AMD Ryzen 2600 Eng Sample leak

[MaktimS]

17 minutes ago, suits said:

Honestly anything less than 4.5 boost/OC is going to be a let down. And I’m really banking on being let down. 

You will be disappointed. It's just a small refresh. 4,5 GHz is best case scenario. I am expecting something around 4,2 max. The architecture is still the same. Clocks around 4,5 or more will not come until Zen 2 in my opinion

[Trixanity]

43 minutes ago, MaktimS said:

You will be disappointed. It's just a small refresh. 4,5 GHz is best case scenario. I am expecting something around 4,2 max. The architecture is still the same. Clocks around 4,5 or more will not come until Zen 2 in my opinion

Depends on what AMD has done with the 12nm process. They'll get 10% just from the process itself (that would give them 4.4 GHz if we assume 4.0 is the rough ceiling right now). They also get 15% reduced area. If they use that to lengthen the pipeline and eliminate other bottlenecks they could potentially be adding another 200 MHz or more. This is of course best case scenario.

They'd probably need to do extreme binning to get a 4.5-4.6 GHz top SKU. Depends on whether they reserve those chips for Threadripper though.

[MaktimS]

19 minutes ago, Trixanity said:

Depends on what AMD has done with the 12nm process. They'll get 10% just from the process itself (that would give them 4.4 GHz if we assume 4.0 is the rough ceiling right now). They also get 15% reduced area. If they use that to lengthen the pipeline and eliminate other bottlenecks they could potentially be adding another 200 MHz or more. This is of course best case scenario.

They'd probably need to do extreme binning to get a 4.5-4.6 GHz top SKU. Depends on whether they reserve those chips for Threadripper though.

I read somwhere that the 12nm architecture isn't really 12nm that it's more like 14nm+ and that the 12nm name is just Global foundries' marketing.  You can find it on older AMD presentations where AMD referred to Ryzen 2 as 14nm+, so I expect jump in performance to be similiar to jump from Skylake to Kaby Lake, but I'd love to be wrong

[Trixanity]

3 minutes ago, MaktimS said:

I read somwhere that the 12nm architecture isn't really 12nm that it's more like 14nm+ and that the 12nm name is just Global foundries' marketing.  You can find it on older AMD presentations where AMD referred to Ryzen 2 as 14nm+, so I expect jump in performance to be similiar to jump from Skylake to Kaby Lake, but I'd love to be wrong

Their 12nm process is advertised as 10% higher performance and 15% reduced area but at the same power (ie. no real power reduction). So it's somewhere in the half node range. The difference between Skylake or Kaby Lake is miniscule.

[Inversion]

On 18/01/2018 at 2:21 PM, bruny06 said:

I just built myself a new RYZEN system with an R5 1600 and I must say I'm quite happy with it, although, I should also say my previous Core i5 2500 at 4.1GHz had a slightly higher single-core performance against my current R5 1600 at stock speeds, which is why I think it would be great to see a clock speed improvement to boost the single-core performance. But still, it performs great in gaming and I don't see the need for a better CPU at the moment. My initial plan was to buy a RYZEN+ CPU when they came out as I already had RAM and MB ready for the upgrade, however, I didn't have a current RYZEN CPU to do the BIOS upgrade, so I guess I'll have to wait for RYZEN 2 in 2019? or is it 2020?

I'm in a pretty similar boat to you although my 1600 is better in every way over my old Athlon X4 860K system. I think I'll probably wait for at least Ryzen 2 and get it just about the time Ryzen 3 comes out for the best price or do the same with Ryzen Gen 3 as whatever comes after is released cause I'll be a student at the time and dropping full price on it is a bit unjustifiable when on a tight budget most of the time. I don't see myself needing a MOBO or RAM upgrade unless I desperately need USB type C (although I'll probably be able to get a PCIe expansion card for that). I love what AMD is doing now and they're still sticking to the budget option plan they've pioneered for so long despite now having competitive stuff. I just need Ryzen Gen 3 to be compatible with the B350 chipset.

