The 2600X is 16.4% behind the 8700K in the link you posted, but if it was clocked at 4.3 all cores and 4.7 single core boost it would still be 16-17% behind , that is at a 16-17% deficit...?.
We'll probably be looking at close to a 20% margin
Well, you have to be careful here. Note that I can run the numbers in the other direction as well, if I want to claim that the Intel gaming IPC advantage is overstated. For example, with 28% of the frame time GPU-bound, I still arrive at the 16% Intel advantage shown by the PT study — but this time with only 5% gaming IPC advantage to Intel.
Frequency advantage for Intel: 16%
Gaming IPC advantage for Intel: 5%
Gaming IPS advantage for Intel: 22%
Intel frame time: 10.0 ms 100%
GPU-bound part of frame time: 2.8 ms 28%
CPU-bound part of frame time: 7.2 ms 72%
AMD frame time: 11.6 ms 100%
GPU-bound part of frame time: 2.8 ms 24%
CPU-bound part of frame time: 8.8 ms 76%
Frame rate for Intel: 100 fps 100%
Frame rate for AMD: 86 fps 86%
Benchmark lead for Intel: 16%
Benchmark loss for AMD: 14%
(Bold numbers estimated/arbitrary, the rest calculated.)
If we could eliminate frequency from the equation, we would get a lower bound on the IPC advantage. So if we could rerun the PT benchmarks at a fixed 4.3 GHz say, and the result now showed a 13% win for Intel, we could claim that the IPC advantage must be at least 13%. Then, to deduce the GPU-bound proportion of the benchmark, we could run the benchmarks with i9-9900K at multiple fixed frequencies to reconstruct the asymptotic curve of frame time over CPU frequency, and hence deduce the GPU-bound portion of the frame time (which would be the lower limit approached as frequency increase). From that we could then estimate the gaming IPC advantage.
You don't expect a 9900K to have a bigger lead at 1080P with a 2080 Ti than a 1080 Ti?!
Look at the 8400 (3.8 ghz all core turbo) vs the 2600x overclocked to 4.1. The lead is 11% at 7% clockspeed deficit, assuming the 8400 maintains full core turbo. So adding those together gives an 18% lead for the 8400.The 2600X is 16.4% behind the 8700K in the link you posted, but if it was clocked at 4.3 all cores and 4.7 single core boost it would still be 16-17% behind , that is at a 16-17% deficit...?.
I wasn't commenting on any particular result — I was just afraid that you were mislead by my post, which demonstrated how to overestimate the IPC lead. I've edited the post to make it clear that you can also underestimate.
PS. I of course agree that with lower GPU-bound part of the frame time, the lead for the faster CPU will grow asymptotically towards the IPC lead. Sorry for the confusion.
Which brings us back to the OP about AMD usurping the 9900K at gaming - we are talking about a 20% IPC uplift in gaming plus 5GHz just to draw level. I don't expect Zen 2 to get over 5GHz, which leaves the 'gaming IPC' as the other factor that can be improved to beat a 9900K. Can AMD really do 20%+ IPC from Zen+ to Zen 2? I have serious doubts, in fact I think it's close to impossible, has a uarch refresh EVER added 20% to gaming?! I cant think of any. AMD would be breaking new ground here if they were to pull it off.
Agree but it doesn't matter much. The mentioned 13% IPC + a 4.5 ghz turbo would probably be good enough. It's on 7nm and no iGPU. The die will be much smaller than intel 9900k while intel has issues with supply due to more cores and probably the iPhone modem. AMD might be a tad slower but the will easily win this as always on price and this round around probably also efficiency.
Actually, even a combo of 5% IPC and 10% frequency would be enough for 3000 series to look very competitive in reviews, and these gains are rather conservative. Sure, we'll still have people staring at 480p 3080Ti benchmarks to highlight that 10% Intel lead, but the market simply won't care anymore. In terms of value over time the 9900K is Achilles with RGB led lighting on his ankle: it will likely be demolished in price/perf by Intel themselves, as they respond to the 3000 series launch.The mentioned 13% IPC + a 4.5 ghz turbo would probably be good enough.
In conclusion, without knowing the GPU-bound part of the frame time, IPC can not be deduced.
