Jump to content
Sign in to follow this  
Keyrock

AMD Ryzen

Recommended Posts

 

4 cores/ ccx makes most sense since it's simplest and has fewest intra ccx connections and 8 would have a lot, but unless they planned for it back when they designed AM4 I'd question if they could go to a full chiplet design- with 2 4 core ccx- while staying with AM4. If they did plan for that I would be very impressed, and AMD did put a lot of stuff into the CPU package which would usually go into the chipset on the motherboard. That would also give an easy way to get 12 cores, threadripper had dummy ccx in gen 1 so you could get 12 core with 3x4 and one dummy.

 

I guess if they went the 8 core/ ccx route they'd have more failed chips to pad out the lower grade SKUs as well as 7nm probably being a bit less reliable as well; my skepticism comes from the extremely low relative failure rate for Zen1. We got a lot of 8 core 1600s here due to them running out of 'bad'/ partially failed chips.

Well, Intel is already offering 8-core CPUs that don't employ the CCX concept. It is more complex so probably more expensive to implement, but I don't think it would affect yields much. Plus the 7nm chiplets are approximately 1/3 the area of the 14nm chips, so that will offset much of the loss in yields of an immature 7nm process.

 

That said Voldemort has retracted his claim on the IO chip; Now he is claiming that these Ryzen 3 products use 7nm chiplets only. More likely that they have slightly larger 7nm chips with IO integrated for PCs, at maybe 1/2 the area of 14nm chips each. This also makes more sense with the combination with Navi which I think would have IO integrated because of bandwidth and energy requirements. Infinity Fabric 2 will do what, 100 GB/s between chips? That may be enough for an integrated GPU, but not for any discrete GPU nowadays.

 

 

The Intel 8 core is very expensive though, in part because you need to have 1x8 'perfect' cores when using ringbus rather than 2 perfect lots of 4 as with the infinity fabric/ ccx system; and it's also expensive on a very mature and refined node. Assuming linear error rates 8 core ccx would double the number of 'bad' ccxes (who knows though, depending on how the intra ccx stuff is handled complexity may go up non linearly and some stuff will have the same error rate whatever the core count). That might remain within acceptable levels, but it's all speculation at the moment.

 

At this point I'm not really sure what to make of the I/O situation at all. Too much rumour and I don't have the technical expertise to evaluate the relative likelihoods.

Share this post


Link to post
Share on other sites

Yeah, the placement says that there will be 2 chiplet designs coming, no reason for that positioning unless they are. I presume they're holding off on the announcement so as not to cannibalise Threadripper 2nd gen sales (albeit TR does have other advantages like quad channel memory) and since the release is still a decent amount of time away.

Share this post


Link to post
Share on other sites

Yeah, the placement says that there will be 2 chiplet designs coming, no reason for that positioning unless they are.

 

Yeah, and looking at a video of the package being held up there is obvious tracing for another chiplet below the top one.

 

 

I presume they're holding off on the announcement so as not to cannibalise Threadripper 2nd gen sales (albeit TR does have other advantages like quad channel memory) and since the release is still a decent amount of time away.

 

 

Makes sense. Even with the quad-channel memory advantage to the Threadrippers, the 16 core Ryzen 3000 will probably be able to beat it in serial or memory insensitive workloads which will be a lot of workloads. That difference between the non-Threadripper 2700X and the test chip in Cinebench is substantial by itself (~1750 points to ~2050 points).

Share this post


Link to post
Share on other sites

16 cores given the power draw of the 8 core suggests they are likely to need an upgraded socket, so I wouldn't be at all surprised to see AM4+ (presumably backcompatible) and maybe x590 chipset as well. Shame there's no frequency information on the 8 core they showed so it could be compared to the 9900k's. Still, clearly a big improvement on the 2000 series even as an Engineering Sample.

  • Like 1

Share this post


Link to post
Share on other sites

So more news from Anandtech: there will be no GPU chiplets for the Ryzen 3000 series. All the APUs will use the existing APU design but on 12nm instead of 14nm. There will be an APU based on Zen 2 cores but at a later date, so probably the Ryzen 4000 series or whatever comes next. AMD also claims that the TDPs will be the same as for the 1000 and 2000 series. I'm not sure that actually means anything. Couldn't they just release SKus with the same TDP at the same prices as their current lineup but with higher TDP CPUs at higher prices? I think I will be surprised if the 16-core SKUs are supported on my B350 mobo.

Share this post


Link to post
Share on other sites

Yeah, I've got a x370 MB and I'm not expecting 16 cores to even be supported on that. 16 cores at the same freqs as the 8 core demo gets the power draw into Bulldozer territory, and it would near have to have twice the TDP of the 1800x 300 series boards were designed for. Pulling that amount of power through the socket itself is very likely feasible, but I'd doubt other components could handle it.

 

APUs lagging by a generation is kind of understandable given that margins are low and 7nm will be fairly expensive (plus there's the GloFo supply agreement), plus there's a lot of inertia in the laptop market. I really wish AMD would fix their naming conventions though as it's as inconsistent as you could get. Ryzen 1x00 is Zen1, Ryzen 2x00 is Zen+, Ryzen 2400g is Zen1, Ryzen 3x00 will be Zen2 and Ryzen 3000 APUs will be Zen+; Vega 10 and 20 are generations, Vega VII is Vega 20, Vega64 is Vega 10, Vega 3/4/8/11 is Vega 10, Vega24/32 on Intel chips is actually Polaris with some Vega 10 features...