[porina]

Base clocks don't tell us much. More interesting is the max turbo clock, that could be indicative of what they're capable of. I'd like to see 4.2+ at sane voltages, not like currently where you have to run them to voltage hell and back.

 

I'll probably get one to play with regardless...

[qepsilonp]

3 hours ago, suits said:

Honestly anything less than 4.5 boost/OC is going to be a let down. And I’m really banking on being let down. 

Well then you are going to be let down, if you already have a Ryzen processor with the right number of cores for you Ryzen was always not going to be a worth while upgrade, wait for Zen 2 / Ryzen 3 like I am if you already have a Ryzen CPU or have a CPU which is good enough for purpose, Zen+ was made so that AMD could stay semi competitive between the long wait between Zen and Zen 2 as a stop gap, as 12LP is really only an improved 14LPP process and that's the only "big" difference, there are the tighter cache timings which should improve AMD's performance in games at extremely high frame rates.

Which will make AMD look better when the tech press run 720p benchmarks with a Titan XP, although it will also help in more reasonable situations also like if you want to run at 144Hz which Ryzen does have a little bit of touble getting to those kinds of frame rates due to the cache latency, memory latency and fabric latency. Although the last two memory latency and fabric latency are the same thing, because the IMC is connected to the rest of the CPU via the infinity fabric, but AMD also promised better RAM support so if you can for example run 3600Mhz memory you improve the latency of the fabric dramatically.

 

But I hope it is something that is properly address in Ryzen 3 / Zen 2 because in specific titles it's even a problem at what are reasonable frame rates, 90 to 120fps because the games are more latency sensitive, this is the optimisation issue you hear in reference to Zen, it's games over using the L3 cache rather than using the L2 cache

 

Although it is a weakness in the Zen architecture, so it's not so much a optimisation issue as just something that Intel is better at, but at the same time it isn't necessary to be so dependent on the L3 cache, it's the same issue that Skylake X had, but as Intel move to higher and higher core counts for the mainstream they wont be able to rely on the old ring bus anymore, so I think game developers need to start moving the dependence of there game engines to L2 cache rather than L3, it is something you need to do anyway to get higher core scaling which means it's probably already being done.

 

This is also the reason AMD holds up much better in content creation because they have a 512KB L2 cache vs Intel's mainstream parts which have 256KB of L2 cache, which as said in applications that scale well with many cores it is necessary to rely on the L2 cache and because AMD also have a low latency link between all the L2 caches on a CCX when you do for example a Cinebench run on all cores a 1800X will beat a 7700k by almost exactly double even though on a single threaded test the 1800X will lose by about 15%.

[Taf the Ghost]

3 hours ago, Trixanity said:

Depends on what AMD has done with the 12nm process. They'll get 10% just from the process itself (that would give them 4.4 GHz if we assume 4.0 is the rough ceiling right now). They also get 15% reduced area. If they use that to lengthen the pipeline and eliminate other bottlenecks they could potentially be adding another 200 MHz or more. This is of course best case scenario.

They'd probably need to do extreme binning to get a 4.5-4.6 GHz top SKU. Depends on whether they reserve those chips for Threadripper though.

The best binned TR parts can get to 4.3 Ghz before the voltage runaway happens. The Golden Ryzen 2000 chip might hit 5 Ghz under water.

3 hours ago, MaktimS said:

I read somwhere that the 12nm architecture isn't really 12nm that it's more like 14nm+ and that the 12nm name is just Global foundries' marketing.  You can find it on older AMD presentations where AMD referred to Ryzen 2 as 14nm+, so I expect jump in performance to be similiar to jump from Skylake to Kaby Lake, but I'd love to be wrong

 

3 hours ago, Trixanity said:

Their 12nm process is advertised as 10% higher performance and 15% reduced area but at the same power (ie. no real power reduction). So it's somewhere in the half node range. The difference between Skylake or Kaby Lake is miniscule.

