|
mayodreams posted:I am actually getting a little excited now!  https://www.youtube.com/watch?v=xzZT2xH3zBk It's a comparison @ 3.0Ghz so already 2.8Ghz baseclock sounds suspect.
|
#
?
Aug 18, 2016 15:12
|
|
|
|
| # ? May 15, 2024 18:45 |
|
|
HalloKitty posted:The problem now is that it's not scheduled to arrive in any quantity until 2017, and there's always time for the chipset to suck, or the pricing to be unrealistic. The best thing for AMD in the long run if they have a big Xeon competitor on their hands could be to almost ignore the consumer market, only push out a $600+ Athlon 64 FX successor, and use as many dies as they possibly can to start undercutting $3k Xeons by 20%. Server market is where the growth is, and if AMD wants to start making investors happy again they need to find a way to get into Amazon / Google / Facebook / Microsoft / Oracle datacenters.
|
|
#
?
Aug 18, 2016 15:14
|
|
|
Twerk from Home posted:The best thing for AMD in the long run if they have a big Xeon competitor on their hands could be to almost ignore the consumer market, only push out a $600+ Athlon 64 FX successor, and use as many dies as they possibly can to start undercutting $3k Xeons by 20%. Server market is where the growth is, and if AMD wants to start making investors happy again they need to find a way to get into Amazon / Google / Facebook / Microsoft / Oracle datacenters. If that is their target, how large of a discount are they going to need over a similarly performing Xeon to match a Xeon's power efficiency advantage considering the high TDP of the chip? Efficiency is a huge driver even in the server world these days.
|
|
#
?
Aug 18, 2016 15:31
|
|
|
Twerk from Home posted:The best thing for AMD in the long run if they have a big Xeon competitor on their hands could be to almost ignore the consumer market, only push out a $600+ Athlon 64 FX successor, and use as many dies as they possibly can to start undercutting $3k Xeons by 20%. Server market is where the growth is, and if AMD wants to start making investors happy again they need to find a way to get into Amazon / Google / Facebook / Microsoft / Oracle datacenters. It's a pretty good idea, but we still have yet to see the power consumption figures. That could toss a spanner in the works. If I can buy what is basically a Broadwell-E from AMD for less money, I won't care if it consumes a couple dozen more watts, I'll be all over that poo poo; a datacentre using a shitton of processors 24/7, on the other hand, probably will. HalloKitty fucked around with this message at 15:48 on Aug 18, 2016 |
|
#
?
Aug 18, 2016 15:31
|
|
|
Chuu posted:If that is their target, how large of a discount are they going to need over a similarly performing Xeon to match a Xeon's power efficiency advantage considering the high TDP of the chip? Efficiency is a huge driver even in the server world these days. 95W for 8C/16T, 150W for 24C/48T and 180W for 32C/64T. I'm not sure why there isn't a 16C/32T chip for what seems to be a 125W slot but I'm not AMD. Basically they seem to trade blows with Broadwell at each turn, so I want to say this will all be a pricing issue. What's the die size on a Broadwell Xeon? Zen is supposed to be 143mm˛.
|
|
#
?
Aug 18, 2016 15:46
|
|
|
drat if they can pull this off, it would be fantastic to have an AMD/Intel Chip battle once again. If anything to start seeing some variety in builds that aren't all only Intel with the random mom and pop PC being AMD. Just so long as they keep their support between the two the same. (NMS and the SSE for isntance. :/ )
|
|
#
?
Aug 18, 2016 16:58
|
|
|
Still salting heavily, I want to see proper benchmarks before we can say AMD is out of the woods. That it is being compared favorably to Broadwell, as opposed to Haswell (like I was saying AMD should be targeting instead) is an encouraging sign, but I really don't like that artificial 3 GHz number. I also still don't like the NUMA L3, and I still want to hear what their competing solution to Intel's ring buses is.
|
|
#
?
Aug 18, 2016 21:46
|
|
|
It's darkly funny once you realize there are a not-insignificant amount of areas fixed-clock Xeons on Broadwell falter compared to their Haswell counterparts I guess the bar being lowered makes internal goals that much easier 
|
|
#
?