  • Like 1

Share this post


Link to post
Share on other sites

Lisa Su announced to be giving the keynote address at Computex towards the end of May. Very likely to be a release date announcement for Zen 2/ Ryzen 3000 series at that time*, some chance for them to be launched at the event. Looks like +10% IPC improvement overall (far more on some workloads) so even a moderate frequency uptick should see Intel's performance advantage gone.

 

*Also the Epyc Rome processors and info on Navi

Share this post


Link to post
Share on other sites

Computex address is just wrapped up. Main Ryzen points:

15% IPC boost which is at the top end of realistic expectations

3700x/3800x 8/16 core/ thread SKUs at 65/105W 4.4/ 4.5 Ghz boost (higher TDP unit has higher base clock). 329 and 399USD RRPs

3900x 12/24 R9 SKU at 105W with 4.6 GHz and USD499 RRP.

Probably not a great watch for Intel, that IPC boost should put Zen2 well ahead clock to clock and the 3900x is half their equivalent's price and TDP. If there's no voltage wall this time around a 5GHz overclock may well be feasible for AMD as well. There was some x570 news as well, but I wasn't really paying much attention to that (or Navi, though I did catch that it's not GCN again as was rumoured).

  • Like 2

Share this post


Link to post
Share on other sites

Time to buy some AMD stock.  Well, or maybe Intel in a bit if all the doom and gloom is right.


Why has elegance found so little following? Elegance has the disadvantage that hard work is needed to achieve it and a good education to appreciate it. - Edsger Wybe Dijkstra

Share this post


Link to post
Share on other sites
27 minutes ago, injurai said:

Can we talk about those cache gains? Oomph!

That's the thing that stood out to me the most about the R9, not the extra 4 cores over the R7s, but the utterly insane 70 MB of cache. I mean, even the lower models have plenty of cache, but that 70 MB... holy crap!

For comparison, Intel's top of the line i9-9980XE, which costs $1800, has 24.75 MB of cache.

Share this post


Link to post
Share on other sites

This new frontier of cache is going to really facilitate heavy parallel workloads. Also I'm liking the SIMD gains as well with Zen 2.

Share this post


Link to post
Share on other sites
2 hours ago, Keyrock said:

That's the thing that stood out to me the most about the R9, not the extra 4 cores over the R7s, but the utterly insane 70 MB of cache. I mean, even the lower models have plenty of cache, but that 70 MB... holy crap!

For comparison, Intel's top of the line i9-9980XE, which costs $1800, has 24.75 MB of cache.

...Is it actually industry standard to work with powers of 10 instead of powers of 2 there?


How many fingers am I holding up, Winston?
 
How to Totally Remove Ignored Users from Your Obsidian Forums.

Share this post


Link to post
Share on other sites

The 70MB is (IIRC) 32MB per chiplet (Level 3) plus 6MB shared (Level 2)*. I presume the big cache is to help with the latency that has been a problem with Zen/+ and perhaps help with the high core count chips being a bit limited by things like dual channel memory. AMD's full computex announcement summary is here, for those interested in a bit more detail/ Epyc and the R5 chips that didn't get a formal announcement in the presentation.

*Not sure how Intel gets 24.75MB but it will be in powers of two as well- somehow.

 

  • Like 2

Share this post


Link to post
Share on other sites

L3 becomes a victim only cache. Meaning to get on an L3, you have to have been cached in the L2 of a particular chip. This is great for heavily disjoint parallel tasks that don't need to synchronize memory access. That's a lot of "personal space." Synchronized workloads will be best dispatched to one chi or the other, not across both. At least if you want to be efficient about your over-all throughput as opposed to execution time.

  • Hmmm 1

Share this post


Link to post
Share on other sites

Surprising few, the 16 core R9 3950X has been announced at E3: 3.5/4.7 base/ boost, 72MB cache, 105W TDP (lol), USD749.

Not all that tempting at that price (I'll either get an 8 core, or go for a MB replacement and 12 core; around Christmas), and with those core numbers I suspect most users would be better off going threadripper for quad channel and more PCIe lanes. Be good for the epeen though. Quite a few benchmarks shown for the other 3000 series entries- plus specs etc for the APUs, 3400G has been beefed up a fair bit over the somewhat anaemic 2400G- which suggest they do out perform Intel chips in games now even in traditionally Intel favoured games like CSGO. Pinch of salt of course, given it was an AMD event.

Share this post


Link to post
Share on other sites

I guess the biggest positive of the R9 3950X, as compared to Threadripper, is that it doesn't require a ludicrously expensive motherboard. Either way, 16 cores is just overkill at this time. Unless you are constantly doing heavy encoding, or something similar where the extra cores will actually make a difference, buying a16 core CPU is simply showing off your **** size.

I'm probably going to wait a while before doing a new desktop build, hopefully the refresh of Ryzen 3 is out by then, and maybe we'll even get a Radeon GPU that's mildly exciting by then so that I could do a fully Team Red build, though I'm not holding my breath on the latter part after seeing the Navi cards they just announced.

Edited by Keyrock

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...
Sign in to follow this  

×
×
  • Create New...