14nm to 14nm+ for Intel (so Skylake to Kaby Lake) was very minimal. Seems more like an improvement in Yields more than anything else. The 14nm++ of Coffee Lake is one of the biggest intra-node boosts we've seen. Taken with Intel's bad 10nm, the 14nm++ is actually better than a node-shrink. Yikes. (Apparently there's going to be a 14nm+++ at some point for certain products. Whoever was in charge of finding improvements in Intel's 14nm node deserves all the bonuses they get.)

 

For AMD & GloFo, the 14nm -> 12nm transition is something of a half-node improvement, but the main thing is that the Pinnacle Ridge parts are on different libraries than the Summit Ridge parts (Ryzen 1000/TR/Epyc). This should allow for better top-end clocks, but that might be under OC conditions. A nice base clock boost should happen, but the important question is where the OC potential hits a wall. I'm expecting the top SKU 2800X (?) to be something like 4.0 Ghz base, 4.4 Boost, and it probably OCs to around 4.8 GHZ before hitting the voltage wall. Taken together with the improvements in the Cache & other IF tweaks, this will probably be a larger real-world & gaming bump than people expect.

 

We'll know more in February when people start testing the 2400G & 2200G against the comparable Ryzen 1400 & 1200 parts.

[Crunchy Dragon]

On 1/18/2018 at 9:21 AM, bruny06 said:

I just built myself a new RYZEN system with an R5 1600 and I must say I'm quite happy with it, although, I should also say my previous Core i5 2500 at 4.1GHz had a slightly higher single-core performance against my current R5 1600 at stock speeds, which is why I think it would be great to see a clock speed improvement to boost the single-core performance. But still, it performs great in gaming and I don't see the need for a better CPU at the moment. My initial plan was to buy a RYZEN+ CPU when they came out as I already had RAM and MB ready for the upgrade, however, I didn't have a current RYZEN CPU to do the BIOS upgrade, so I guess I'll have to wait for RYZEN 2 in 2019? or is it 2020?

I would advise you flash the BIOS so it has support for the new CPUs, then you can just upgrade to Ryzen+ whenever you want after they hit the market.

[Trixanity]

2 hours ago, Taf the Ghost said:

We'll know more in February when people start testing the 2400G & 2200G against the comparable Ryzen 1400 & 1200 parts.

Aren't the APUs 14nm? I was under the impression that they're technically halfway between Ryzen 1 and 2 but are called 2 to fit in with the 2018 lineup and to highlight the improvements of course.

[qepsilonp]

On 18/01/2018 at 2:21 PM, bruny06 said:

I didn't have a current RYZEN CPU to do the BIOS upgrade, so I guess I'll have to wait for RYZEN 2 in 2019? or is it 2020?

Erm I think your talking about Zen 2, Zen 2 will be Ryzen 3 as Zen+ is Ryzen 2 and it will almost certainly come in 2019, although my guess would be at easilest it will come is May 2019, as on AMD charts right now they have Zen 3 coming in 2020, the chart isn't numbered but it ends in 2020 and it has Ryzen 3 on it, also given comments by AMD saying that they will have new CPU's on AM4 until 2020 to 2021 late 2020 for Zen 3 makes sense, this was back when Ryzen had just launched though so there was less certainty than there is now.

 

Although I believe Zen 3 will not be compatible with AM4 because I would think they would want to jump on PCIe 5 and DDR5 and yes I am actually talking about PCIe 5, PCIe 4 was orignally planned for 2013 I think, but they had a load of toubles, but according to PCI SIG, PCIe 5 is going to be finalised in 2019, given usual development times devises and motherboards should be available about a year after that.

 

So if Zen launches during 2020 or after it AMD will want to implement PCIe 5 and will simply skip PCIe 4, also DDR5 is another thing they will want to implement which will by available by then, which would make the new CPU's incompatible with the old boards, unless AMD wanted to waste quite a bit of die space and a lot of development time having the CPU be compatible with PCIe 3, PCIe 5, DDR4 and DDR5, then make a new series of boards that support the PCIe 5 and DDR5 which AMD is already behind Intel on there IMC unnecessarily increasing complexity would be dumb as you would likely also take a performance hit.