Aug 18, 2016 21:48
|
|
|
SwissArmyDruid posted:Still salting heavily, I want to see proper benchmarks before we can say AMD is out of the woods. That it is being compared favorably to Broadwell, as opposed to Haswell (like I was saying AMD should be targeting instead) is an encouraging sign, but I really don't like that artificial 3 GHz number.
|
|
#
?
Aug 18, 2016 21:56
|
|
|
In the end, I guess this is an upside down and backwards version of wanting AMD to shoot for the stars. Too often for my liking, AMD tries some really ambitious stuff. Most of which sounds good on paper, but the supporting tech for which just doesn't exist yet. APUs might finally be what they should have been from the get-go, once they get some HBM on them to feed the GPU cores properly, for example. Then everyone else just comes in and swoops in and copies them. I mean, most recently, Intel just acquired a company called Nervana, whose has a product that has a striking resemblance to Fiji XT, right down to the interposer and stacked DRAM. This time, I kind of would have liked them to aim for Cannon Lake performance, and fall a bit short around Broadwell, instead of aiming for Haswell like everyone else has been saying and running into problems like with the the RX 480s needing to have their voltage cranked to hit the clocks AMD needs them to hit, thus coming out with yet another deficient part. I dunno. I just want competition in the CPU and GPU markets again and AMD is moving at a glacial pace, and there's nothing that can be done externally to get them there.
|
|
#
?
Aug 18, 2016 21:56
|
|
|
Nothing a good dose of corporate espionage couldn't fix.
|
|
#
?
Aug 18, 2016 23:35
|
|
|
I'm not sure they even need to spy. Intel puts all their ideas there on the table, you could do a lot worse than copying the number of ALUs and AGUs, or cache sizes, or whatever. Also, I'm hearing through the grapevine that, much to my dismay, Haswell was the target and AMD hit the jackpot.
|
|
#
?
Aug 19, 2016 00:22
|
|
|
SwissArmyDruid posted:I'm not sure they even need to spy. Intel puts all their ideas there on the table, you could do a lot worse than copying the number of ALUs and AGUs, or cache sizes, or whatever. We were always saying though that if they'd hit Haswell-level performance, that'd be great, so why the frustration?
|
|
#
?
Aug 19, 2016 00:29
|
|
|
Going for Haswell IPC and undercutting in price vs the concurrent intel SKU by around 35-45% is the most realistic and likely outcome considering the position AMD is in. As much as we all would like AMD to pull out something out of their rear end ala POWER9 and give Intel some real trouble, it ain't happening. Anime Schoolgirl fucked around with this message at 00:33 on Aug 19, 2016 |
|
#
?
Aug 19, 2016 00:31
|
|
|
Said it before I could post. Haswell is a great place to be, and progress is slowing down generation by generation. Now for AMD to price this realistically - AND beyond where Intel would simply price drop old hardware and kill Zen in the cradle.
|
|
#
?
Aug 19, 2016 00:48
|
|
|
Anime Schoolgirl posted:Going for Haswell IPC and undercutting in price vs the concurrent intel SKU by around 35-45% is the most realistic and likely outcome considering the position AMD is in. this is all anyone cares abt. haswell to sky lake is a minimal gain and if zen is as good as broadwell, well, its gonna be good also intel doesn't price drop old HW since no one would buy new HW.
|
|
#
?
Aug 19, 2016 00:56
|
|
|
Who is going to actually build Zen chips for AMD? If it is Global Foundries again, I will be very surprised if the chips attain clockspeeds that allow them to compete. GF has done a terrible job producing 14nm graphics chips for AMD. Their 14nm process tech comes from Samsung, who I would imagine optimized for low power ARM chips, DRAM, and NAND chips. If TSMC is building Zen chips for AMD, there is a lot more reason for optimism, IMO.
|
|
#
?
Aug 19, 2016 01:24
|
|
|
SwissArmyDruid posted:I'm not sure they even need to spy. Intel puts all their ideas there on the table, you could do a lot worse than copying the number of ALUs and AGUs, or cache sizes, or whatever. I thought the info out now is just that it can kinda hit Haswell under very tightly designed benchmarks, and probably doesn't manage that performance in more typical workloads?
|
|
#
?
