The only reason to prefer an older, power-hungry architecture over Zen 2 was 1080p gaming, value be damned.
Now that it appears Zen 3 will hold all the crowns *including gaming and single-threaded performance* people are arguing that an unreleased, backported architecture on 14++++++++++ will somehow do better... beggars belief.
Sorry, but anyone who cares about value won't buy either Zen 3 or Infinity Lake+++. They're probably already smiling while gaming on their $99 R5 2600 or similar budget buy.
Now that it appears Zen 3 will hold all the crowns *including gaming and single-threaded performance* people are arguing that an unreleased, backported architecture on 14++++++++++ will somehow do better... beggars belief.
Well, if Zen 3 is 5-7% faster than Comet Lake in most games, maybe a tad less when OCed... you'd think Rocket Lake would be able to do better than that.
I have seen it multiple places. I just don't want to bother finding all the links. One test is not enough for you ? why do you care, since you only buy Intel ?
Well, if Zen 3 is 5-7% faster than Comet Lake in most games, maybe a tad less when OCed... you'd think Rocket Lake would be able to do better than that.
Why would I think anything on 14nm would do better than the most advanced (to date) 7nm processor ? I even got "beat up" so to speak when I did not realize that Raspberry PI was 28 nm. (other thread)
Wow, are people really arguing about 2 products that have yet to be out in the wild and available for 3rd party benchmarks? We don't have real numbers for either of them, though it is likely the AMD numbers from the event are close enough.
And the only actual numbers we have on Rocketlake are speculation, and marketing from Intel? We don't even have an actual release date, just some vague hand waving.
Settle down folks, sheesh. We have less than a month for real numbers for Ryzen 3, and some unknown amount of time for the Intel part. Wait for them to be benchmarked, then start arguing.
Settle down folks, sheesh. We have less than a month for real numbers for Ryzen 3, and some unknown amount of time for the Intel part. Wait for them to be benchmarked, then start arguing.
The MHz war is next! Performance don't count it's all about the Mhz! Makes the e-peen bigger! Performance is nothing more than those low hanging fruit.
Vermeer's not going to work for enterprise. You need the IGP. Doesn't need to be much, Vega 3 would be perfect, but you need the IGP. Given the demand destruction brought on by The Rona, you have to think that AMD will not focus too much on enterprise desktop despite being most of the market.
Isn't Van Gogh really a IoT type chip that might also be a Dali replacement? The whole point of it is to use the GPU features on the edge, the CPU is just there. The competitive Intel product is Elkhart Lake maybe?, not Tiger Lake.
But with Matisse that was with two CCX per CCD which each had to have a separate connection to the IOD. So Matisse had two full read connection and two half write connections to each CCD. Since with Vermeer the IOD is supposedly unchanged this would mean each CCX (which now equals a CCD) gets two full read and one full write connection to the IOD.
The confidential slides with Adobe Premier and CAD were interesting if true. If AMD were not fibbing in those, then that would indicate their ST changes void the need of having a higher speed processor to be faster, especially vs. a processor with QuickSync due to the igpu?
From your own link going from 3600CL16 to 3733CL14 on Zen 2 leads to a 2% gain, while on Skylake going from 3600CL16 to 4133CL17 leads to a 13% gain. So Skylake gains more from memory overclocking than Zen 2, as for Zen 3 the jury is out there as to how much it can gain from overclocking memory, but we know that the IO die is the same, meaning that no improvements have been made to the memory controller, so chances of Zen 3 suddenly becoming better than Zen 2 in memory scaling don't seem very high.
Not even true and that goes to Intel number aswell. You can way more on both systems. I gain with 3900x from stock 3600mhz XMP (CL16) to CL14 with dram calc fast subtimings.. 20-30 fps better low fps in 64 player BFV.
Problem is no reviewer actually test multiplayer scenarios. I had, now sold 3900x, like 130 fps low fps in 64 player BFV. A overclocked 9900k with 4300mhz low timings ram has like 150fps. So it will be interesting to see what my 5900x will give in same scenario.
Btw guys DX12 in battlefield on ryzen is garbage, no one uses that. So if Zen3 is that close to 10900k in BFV with DX12 means its ahead when it uses dx11.
That was my 6th post in the thread. I have no data, that's why I used the word maybe. Besides, what's the alternative? Swallowing all the data from the launch slides because it comes from a chipmaker? Where have you been when Intel launches their products then? Oh, wait! You're usually right there in the mix, shooting everything down. Also, for your information, I called nobody out. I simply speculated (as everyone is doing) that the omission of a cpu bound game by AMD may not have been an accident. The slides are marketing slides, afterall.
