When will we see an 8-core Ryzen 5?
What could the next few months hold for CPUs?
I think this is something that's been on a lot of people's minds lately especially with Intel's 10th gen coming out recently. I even went as far in a recent blog post as saying right now is the best contest in CPU history... at least until Zen3 comes out that is. No matter which way you cut it, after such a long period of time with only incremental updates from Intel who held a nearly uncontested lead for a decade (with only a brief discussion between Sandy Bridge and Bulldozer in 2012), the events unfolding right now are making history, and the results of which will determine the landscape for a long time to come.
Note: This is almost entirely an opinion piece. I don't have any industry sources or fresh leaks. I'm simply looking at available information and offering my personal analysis.
AMD holds the performance advantage with Zen2
Right now, it's no secret that between AMD and Intel, it's team blue that's on the back foot. In that mentioned blog post I take a deep look at the current CPU landscape and make some predictions for how 10th gen would turn out. I was overall very close, and Intel both surprised and disappointed me with better than expected performance, lower than expected temperatures, higher than expected prices, and lower than expected availability. Ultimately their most compelling chip is the 10600k, which we think will hit stores with a $280 price tag, but don't even know for sure because it literally does not exist. No store has had a single listing for it go up.
Before Zen2, Intel held the per-core advantage. If you wanted the highest performance regardless of the cost, you bought Intel. If you wanted great, but not-quite performance at much better prices, you bought Ryzen. This changed as Zen2 started taking most if not all per-core productivity workloads, and with 12 and 16-core options, utterly eclipsed Intel in performance across the board except in gaming where Intel maintained that slim edge due to its ring bus and lower RAM latency. Intel has since done everything in their power to flaunt that gaming advantage. Even after Intel's 10th gen, we're still looking at much the same landscape as before.
Enter, Matisse 2
It still blows my mind that they're doing this, if rumors are accurate. Depending on who you listen to, AMD is preparing either a 3750X and 3850X, or a 3600XT, 3800XT, and 3900XT (the latter of which does sound 'more' plausible to me.) It seems to be little more than AMD taking the bait of Intel's paper launch, but in my opinion, just like the 10th gen launch, really doesn't do a whole lot to change the landscape. The best that we can hope for in my opinion is an incremental 100-200MHz clock speed bumps. Going with the XT rumor, thus refreshing the 3600X, 3800X, and 3900X, this is what we're currently looking at:
CPU
|
Base Clock
|
Single-core boost
|
Overclock
|
3600X
|
3.8GHz
|
4.4GHz
|
~4.2GHz
|
3800X
|
3.9GHz
|
4.5GHz
|
~4.3GHz
|
3900X
|
3.8GHz
|
4.6GHz
|
~4.3GHz
|
The highest clock speed in the entire Zen2 lineup is the 4.7GHz on the very tightly-binned 3950X, which I seriously doubt we're going to breach. Some people are claiming we'll see a 200-300MHz clock speed increases, but I think that's very ambitious. It's possible that it's on a more mature version of the node with better binning, as well as improved boost algorithms.
AMD doesn't advertise the maximum boost clock that their CPUs will reach all-core, as it has to with several factors including cooling and power draw, but I just tested a stock 3600X with a stock cooler on it in an X470 board, and it hit 4.05GHz all-core in cinebench despite a 3.8GHz base clock. Based on this I think it's likely that AMD could get some marketing wins by advertising 4.0 or 4.1GHz base clocks given some slight improvements in binning.
As for actual performance increases, I expect it to be extremely minor over the current, as AMD didn't leave a whole lot in the tank the first time around. I could see a 3600XT hitting 4.6GHz single core, and a 2900XT hitting 4.7GHz, but nothing radical, and certainly nothing that can close the gap on Intel's gaming edge.
Zen 3 timeline?
Currently, rumors are saying that Zen3 is expected to launch in September. Given that Matisse 2 is expected to launch in July, this really doesn't leave much room. Based on the expected launch window of Matisse 2, I wouldn't be surprised if Zen3 was late, maybe october or november. However that is speculation. I think we will continue with the assumption that Zen3 is on schedule for September for now.
Zen 3 architecture and clock speeds and rough performance guesses
Not a whole lot is known about Zen 3. We know that it's a more major redesign with significant IPC improvements, and we know that it has a unified 8-core CCX (ultimately rendering the term CCX redundant) and along with it a unified 32MB cache. While it's the same amount of cache per core, the unified cache does make more data available to each core. There is no increase expected in number of cores or threads. It's my personal opinion that we're going to see a further evolution of Infinity Fabric which will further reduce the latencies for core to core communication, die to die communication, and notably for gaming, memory accesses.