Aug 19, 2016 01:27
|
|
Well the actual cost per die will probably be around $75 assuming yield loss with glofo isn't reimbursed and yields are where other 14nm products are... Those are big chips! With packaging costs, I expect the material cost of the flagship chip will be above $100, so with R&D, marketing, etc I wouldn't be that surprised to see a $400 price point for that, especially to make up for what is likely to be a rather large share of non bin 1 good dies.PBCrunch posted:Who is going to actually build Zen chips for AMD? If it is Global Foundries again, I will be very surprised if the chips attain clockspeeds that allow them to compete. GF has done a terrible job producing 14nm graphics chips for AMD. Their 14nm process tech comes from Samsung, who I would imagine optimized for low power ARM chips, DRAM, and NAND chips. Samsung's14nm process is mainly used to manufacture SOCs, yes, but that's not thing only thing they make at 14nm. There are many different 14nm processes that are suited for different applications as well, like lpp, lpe, etc. And remember, TSMC makes SoCs too! Watermelon Daiquiri fucked around with this message at 01:34 on Aug 19, 2016 |
|
|
#
?
Aug 19, 2016 01:29
|
|
|
PBCrunch posted:Who is going to actually build Zen chips for AMD? If it is Global Foundries again, I will be very surprised if the chips attain clockspeeds that allow them to compete. GF has done a terrible job producing 14nm graphics chips for AMD. Their 14nm process tech comes from Samsung, who I would imagine optimized for low power ARM chips, DRAM, and NAND chips. It's not confirmed that Global Foundries is making Zen but this quote makes it seem that way. Mark Papermaster AMD posted:...14LPP process technology across a broad set of our CPU, APU, and GPU products. http://www.globalfoundries.com/news...on-amd-products
|
|
#
?
Aug 19, 2016 01:33
|
|
Tanreall posted:It's not confirmed that Global Foundries is making Zen but this quote makes it seem that way. yeah, 14nm is either Glofo, Intel, or Samsung iirc. I highly doubt intel will be a foundry for them, and I really don't think its samsung.
|
|
|
#
?
Aug 19, 2016 01:36
|
|
|
Watermelon Daiquiri posted:Well the actual cost per die will probably be around $75 assuming yield loss with glofo isn't reimbursed and yields are where other 14nm products are... Those are big chips! With packaging costs, I expect the material cost of the flagship chip will be above $100, so with R&D, marketing, etc I wouldn't be that surprised to see a $400 price point for that, especially to make up for what is likely to be a rather large share of non bin 1 good dies.
|
|
#
?
Aug 19, 2016 01:53
|
|
|
What makes a process suitable for high/low power stuff anyways? Fin dimensions? # fins per transistor? materials? sige growth?
|
|
|
#
?
Aug 19, 2016 01:55
|
|
|
Watermelon Daiquiri posted:yeah, 14nm is either Glofo, Intel, or Samsung iirc. I highly doubt intel will be a foundry for them, and I really don't think its samsung. quote:AMD has strong foundry partnerships and our primary manufacturing partners are GLOBALFOUNDRIES and TSMC. We have run some product at Samsung and we have the option of enabling production with Samsung if needed as part of the strategic collaboration agreement they have with GLOBALFOUNDRIES to deliver 14nm FinFET process technology capacity. I don't think there's any difference between Global Foundries and Samsung other than production output.
|
|
#
?
Aug 19, 2016 01:57
|
|
|
Yeah, that depends on what all the agreement between samsung and glofo is; whether or not it includes new processes, yield improvements, etc or if it was just samsung going "Ok, here is what we have so far, have fun!" and glofo engineers have been futzing around with it since.
|
|
|
#
?
Aug 19, 2016 02:24
|
|
|
Anime Schoolgirl posted:Going for Haswell IPC and undercutting in price vs the concurrent intel SKU by around 35-45% is the most realistic and likely outcome considering the position AMD is in. Gotta love when Zen is rumored to be competitive with BDW-E, the minds of fanboys go straight to "gimme that for <$200", because competitive performance = competitive pricing doesn't apply to AMD. With a healthy fanbase like this small wonder why AMD has been bleeding money since forever.
|
|
#
?
