to go with. On productivity, it depends how the software was optimized. If it is really good at multi-threaded workloads, then AMD wins many times (but considering Intel has been optimized for for decades, that doesn't always hold true). If it needs higher IPC/single core speeds, then Intel is almost always the winner. But, then you see the 2700X at 4.2GHz beating Intel 7820X running at 4.7GHz sometimes. It really is use dependent and why a person needs to understand their workloads when picking for the absolute best for their needs.

 

thing is, most gamerbois with mainstream desktops rather go for the big clock numbers instead for real, actual, overall performance. so whatever overclocks better, thats the buy they go for

 

Nvidia is holding back, one would assume, rather than not having higher performance products waiting to go into production.

Nvidia isn't in the same bad place as Intel as far as process node and foundry troubles, so I expect Nvidia to be able to counter AMD's GPU wonders for the next couple of generations moving forward.

Hopefully AMD will continue to shave off the differences and get closer and closer, but I would be - and most everyone else would be - surprised to see AMD one up Nvidia in performance any time soon.

That doesn't mean you have to buy Nvidia, you can buy AMD GPU's and have just as much fun, if you can find them - and they are at MSRP. For now the miners are killing gamer access to AMD GPU's.

It's sad.

 

Most games don't multithread as well as Doom in Vulkan for example, so they would be justified in going for the highest clocks since that usually means the best performance.

 

sure, but thats kinda shortsighted. in the long run, more cores will serve you better than high clocks. especially with higher core count cpus flooding the market (and continuing to do so in the near future), games will most likely be more optimized towards higher core counts.

in the end, id gladly accept a "loss" in performance of like 5% right now due to lower clocks rather than taking a hit of 50-100% due to less cores in the near future

 

In a gaming scenario, the performance delta will never be that big when you're talking the difference between 2 extra cores at 6+ core count.

 

I'd gladly sacrifice a good chunk of frequency for 2 extra cores. It's revolting Intel milked quad cores for this long.

 

More news on TSMC ang GLOFO 7nm processes:
TSMC and GloFo may compete to produce the most overclockable AMD Zen 2 CPUs

https://pcgamesn.com/amd-globalfoundries-tsmc-7nm-production?amp

 

Yes and no here. They were not about to just give away the shop, this is where Ryzen put Intel to the test. Now clocks and core counts are much more reasonable. I say just let the fight continue.

 

yep, AMD forcing intel to go for more cores more quickly, whereas Intel´s high clocks pressures AMD to improve on their max clocks (which they did with Zen+). good times

 

Por que no los dos?