But imagine AMD gave you 20 PCIe gen 5 lanes right to the CPU... that would be nuts you could run 10 Titan Xp's off that much bandwidth as 2 PCIe gen 5 lanes would equate to roughly 8x PCIe gen 3 lanes, and there is not a measurable difference for running in 8x vs 16x with the Titan Xp.

Like 10 virtual machines all with there own Titan Xp assuming AMD move to 12 cores you could only have 1 core each, but still, that is a stupid amount of bandwidth, or ThreadRipper with 64 PCIe Gen 5 lanes thats 32 Titan Xp's or 16 Titan V's, Epyc 64 Titan Xp's or 32 Titan V's.

Thats Nuts, I think even data centre / deep learning guys would even be like, "THAT'S TOO MUCH BANDWIDTH, WHAT AM I SUPPOSED TO DO WITH THIS?"

[Taf the Ghost]

1 hour ago, Trixanity said:

Aren't the APUs 14nm? I was under the impression that they're technically halfway between Ryzen 1 and 2 but are called 2 to fit in with the 2018 lineup and to highlight the improvements of course.

They have the Zen+ optimizations on the exact same process node & production libraries as all of the Ryzen 1 parts, which is why we'll know a lot about Ryzen 2000s series from poking them with a bunch of sticks. The Ryzen vs Intel difference isn't down to IPC; it's down to some structure, latency & optimization differences that Nvidia's driver uses to exploit game engines for faster performance.

 

People already forget that X299 showed the same type of issues, in most games, that Ryzen has shown, at least until the patches started rolling out. You could solve a lot of the problem with some uncore OC, as it dropped the latency down enough that the optimizations for the Core uArch caught up, but it took work for a lot of the CPUs to really keep up in a solid number of games. We'll see what the first generation of enhancements will do with the G-models, but we won't know exactly what the full IF effect is until we get the Ryzen 2000 models.

 

Still, if AMD can find 15% in Gaming, they oblierate Intel out of everything but the 8700k OC'd + 1080 TI market.

[Taf the Ghost]

2 hours ago, qepsilonp said:

Erm I think your talking about Zen 2, Zen 2 will be Ryzen 3 as Zen+ is Ryzen 2 and it will almost certainly come in 2019, although my guess would be at easilest it will come is May 2019, as on AMD charts right now they have Zen 3 coming in 2020, the chart isn't numbered but it ends in 2020 and it has Ryzen 3 on it, also given comments by AMD saying that they will have new CPU's on AM4 until 2020 to 2021 late 2020 for Zen 3 makes sense, this was back when Ryzen had just launched though so there was less certainty than there is now.

 

Although I believe Zen 3 will not be compatible with AM4 because I would think they would want to jump on PCIe 5 and DDR5 and yes I am actually talking about PCIe 5, PCIe 4 was orignally planned for 2013 I think, but they had a load of toubles, but according to PCI SIG, PCIe 5 is going to be finalised in 2019, given usual development times devises and motherboards should be available about a year after that.

 

So if Zen launches during 2020 or after it AMD will want to implement PCIe 5 and will simply skip PCIe 4, also DDR5 is another thing they will want to implement which will by available by then, which would make the new CPU's incompatible with the old boards, unless AMD wanted to waste quite a bit of die space and a lot of development time having the CPU be compatible with PCIe 3, PCIe 5, DDR4 and DDR5, then make a new series of boards that support the PCIe 5 and DDR5 which AMD is already behind Intel on there IMC unnecessarily increasing complexity would be dumb as you would likely also take a performance hit.

But imagine AMD gave you 20 PCIe gen 5 lanes right to the CPU... that would be nuts you could run 10 Titan Xp's off that much bandwidth as 2 PCIe gen 5 lanes would equate to roughly 8x PCIe gen 3 lanes, and there is not a measurable difference for running in 8x vs 16x with the Titan Xp.