Actually, even a combo of 5% IPC and 10% frequency would be enough for 3000 series to look very competitive in reviews, and these gains are rather conservative. Sure, we'll still have people staring at 480p 3080Ti benchmarks to highlight that 10% Intel lead, but the market simply won't care anymore. In terms of value over time the 9900K is Achilles with RGB led lighting on his ankle: it will likely be demolished in price/perf by Intel themselves, as they respond to the 3000 series launch.
Unless AMD made some outstanding mistake in Zen 2, next year will be a very good year for buying CPUs, from either AMD or Intel.
Sure, that is where you see the biggest difference. But even with generally GPU bound game there are passages which are completely CPU bound.
Practically.
In 2015 I had a old PC (I had a good one in work) with C2D e6300@3,16GHz from 2006 with 8GB DDR2-800 slightly oced to 850 MHz. I had gtx260 GPU which failed so I bought gtx760 in 2014.
In 2015, a game I wanted to play came out- Witcher 3. In 1680x1050 (NEC WGX2Pro IPS panel) it was quite a mess on death march difficulty. The running around and city was ok, but boss/general fights were like 20 fps, unplayable even on low details...
Benchmarks on Core2Quad were about 70-80 fps in medium details 1080p around the web, So I upgraded to c2q 9650 and oced it to 3.9GHz (got a good price 80EUR and I waited for Skylake to come out to buy a new platform).
Fps in fight went up, to like 25-30, nothing special. Average went up massively, much smoother in general areas.
Then I bought new board (asrock extreme4 z170, 32GB DDR4 2666- needed that as a replacement for current work CFD precalc machine and i5-6600K, which I oced to 4,4GHz). Average FPS went up a little (10 fps up, GPU bound), but fights with the same GTX760 went up from 25 to 70-80 fps. Finally the death march difficulty was playable and the game enjoyable.
The web reviews didn't show that massive difference at all. It was like 30-40% upgrade, which is a lot, but not 300% upgrade I experienced in the critical CPU intensive scenes- boss/fights where I needed the fps as high as possible.
If you asked around the web for upgrade for my system, pretty much 90% of the people recommended upgrade the gtx760, that the low low end for such a demanding game like wither 3. But that wasn't the case.
I am not saying that is the case with ryzen vs cores now, but I am saying that the average benchmarking can be very misleading.
https://www.hardware.fr/articles/965-3/performances-jeux-3d.htmlIn conclusion, without knowing the GPU-bound part of the frame time, IPC can not be deduced.
Yup but you look at a lot of benchmarks from different sources and you still find differences from minimal to 30% ahead.I should also mention that 'gaming IPC' is just an estimate because the choice of games greatly influences the result of such a metric.
Its probably easy to 'cherry pick' games that shows Intel to be 30% ahead per clock, and perhaps just as easy to hand pick games that shows AMD to have minimal deficit per clock.
2400G: 4MB L3 for 8 threadsOn the AMD side the 2600x is almost 40% faster than the 2400g while the difference in cores is 50%,so the GPU starts to bottleneck at -10% of 6c/12t at ... 4Ghz?
soon, it begins, better strap in folks, this one's a hot one
Well, here is an 8700K running CB multi at a 110W package TDP:Care to elaborate a bit more?
Indeed sir, that was I wanted to say. Intel has already a hard time with AMD, but now they have 3 other more to watch. And fullfil all their disadvantages in order to maintain their leadership is impossible.Please stay on topic. If you diverge, try at least to relate it to the discussion. Thanks.
Will Zen 2 with a 12% IPC uplift at 4.5ghz best a 9900k at 5.0?
It's misleading if it isn't grounded in reality. Base frequency has no use case in reality anymore since the CPU tries to drive higher frequencies whenever possible which is essentially always. Compare this to AMD's TDP which for Ryzen 2xxx is a hard upper limit that is always enforced unless deactivated by enabling PBO (Precision Boost Overdrive).BTW, Intel has always clearly stated how they arrive at the published TDP number.
It's only misleading of you don't read what Intel posts.
Click on the ? next to Processor Base Frequency and TDP and Thermal Solution Specification.
https://ark.intel.com/products/186605/Intel-Core-i9-9900K-Processor-16M-Cache-up-to-5-00-GHz-
So what's your proof that cache has any effect at all on these numbers?I mean these games are all coded for jaguar with zero cache per core...2400G: 4MB L3 for 8 threads
2600X: 16MB L3 for 12 threads
People aren't even trying anymore.