It's clear you weren't paying attention so go back and do your home work before attacking me. You didn't even know which is my first post in this thread.
I had good things to say about Zen, Zen 2, and Renoir. Skepticism is still allowed on Anandtech, right? Or, is it only allowed in Intel threads?
2. Regarding Rocket Lake, I said that it will be comfortably ahead in gaming IF it clocks to ~5GHz. That's speculation, true - but it is speculation based on known facts which makes it likely to be a highly probable outcome. What response did it provoke from the AMDeniers? - "nope, nada, can't be true", accusations of bias, refusal to acknowledge outliers in the data, appeals to authority etc. If you find my approach problematic, what do you have to say about these people?
Your claim that Rocket Lake (running at 5 Ghz) will be significantly faster in gaming than Skylake (running at 5.1-5.3 Ghz) is still pure speculation and there are plenty of early hints around that say that this might not be the case.
Yes, it will be faster in ST and lightly threaded benchmarks (e.g. Cinebench ST). It will be siginificantly faster in software supporting AVX-512 (and some AVX-2 tests). Known Sunny- and Willow Cove benches (+ Rocket Lake frequency leaks) are enough to deduce that with a high degree of accuracy.
But it's waaay too early to claim anything about gaming yet:
Gaming doesn't really benefit much from general "higher IPC" especially if a lot of it comes from wider vector units (as is the case with Zen 2 and Sunny Cove). It cares about memory subsystem, particularily the cachess and memory latency [1]
We know for certain that Rocket Lake has Sunny Cove memory-hierarchy not Willow Cove. This means that it will still have 16MB of L3. Only the L2 has doubled in size.
How does the Ice Lake memory subsystem perform vs Skylake and Tiger Lake? Well luckily Anandtech has an article comparing these three in detail. Willow Cove improves things quite a bit but Sunny Cove? Not so much really.
TL;DR:
A 5.0 Ghz Rocket Lake will probably be ever so slightly faster in gaming than a 5.3 Ghz Skylake But as the memory-interface has not changed much the gains will be very limited (it might end up being almost a wash). That's not enough to put Intel "Comfortably ahead", particularily if you consider that AMD use to be 10% down at stock speed and now is 5-6% ahead instead. And that's when you ignore up to ~20% in cache-sensitive competitive titles.
Even if Rocket Lake somehow wins (which is not a given) it will not be "comfortable". It will only be low single-digit percent on average while still losing in a lot of titles. Heavily in some competitive ones (that actually have quite large player-bases).
[1] See this 3200G vs 4350G comparison with a discrete GPU . 4350G is miles faster in ST CPU-Z (both have the same caches and boost-clock), yet in games it's barely any better (despite having the SMT edge). Now compare this to 3200G vs 3300X, where the latter wins by 30-50%. The only real difference between 4350G and 3300X is cache.
Appendices
[1] See this 3200G vs 4350G comparison with a discrete GPU . 4350G is miles faster in ST CPU-Z (both have the same caches and boost-clock), yet in games it's barely any better (despite having the SMT edge). Now compare this to 3200G vs 3300X, where the latter wins by 30-50%. The only real difference between 4350G and 3300X is cache.
RocketLake is a gimped SunnyCove core, there is no comaprison with Zen3. Zen3 is ~5-7% faster than latest iteration of Coves (Willow Cove) so whatever Cypress Cove( the thing in RocketLake) brings will be too little, too hot. Pun intended.
The whole point of the message was to say that as Rocket Lake doesn't increase the L3 cache size or improve anything considerably in the memory subsystem, it won't see much gains in games, even when it has a significantly higher IPC vs Skylake.
4350G struggles to really differenciate itself from 3200G despite having 15% higher IPC (and SMT enabled) with the cache-sizes and clocks were equal. 3300X blows the doors off 3200G despite being essentially a 3450G with 4x the cache and a single-digit % boost-clock bump.
Geez, people are comparing performance between two CPUs that aren't even for sale yet.
Comparing gaming data (FPS) from different games and treating them like as if they are from as if they are related data points (they are not, they are independent).
And other general debates about essentially unknowable information (well, unless you work at AMD or Intel).
This is embarrassing.
WillowCove uses the non-inclusive cache subsystem like Zen3 and Skylake-X which performs better on independent threads because L2 does not contain L1 copies and L3 does not contain L2 copies. And when it comes to CypressCove, the cache system is inclusive like Skylake and SunnyCove, which works better in interdependent threads, i.e. in games. I think it is quite possible that CypressCove will have an IPC comparable or higher than Zen3 at least in games is this chance.
WillowCove can gain compared to SunnyCove, but it can also significantly lose, mainly due to the non-inclusive cache system used.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.