Like 10 virtual machines all with there own Titan Xp assuming AMD move to 12 cores you could only have 1 core each, but still, that is a stupid amount of bandwidth, or ThreadRipper with 64 PCIe Gen 5 lanes thats 32 Titan Xp's or 16 Titan V's, Epyc 64 Titan Xp's or 32 Titan V's.

Thats Nuts, I think even data centre / deep learning guys would even be like, "THAT'S TOO MUCH BANDWIDTH, WHAT AM I SUPPOSED TO DO WITH THIS?"

AMD is going to release a set of Zen 2-based APUs on AM4 in 2020. It's why they use the words "support" for AM4 through 2020. DDR5 will require a new socket, so we know that much. But AMD is always hedging with those statements. Realize that DDR5 is supposed to launch in 2020, but it could be late. There could end up being a Zen2+ model in 2020 if DDR5 runs horribly late. Those decisions will be finalized in early to mid-2019. (Intel has this exact same issue as well.)

 

PCIe 4.0 should be in Zen2. Whether any consumer space Motherboards use it is another issue. PCIe 5.0 is a wholly different issue. The design needs to be finalized before they can lock in the CPU design, which means it's at least 18 months from that point until CPUs are using it. (Non-CPU parts get there earlier, as there's also certain PCIe 4.0 products in the server space.)

[cj09beira]

6 minutes ago, Taf the Ghost said:

AMD is going to release a set of Zen 2-based APUs on AM4 in 2020. It's why they use the words "support" for AM4 through 2020. DDR5 will require a new socket, so we know that much. But AMD is always hedging with those statements. Realize that DDR5 is supposed to launch in 2020, but it could be late. There could end up being a Zen2+ model in 2020 if DDR5 runs horribly late. Those decisions will be finalized in early to mid-2019. (Intel has this exact same issue as well.)

 

PCIe 4.0 should be in Zen2. Whether any consumer space Motherboards use it is another issue. PCIe 5.0 is a wholly different issue. The design needs to be finalized before they can lock in the CPU design, which means it's at least 18 months from that point until CPUs are using it. (Non-CPU parts get there earlier, as there's also certain PCIe 4.0 products in the server space.)

i am guessing am4+ brings pcie 4 

[qepsilonp]

4 minutes ago, Taf the Ghost said:

AMD is going to release a set of Zen 2-based APUs on AM4 in 2020. It's why they use the words "support" for AM4 through 2020. DDR5 will require a new socket, so we know that much. But AMD is always hedging with those statements. Realize that DDR5 is supposed to launch in 2020, but it could be late. There could end up being a Zen2+ model in 2020 if DDR5 runs horribly late. Those decisions will be finalized in early to mid-2019. (Intel has this exact same issue as well.)

 

PCIe 4.0 should be in Zen2. Whether any consumer space Motherboards use it is another issue. PCIe 5.0 is a wholly different issue. The design needs to be finalized before they can lock in the CPU design, which means it's at least 18 months from that point until CPUs are using it. (Non-CPU parts get there earlier, as there's also certain PCIe 4.0 products in the server space.)

Errr you think they are going to implement PCIe 4 on a AM4 board which would mean that they would need to support both PCIe 3 and PCIe 4 on the CPU, or did you mean Zen 3 and considering Zen 3 is probably not going to come out early in 2020 depending on when PCIe 5 is finalised in the year say it's finalised in Jan 2019 if AMD went for a late 2020 launch which they probably will anyway then that leave lots of time, and according to other sources it's 12 months, although that was Tech quickie and LTT isn't where you should get your tech news on anything more than a superficial level.  