Word on the street is anywhere from 15 to 20% IPC increase, but let's be conservative and assume a roughly 15% IPC boost. In terms of node, I think we're looking at N7+, which is claimed to offer 10% higher clock speeds at a specified voltage (this does not mean 5.1GHz confirmed, when they say 10% it's typically at a lower voltage and clock speed that makes the process look good, not necessarily at the top) however, we may see base clock speeds around that 4.0GHz mark, single core boosts around 4.6-4.8GHz, and all-core overclocks I'd guess in the ~4.5GHz region.
Overall, I'd say we should expect that given the same number of cores, Zen 3 should perform on average 20% faster than Zen 2. This would mean an 8c/16t Cinebench R20 run in the ~5800 Cinebench R20, just short of the 10900k's score, but with a leading ~600 single-threaded score.
Like everything else, these are vague estimates based on information that everyone has at their disposal, so take them with a shovel-full of salt, and watch out for the reviews when they come. However, I think it would be more controversial than not at this point to suggest that AMD will not properly eclipse Intel's core architecture and 14nm node in performance in nearly if not every notable metric. I think the best Intel can hope for in gaming after Zen 3 is only the occasional win.
Zen3 segmentation and pricing
I have a somewhat controversial stance on Zen3. AMD absolutely has the performance crown here, per core, and per segment, as outlined above. To many, this means that AMD will not increase the core counts per segment. They might be right, too, after all, since they will be at the top, there's no need to increase core counts. AMD could release a $500 12-core Ryzen 9 4900X and it would sell, because it would beat out everything from Intel, in every metric, making the 10900k a very hard sell for any purpose. It would be easy, it would be clean, and it would be profitable. Shareholders are likely pushing for exactly this.
8 cores could be coming to midrange
I have several points on why I believe AMD should increase the cores in the Ryzen 5 lineup as of Zen 3 launch to have 8 cores.
The threat of Intel
First, is competition with Intel. AMD has largely built the brand of Ryzen around having higher core counts per segment, which offers better multi-core performance at lower prices. With Intel's 10th gen, they've finally largely caught up in core counts. While the i9 lags, the i3, i5, and i7 match the 4c/8t, 6c12t, and 8c16t of their ryzen counterparts, respectively. In the face of competition from AMD, Intel has consistently increased core counts every release since Ryzen. If Ryzen does not increase core counts per segment with Zen 3, they leave the door open for Intel to increase their core counts per segment. With Intel lagging behind in per-core performance, I would see that as a necessary step for Intel to claim some advantage, if AMD gives them the opportunity.
Consoles bring 8 cores to everyday gamers
First is consoles. In November (probably) we're going to be getting two consoles each equipped with 8 cores and 16 thread Zen 2 CPUs. Consoles are typically around midrange, as they have to target a price of $400-500. While the Zen 3 6 cores will be closest to the consoles in all-core performance (like how we see a 6-core 3600X nearly matching the 8-core performance of the 2700X), there's a lot to be said of marketing. With the release of the consoles, PC gamers are going to be looking to upgrade their rigs to specs that they know will not just match but stay ahead of the consoles.The market is ripe for the core increase
For a decade, Intel released quad-core CPUs as the default core-counts in mainstream CPUs. AMD burst on the scene in 2017 with the first 8-core mainstream part. Now it's AMD's opportunity to say that 8-core part is midrange and offer the pricing to match. Furthermore, according to Steam Hardware Survey, 48% of users still have 4-core CPUs in their machines with barely over 7% having more than 6c. This means if they released an 8c midrange CPU with modern features and performance, there's a huge market of people that would benefit from upgrading to that, and making it affordable would only increase the draw for them to do so, and allow AMD to sell an incredible number of chips.
They can likely maintain adequate margins
It's currently expected that Zen 3 will be manufactured on TSMC's N7+ node, which offers 20% density improvement over N7 which is what Zen 2 is manufactured on. It also offers notable improvements in clock speed and power efficiency over N7, which will help Zen 3 hit those higher clock speeds at the same or lower power. The increase in density, as long as the cost per wafer is not increased, will increase yields and decrease prices. Even on N7, the yields were so high that it took nearly a full year before it made sense to release 4-core Zen 2 parts. N7+ is only going to be better.