Aug 19, 2016 03:30
|
|
|
1100 x .65 = 715 It's a bit naive to think that competitive performance = competitive price. You can see it in most markets where brand names demand a premium price.
|
|
#
?
Aug 19, 2016 05:14
|
|
|
Given that the best thing I can buy for $700 right now is either E5-2360 v3 8C16T @ 2.4GHz, an E5-1650 v4 6C12T @ 3.6GHz, or an i7-6850K 6C12T @ 3.8GHz? I'd say that Monsieur AMD has met the minimum buy in at the poker table. Might not be the high-roller table, but there are a couple of Chinese fat cats in the same room throwing around a couple thousand playing pai gow.
SwissArmyDruid fucked around with this message at 07:14 on Aug 19, 2016 |
|
#
?
Aug 19, 2016 06:13
|
|
|
AnandTech have a thing about AMD today. I didn't see it posted. Some standout quotes: Along with this diagram:  quote:The INT side of the core will funnel the ALU operations as well as the AGU/load and store ops. The load/store units can perform 2 16-Byte loads and one 16-Byte store per cycle, making use of the 32 KB 8-way set associative write-back L1 Data cache. AMD has explicitly made this a write back cache rather than the write through cache we saw in Bulldozer that was a source of a lot of idle time in particular code paths. AMD is also stating that the load/stores will have lower latency within the caches, but has not explained to what extent they have improved. quote:However, we have recieved information from a source (which can't be confirmed via public AMD documents) that states that Zen will feature two sets of 8MB L3 cache between two groups of four cores each, giving 16 MB of L3 total. This would means 2 MB/core, but it also implies that there is no last-level unified cache in silicon across all cores, which Intel has. quote:What this means, between the L2 and the L3, is that AMD is putting more lower level cache nearer the core than Intel, and as it is low level it becomes separate to each core which can potentially improve single thread performance. The downside of bigger and lower (but separate) caches is how each of the cores will perform snoop in each other’s large caches to ensure clean data is being passed around and that old data in L3 is not out-of-date. AMD’s big headline number overall is that Zen will offer up to 5x cache bandwidth to a core over previous designs. quote:For Zen, the micro-op cache will save power by not having to go further out to get instruction data, improved prefetch and a couple of other features such as move elimination will also reduce the work, but AMD also states that cores will be aggressively clock gated to improve efficiency. Interesting article and pretty well balanced. Definitely has given me a little more tempered optimism given the stuff above about efficiency. E: while I'm here, motherboard stuff for Zen server and consumer stuff: http://www.anandtech.com/show/10581/early-amd-zen-server-cpu-and-motherboard-details-codename-naples-32cores-dual-socket-platforms-q2-2017 NewFatMike fucked around with this message at 08:51 on Aug 19, 2016 |
|
#
?
Aug 19, 2016 07:43
|
|
|
Palladium posted:Gotta love when Zen is rumored to be competitive with BDW-E, the minds of fanboys go straight to "gimme that for <$200", because competitive performance = competitive pricing doesn't apply to AMD. The dream scenario of AMD reaching price parity with Intel will happen when they overtake Intel in IPC/tech/efficiency, which they might on the last one if their 32 core part indeed only uses 180w. AMD could ask the same price for the highest E7 SKU with that kind of chip. Anime Schoolgirl fucked around with this message at 16:26 on Aug 19, 2016 |
|
#
?
Aug 19, 2016 09:14
|
|
|
NewFatMike posted:E: while I'm here, motherboard stuff for Zen server and consumer stuff: http://www.anandtech.com/show/10581/early-amd-zen-server-cpu-and-motherboard-details-codename-naples-32cores-dual-socket-platforms-q2-2017 Oh my god, green motherboards, I can't remember the last time I saw a green motherboard.
|
|
#
?
