The AMD Zen and Ryzen 7 Review: A Deep Dive on 1800X, 1700X and 1700
by Ian Cutress on March 2, 2017 9:00 AM ESTFetch
For Zen, AMD has implemented a decoupled branch predictor. This allows support to speculate on incoming instruction pointers to fill a queue, as well as look for direct and indirect targets. The branch target buffer (BTB) for Zen is described as ‘large’ but with no numbers as of yet, however there is an L1/L2 hierarchical arrangement for the BTB. For comparison, Bulldozer afforded a 512-entry, 4-way L1 BTB with a single cycle latency, and a 5120 entry, 5-way L2 BTB with additional latency; AMD doesn’t state that Zen is larger, just that it is large and supports dual branches. The 32 entry return stack for indirect targets is also devoid of entry numbers at this point as well.
The decoupled branch predictor also allows it to run ahead of instruction fetches and fill the queues based on the internal algorithms. Going too far into a specific branch that fails will obviously incur a power penalty, but successes will help with latency and memory parallelism.
The Translation Lookaside Buffer (TLB) in the branch prediction looks for recent virtual memory translations of physical addresses to reduce load latency, and operates in three levels: L0 with 8 entries of any page size, L1 with 64 entries of any page size, and L2 with 512 entries and support for 4K and 256K pages only. The L2 won’t support 1G pages as the L1 can already support 64 of them, and implementing 1G support at the L2 level is a more complex addition (there may also be power/die area benefits).
When the instruction comes through as a recently used one, it acquires a micro-tag and is set via the op-cache, otherwise it is placed into the instruction cache for decode. The L1-Instruction Cache can also accept 32 Bytes/cycle from the L2 cache as other instructions are placed through the load/store unit for another cycle around for execution.
Decode
The instruction cache will then send the data through the decoder, which can decode four instructions per cycle. As mentioned previously, the decoder can fuse operations together in a fast-path, such that a single micro-op will go through to the micro-op queue but still represent two instructions, but these will be split when hitting the schedulers. The purpose of this allows the system to fit more into the micro-op queue and afford a higher throughput when possible.
The new Stack Engine comes into play between the queue and the dispatch, allowing for a low-power address generation when it is already known from previous cycles. This allows the system to save power from going through the AGU and cycling back around to the caches.
Finally, the dispatch can apply six instructions per cycle, at a maximum rate of 6/cycle to the INT scheduler or 4/cycle to the FP scheduler. We confirmed with AMD that the dispatch unit can simultaneously dispatch to both INT and FP inside the same cycle, which can maximize throughput (the alternative would be to alternate each cycle, which reduces efficiency). We are told that the operations used in Zen for the uOp cache are ‘pretty dense’, and equivalent to x86 operations in most cases.
Facebook
Twitter
RSS
574 Comments
View All Comments
bobsta22 - Saturday, March 4, 2017 - link
Office with 20 PCs - all developers - loads of VMs and containers.All the PCs are due a CPU/Gfx refresh, but ITX mobos required.
Cant wait tbh. This is a game changer.
prisonerX - Saturday, March 4, 2017 - link
What if they come out with a 16 core line next year!bobsta22 - Saturday, March 4, 2017 - link
What?lilmoe - Tuesday, March 7, 2017 - link
It really is. As a freelance developer, I can't wait.ericgl21 - Saturday, March 4, 2017 - link
For me, the more important thing to see from AMD is if they can come up with a chip that can beat the mobile Core i7-7820HQ (4c/8t no ECC) & the Xeon E3-1575M v5 (4c/8t with ECC), for less money.And the number of PCIe gen3 lanes is very important, especially with the rise of M.2 NVMe storage sticks.
cmagic - Sunday, March 5, 2017 - link
Will anandtech review Ryzen in gaming? I would really like Anandtech view, since I don't really trust other sites especially those "entertainment" sites. Want to see how Anandtech dive into its main cause.Tchamber - Sunday, March 5, 2017 - link
@cmagicPage 15
2017 GPU
The bad news for our Ryzen review is that our new 2017 GPU testing stack not yet complete. We recieved our Ryzen CPU samples on February 21st, and tested in the hotel at the event for 6hr before flying back to Europe.