They still have room to grow in the following generations
I'll be talking more about Zen 4 in a future article, but as a teaser, it's very likely that we're going to see more cores on the package. With further die size reductions for both CCDs and the IOD, it should be very possible to fit at least 3 CCDs on the AM5 package, for 24 core designs. At this point, I think it would be controversial to say that Zen 4 will not have at least 8 cores on Ryzen 5-tier, the question is when they'll make that change; Zen 3 or Zen 4. Given that Zen 3 doesn't have any more cores on it to offer at the high end but Zen 4 will, to me it would make the most sense to split the hype between them, offering better segmentation of the cores in Zen 3, and then expanding the high end substantially with Zen 4.
If I was head of AMD...
This is what I personally would do:
Ryzen
|
Cores
|
Threads
|
Price
|
TDP
|
3 3300X
|
4
|
8
|
$100
|
65W
|
5 3600XT
|
6
|
12
|
$140
|
115W
|
3 4300X
|
6
|
12
|
$160
|
95W
|
7 3800XT
|
8
|
16
|
$200
|
125W
|
5 4500X
|
8
|
8
|
$220
|
95W
|
5 4600X
|
8
|
16
|
$260
|
95W
|
9 3900XT
|
12
|
24
|
$300
|
125W
|
7 4700X
|
12
|
12
|
$330
|
95W
|
7 4800X
|
12
|
24
|
$400
|
95W
|
9 4900X
|
16
|
32
|
$500
|
115W
|
After Zen 3 comes out, drop prices on Matisse 2 refresh, and offer then as entry levels into SMT. Some people will disagree with my choice to segment using SMT, but having SMT off allows for better temps, and higher clock speeds, which those units should leverage, being slightly faster per core than their SMT-enabled counterparts with the exception of the 4900X which would not have an SMT disabled counterpart because it's intended to be the top. Like the 3950X it replaces, it should be reserved for the best-binned dies, and have the highest peak clock speeds. This lineup is also designed for that 16 core part to sell in greater numbers and give more tangible improvements for buying higher. The current 3800X makes very little sense in the face of the 3700X being nearly as high of a performer for much less money.
This is also not a great departure from the current prices. Currently, you can buy a 3900X for $390, a 3700X for $295, and a 3600 for $160.
What do I predict AMD will do?
Well I expect them to continue to make some of the questionable naming and segmentation decisions they're known for. I also don't actually expect them to segment on SMT (to the relief of some, I'm sure), maybe like this:
Ryzen
|
Cores
|
Threads
|
Price
|
TDP
|
3 4300X
|
6
|
12
|
$160
|
65W
|
5 4600
|
8
|
16
|
$220
|
65W
|
5 4600X
|
8
|
16
|
$260
|
95W
|
7 4700X
|
12
|
24
|
$360
|
65W
|
7 4800X
|
12
|
24
|
$420
|
105W
|
9 4900X
|
16
|
32
|
$500
|
105W
|
While this may seem better for some as it offers better prices for SMT-enabled chips, if you notice the difference in TDP, that will affect how the chips perform. I don't see a lot of value in 65W chips for desktop, though AMD seems to like segmenting on power consumption instead of SMT. While I disagree with it as a decision, I would not be disappointed to see this lineup at these prices.
Also, you may remember discussions around AdoredTV's leaks for Zen2. I'm not convinced those leaks were entirely illegitimate. While they were out to lunch on clock speeds, I think AMD was considering that segmentation for Zen 2. These prices and core counts are a lot like what he was predicting for Zen 2, and I think it has a good chance of panning out this time.
And, just like Jim, I could be completely out to lunch as well. Just because this is what i think would be a good idea doesn't mean it's the same decisions AMD will actually make. We could easily end up seeing a Zen 3 8-core Ryzen 7.
And, just like Jim, I could be completely out to lunch as well. Just because this is what i think would be a good idea doesn't mean it's the same decisions AMD will actually make. We could easily end up seeing a Zen 3 8-core Ryzen 7.
Thank you for reading!
As always, come visit me in Discordland! I've got a little flourishing community that's so far been rather active and we've had a lot of great tech discussions with, and I'd love more people around to tell me why I'm wrong.
Twitter: https://twitter.com/MeyerRants
thnks meyer! keep up the good work!
ReplyDelete