[cj09beira]

4 minutes ago, qepsilonp said:

Errr you think they are going to implement PCIe 4 on a AM4 board which would mean that they would need to support both PCIe 3 and PCIe 4 on the CPU, or did you mean Zen 3 and considering Zen 3 is probably not going to come out early in 2020 depending on when PCIe 5 is finalised in the year say it's finalised in Jan 2019 if AMD went for a late 2020 launch which they probably will anyway then that leave lots of time, and according to other sources it's 12 months, although that was Tech quickie and LTT isn't where you should get your tech news on anything more than a superficial level.  

as far as i know pcie versions mostly only differ in pin frequency, and thats how they keep backwards compatibility, older versions simply respond a few cycles latter, so they can have a cpu supporting both no problem, all cpus already do that, and yes he is talking about zen 3, with probably an updated socket (possibly a am4+)

[Taf the Ghost]

6 minutes ago, cj09beira said:

i am guessing am4+ brings pcie 4 

Possible, though I'm not sure they'll even call it AM4+. It'll just be the 500-series Chipsets. At the same time, it's not assured they'll run the consumer ones at PCIe 4.0 speeds until Zen 3. There's literally no use in normal Desktop space in the current roadmaps for products, but that's not true in the server space. Which is why I would expect it on Epyc 8000 series parts. 

2 minutes ago, qepsilonp said:

Errr you think they are going to implement PCIe 4 on a AM4 board which would mean that they would need to support both PCIe 3 and PCIe 4 on the CPU, or did you mean Zen 3 and considering Zen 3 is probably not going to come out early in 2020 depending on when PCIe 5 is finalised in the year say it's finalised in Jan 2019 if AMD went for a late 2020 launch which they probably will anyway then that leave lots of time, and according to other sources it's 12 months, although that was Tech quickie and LTT isn't where you should get your tech news on anything more than a superficial level.  

PCIe 4.0 is fully backwards compatible with PCIe 3.0. I expect them to simply run the lanes at PCIe 3.0 speeds by some microcode approach. Turns out X370 chipsets are running at PCIe 2.0 x4 off the CPU. I'm not sure of the technical hurdles to this, mostly as microchip control code isn't something I've studied deeply, but I'm fairly certain we've seen this before. As PCIe 4.0 has no benefit in the consumer space for a rather long time.

 

However, if AMD is splitting the AM4 and Server dies in Zen2, which I also expect, they could just not bother with PCIe 4.0 at all on the Consumer die.

[Lotus]

I'm hoping for something akin to the 3.9 Ghz i7-4770k to 4.4 GHz 4790k. A 500mhz increase in boost clock would be great. That was a refresh too, and Ryzen was probably released a bit on the edge of premature too.

[qepsilonp]

4 minutes ago, Taf the Ghost said:

Possible, though I'm not sure they'll even call it AM4+. It'll just be the 500-series Chipsets. At the same time, it's not assured they'll run the consumer ones at PCIe 4.0 speeds until Zen 3. There's literally no use in normal Desktop space in the current roadmaps for products, but that's not true in the server space. Which is why I would expect it on Epyc 8000 series parts. 

PCIe 4.0 is fully backwards compatible with PCIe 3.0. I expect them to simply run the lanes at PCIe 3.0 speeds by some microcode approach. Turns out X370 chipsets are running at PCIe 2.0 x4 off the CPU. I'm not sure of the technical hurdles to this, mostly as microchip control code isn't something I've studied deeply, but I'm fairly certain we've seen this before. As PCIe 4.0 has no benefit in the consumer space for a rather long time.

 

However, if AMD is splitting the AM4 and Server dies in Zen2, which I also expect, they could just not bother with PCIe 4.0 at all on the Consumer die.

Ehhh I dunno PCIe Gen 3 x4 is already becoming a limitation for NVMe although I must admit a doubling would ease that for a while which can be done with PCIe 4 and the Titan V has already run into limitations with PCIe gen 3 8x assume the next high end GPU adds another 70% performance and the one after that then you are going to be limited by PCIe gen 3 16x

 

PCie gen 4 is definitely something that consumers will need soon after Zen 2 like another year maybe.