I just ordered my 1700X, I plan to keep it for at least 5 years, as my needs don't change much. My current Intel 6 core is coming up on 7 years old now. I like to buy high end and use it a long time.
Lazlo Panaflex - Monday, March 6, 2017 - link
Same here...probably gonna grab a 1700 at some point and put this here i5-2500 non-k in the kids computer.asH98 - Sunday, March 5, 2017 - link
'''The BIG QUESTION is WHY are the HEDT benchmarks (professional ie Blender) fairer than gaming benchmarks??Bottom line is that CUTTING-EDGE CODING is happening NOW in AI, HPC, data, and AV/AR, game coders because of $$$ are the last to change or learn unless forced (great for NVidia Intel) so most of the game coding is stuck in yesteryear- Bethesda will be the test bed for game coders to move forward
Hence the difference in game benchmarks vs 'professional' (HEDT) benchmarks. Game coders can get stuck using yesterday's code without repercussions and consequences as long as old hardware dominates and there are no incentives to change or learn new skills. The same Cant happen in the Professional area where speed is tantamount to performance and $$$
TheJian - Sunday, March 5, 2017 - link
I hope you're going to test a dozen games at 1080p where most of us run for article #2 and the GAMING article should come in a week not 1/2 year later like 1080/1070 gtx reviews...LOL. As this article just seems like AMD told you "guys, please don't run any games so we can sell some chips to suckers before they figure out games suck". And you listened. No point in testing 1440p or 4k for CPU, and 95% of us run 1920x1200 and BELOW so you should be testing your games there for a CPU test.The fact they are talking Zen2 instead of fixing Zen1 kind of makes me think most of the gaming is NOT going to be fixable.
http://www.legitreviews.com/amd-ryzen-7-1800x-1700...
149fps for 7700 in theif vs. 108 for 1800x? JEEZ. GTA5 again, 163 to 138. Deus ex MD 127 to 103. These are massive losses to Intel's chip and Deus was clearly gpu bound as many of Intel's chips hit the same 127fps including my old 4790k :( OUCH AMD.
https://www.guru3d.com/articles_pages/amd_ryzen_7_...
Tombraider same 7700k vs. 1800x 132fps to 114 (never mind 6850 scoring 140fps). This will probably get worse as we move to 1080ti, vega, nvidia refresh for xmas, Volta, 10nm etc. If you were using a faster gpu the cpus will separate even more especially if people are mostly gaming at 1080p. Even if many move to 1440p, that maybe fixes some games (tombraider is one with 1080 regular that hits a wall at 90fps), but again goes back to major losses as we move to 10nm etc. We get 10nm chips for mobile now and gpus probably next year at 12nm (real? fake 12nm? Either way) and might squeak into 2017 (volta, TSMC). 10nm gpus will likely come 2018 at the latest. Those gpus will make 1440p look like 1080p today surely and cpus will again spread out (and no, we won't all be running 4k then...LOL). You could see cpus smaller than 10nm BEFORE you upgrade your cpu again if you buy this year. That could get pretty ugly if the benchmarks around the web for gaming are not going to improve. One more point you'll likely be looking at GDDR6 (16Gbps probably) for vid cards allowing them to possibly stretch their legs even more if needed. Again, all not good for a gamer here IMHO.
“But Senior Engineer Mike Clark says he knows where the easy gains are for Zen 2, and they're already working through the list”
So maybe no fix in sight for Zen1? Just excuses like "run higher res, and code right guys"...I hope that isn't the best they've got. I could go on about games, but most should get the point. I was going to buy ryzen purely for Handbrake, but I'll need to see motherboard improvements and at least some movement on gaming VERY soon.
One more ouch statement from pcper.
https://www.pcper.com/news/Processors/AMD-responds...
"For buyers today that are gaming at 1080p, the situation is likely to remain as we have presented it going forward."
So they don't think a fix is coming based on AMD info and as noted as gpus get much faster (along with their memory speeds) expect 1440p to look like today's 1080p benchmarks at least to some extent.
The board part is of major interest to me, so I can wait a bit and also see Intel's response. So AMD has be hanging for a bit here, but not for too long. I do like the pro side of these though (handbrake especially, just not quite enough).