Aug 19, 2016 09:42
|
|
|
Anime Schoolgirl posted:Yes, surely because people want it for cheap and not because they've been behind way beyond the margin of competition for forever that they need to regain mindshare by undercutting the high-end just enough while keeping a healthy margin. The dream scenario of AMD reaching price parity with Intel will happen when they overtake Intel in IPC/tech/efficiency, which they might on the last one if their 32 core part indeed only uses 180w. I'm betting AMD manages to get into the mindscape of average consumer and server parts, where only the HEDT users notice Zen lagging behind in IPC. It's going to be kind of unnoticeable in games or office workloads, and if it's 32C/64T @ 2.9Ghz and 180W TDP for ~1200$ then servers will eat that up. I actually think that;'s what Zen's problem might be is clock scaling - it's clearly not an issue with thermals if baseclock between the 32C part and the 8C isn't that different while TDP is less than double. Zen's going to hit 3.5Ghz to 3.8Ghz IMHO, especially for the C stepping. Expect something more geared to clockspeed with D/E Zen stepping and Zen+ (which is apparently a fast follow on to Zen).
|
|
#
?
Aug 19, 2016 11:22
|
|
|
Well, more like $6000, considering the price of E7 Xeons. This is one of the places where AMD benefits from Intel's pricing.
|
|
#
?
Aug 19, 2016 11:26
|
|
|
Also LOL at Intel actually dropping the price of their Prized 8C and 10C BW-E's even if Zen comes in and kicks their teeth in somehow. Didn't Intel still charge like $1000 for a Pentium 4 EE back in the day when just the Athlon 4000 was kicking its rear end all over the place before it even launched? I continue to really hope for the best here though. Intel needs a kick in the nuts and AMD needs a winning architecture that can bring them back into the game full swing. If Zen ends up being great, and they make some APU's with HBM2 for the mobile market that can swing within striking distance of say 25% slower than an Nvidia 1060 (would that be possible?) then they could really have some killer products on the market in the next year or two. I could picture a good bang for the buck laptop being a 4C/8T Zen powered laptop with a 460X level APU with HBM on the die, but would they/could they make such a thing before Nvidia/Intel somehow beat them to the punch? (Intel sure won't, they have pretty much dropped updating the IGPU past what you see with the Iris Pro 580 from what I have seen.)
|
|
#
?
Aug 19, 2016 17:55
|
|
|
SwissArmyDruid posted:Oh my god, green motherboards, I can't remember the last time I saw a green motherboard. Supermicro boards are still all green.
|
|
#
?
Aug 19, 2016 19:03
|
|
|
Isn't the Iris Pro 580 about R7 260/GTX 750 performance?
|
|
#
?
Aug 19, 2016 19:09
|
|
|
 Most server boards are, it's really sad. Part of the reason that I'm waiting for Zen is hoping that core counts on normal sockets will go up, and that maybe more inventive motherboard designs will be more common. I want an ITX board with 4 SO-DIMM slots, and for there to be 32GB SO-DIMMs. Before anyone laughs at me, I'm running out of memory with the 48 GB I have in my desktop, and I'd want to go with the nuclear option instead of 'just' 64 GB. I want to fit 8 cores I mean, that's not actually possible right now, but it's what I want... I want a good video card and a lot of cores and ram and drive space in the smallest possible space. I hope that the Kimera Cerberus becomes a thing instead of just having one run once that will probably come when I have no money, either more cases like it or just continued production... But this is way better for the SFF thread. Sinestro fucked around with this message at 19:20 on Aug 19, 2016 |
|
#
?
Aug 19, 2016 19:13
|
|
|
FaustianQ posted:Isn't the Iris Pro 580 about R7 260/GTX 750 performance? Yes and no. It's massively memory bottlenecked and falls off at high resolution / AA / any other memory-intensive situations. I think on paper it's even faster than those!
|
|
#
?
Aug 19, 2016 19:13
|
|
|
Sinestro posted:
It looks mighty hard to fit a pcie card and use sata on that board, inventive indeed.
|
|
#
?
Aug 19, 2016 19:18
|
|
|
|
| # ? May 15, 2024 18:45 |
|
|
Well, I can't fit more than three drives in there at most, so I think I'd be fine. I mean, spinning ones. I could stick a SATA SSD somewhere in there just hiding since it doesn't need to be that secure. Wish there was an M2 slot somehow, though. Why I really wish MATX cases that weren't nearly as big as ATX cases (or even ATX cases that built to be as small as possible) were a thing.
|
|
#
?
Aug 19, 2016 19:21
|
|