[Taf the Ghost]

5 minutes ago, cj09beira said:

as far as i know pcie versions mostly only differ in pin frequency, and thats how they keep backwards compatibility, older versions simply respond a few cycles latter, so they can have a cpu supporting both no problem, all cpus already do that, and yes he is talking about zen 3, with probably an updated socket (possibly a am4+)

Zen3 will be AM5. The DDR5 will necessitate a fairly different socket. DDR5 is a big upgrade, compared to the relatively smaller clock boost with DDR4. (This is why DDR5 being "on time" in 2020 is going to be the biggest story from Computex 2019 through 2020 in the Computer space. If it isn't, nearly the entire industry is going to be scrambling to adapt.) http://www.pcgamer.com/ddr5-memory-is-twice-as-fast-as-ddr4-and-slated-for-2019/  The first working prototypes only were announced recently.

 

Or Zen3 ends up being still on DDR4 if DDR5 is really late. Or AMD launches a Zen2+ CPU in early 2020 and Zen3 late in 2020. (It's not like Intel hasn't just done that.)

[Taf the Ghost]

2 minutes ago, qepsilonp said:

Ehhh I dunno PCIe Gen 3 x4 is already becoming a limitation for NVMe although I must admit a doubling would ease that for a while which can be done with PCIe 4 and the Titan V has already run into limitations with PCIe gen 3 8x assume the next high end GPU adds another 70% performance and the one after that then you are going to be limited by PCIe gen 3 16x

 

PCie gen 4 is definitely something that consumers will need soon after Zen 2 like another year maybe.

PCIe 4.0 is useful in the Workstation/HEDT and server space, most definitely. Which is why I fully expect AMD to have it operating in Zen2. (Notably, they finished the spec and then, soon after, AMD was talking about "taped out" chips.)  At the rate that GPUs are increasing for performance, it's going to be another couple of generations before a PCIe 3.0 x16 slot is bandwidth limited by the socket. That just isn't a worry in the consumer space.

 

The one thing we don't know is how many dies is AMD producing on 7nm in the first go with Zen2? My assumption is there is at least 2 designs. Maybe not "mainstream + server" but "little + big". Then it'll be up to the motherboard manufacturers about what they bring to Mainstream. I expect PCIe 4.0 on Threadripper 3 and Epyc 8000 series, however. (Easy prediction: those two parts will have a new Socket.) 

[qepsilonp]

19 minutes ago, Taf the Ghost said:

PCIe 4.0 is useful in the Workstation/HEDT and server space, most definitely. Which is why I fully expect AMD to have it operating in Zen2. (Notably, they finished the spec and then, soon after, AMD was talking about "taped out" chips.)  At the rate that GPUs are increasing for performance, it's going to be another couple of generations before a PCIe 3.0 x16 slot is bandwidth limited by the socket. That just isn't a worry in the consumer space.

 

The one thing we don't know is how many dies is AMD producing on 7nm in the first go with Zen2? My assumption is there is at least 2 designs. Maybe not "mainstream + server" but "little + big". Then it'll be up to the motherboard manufacturers about what they bring to Mainstream. I expect PCIe 4.0 on Threadripper 3 and Epyc 8000 series, however. (Easy prediction: those two parts will have a new Socket.) 

Well if AMD want PCIe 4 for Zen 2 or Zen 3 on the workstation and server they might as well do it on desktop as well, well as long as they go the same route of Ryzen 1 of using the same die for the consumer space as the workstation and server, the only problem then is designing the boards which isn't terribly difficult for a PCIe change. I suppose you could ague the reason they when for 64 and 128 PCIe lanes on TR and Epyc was because they were not going to go for PCIe gen 4, although PCIe gen four was finalised PCIe gen 4 in Oct 2018 and AMD only recently announced the design was finished so Zen 2 with PCIe 4 is not all that unlikely.

 

And there is a reason to use the same die for HEDT and Data Centres because you can use the best dies for server and workstations and don't need to throw away once that don't meet the spec, you can just use them in the consumer space and having an upgrade path in data centres is even more important than it is in the consumer space, as you average consumer will sit on a CPU for like 3 - 5 years, by which time you would have already changed sockets, where as data centres usually upgrade much more regularly and completely replacing all there hardware is a lot more difficult and expensive.

 

Why would they keep the socket for consumers and change the socket for data centres?

[Inversion]

17 hours ago, qepsilonp said:

Well then you are going to be let down, if you already have a Ryzen processor with the right number of cores for you Ryzen was always not going to be a worth while upgrade, wait for Zen 2 / Ryzen 3 like I am if you already have a Ryzen CPU or have a CPU which is good enough for purpose, Zen+ was made so that AMD could stay semi competitive between the long wait between Zen and Zen 2 as a stop gap, as 12LP is really only an improved 14LPP process and that's the only "big" difference, there are the tighter cache timings which should improve AMD's performance in games at extremely high frame rates.

Which will make AMD look better when the tech press run 720p benchmarks with a Titan XP, although it will also help in more reasonable situations also like if you want to run at 144Hz which Ryzen does have a little bit of touble getting to those kinds of frame rates due to the cache latency, memory latency and fabric latency. Although the last two memory latency and fabric latency are the same thing, because the IMC is connected to the rest of the CPU via the infinity fabric, but AMD also promised better RAM support so if you can for example run 3600Mhz memory you improve the latency of the fabric dramatically.

 

But I hope it is something that is properly address in Ryzen 3 / Zen 2 because in specific titles it's even a problem at what are reasonable frame rates, 90 to 120fps because the games are more latency sensitive, this is the optimisation issue you hear in reference to Zen, it's games over using the L3 cache rather than using the L2 cache

 

Although it is a weakness in the Zen architecture, so it's not so much a optimisation issue as just something that Intel is better at, but at the same time it isn't necessary to be so dependent on the L3 cache, it's the same issue that Skylake X had, but as Intel move to higher and higher core counts for the mainstream they wont be able to rely on the old ring bus anymore, so I think game developers need to start moving the dependence of there game engines to L2 cache rather than L3, it is something you need to do anyway to get higher core scaling which means it's probably already being done.

 

This is also the reason AMD holds up much better in content creation because they have a 512KB L2 cache vs Intel's mainstream parts which have 256KB of L2 cache, which as said in applications that scale well with many cores it is necessary to rely on the L2 cache and because AMD also have a low latency link between all the L2 caches on a CCX when you do for example a Cinebench run on all cores a 1800X will beat a 7700k by almost exactly double even though on a single threaded test the 1800X will lose by about 15%.

My main issue with what you're saying here is that Ryzen was never gonna be a worthwhile upgrade. If you were coming from an older system (such as myself from an Athlon X4 860K) it was a huge upgrade. And if you were coming from an i3 or i5 system and were wanting to content create on a budget then again Ryzen was a godsend. 6C/12T for about £180 normally or £140 on sale was a bargain and still is. I think you need to get that AMD really hit it out of the park on Ryzen from both a value point of view and when compared to their older hardware. It was a massive leap forwards and let's be real. If Ryzen and Threadripper hadn't been as big a game changer, Intel would most likely only now be releasing X299 about now and Coffee Lake would still be a few months away. It was the kick in the pants the industry needed and I can't wait to see exactly what AMD do next.

[JediFragger]

46 minutes ago, Inversion said:

6C/12T for about £180 normally or £140 on sale was a bargain and still is.

My only question is, if Ryzen is such a great seller then why are they having to have a 'sale' on them? Pricing has been rather 'flexible' for the last 6 months or so though from what I've seen here in the UK.

[MaktimS]

51 minutes ago, JediFragger said:

My only question is, if Ryzen is such a great seller then why are they having to have a 'sale' on them? Pricing has been rather 'flexible' for the last 6 months or so though from what I've seen here in the UK.

Because new CPUs are coming. Current Ryzen has excellent selling numbers. For example in Germany, where AMD is selling more CPUs than Intel