AMD Ryzen (Summit Ridge) Benchmarks Thread (use new thread)

Page 238 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Status
Not open for further replies.

OrangeKhrush

Senior member
Feb 11, 2017
220
343
96
Looks like the same source that Dresdenboy and I were speculating on a month ago that everyone denounced as "fake". I tried to say it may not be at Overclock.net and they were still locking threads and editing posts.


Alva is pretty well known.

People "fanboys" will denounce anything that doesn't suite their inuendo as fake
 

OrangeKhrush

Senior member
Feb 11, 2017
220
343
96
OrangeKhrush - "See guys, ignore the multitude of credible benchmarks showing Ryzen as competitive and focus on this cropped screenshot posted anonymously on chinese forums."


The rear guard action is strong with this one. You keep fighting that good fight, soldier. Never give up, even in the face of insurmountable odds.

What? It is a very credible leak that falls in with the numbers
 

lolfail9001

Golden Member
Sep 9, 2016
1,056
353
96
Do you know what it would use these instructions for?
I have little clue (because i did not touch it yet) but it is the only viable explanation of jump between Haswell and Ivy, don't you think? Well, either that or
AVX2 is largely an integer SIMD extension, so Intel’s architects applied the same conceptual technique to achieve single cycle 256-bit SIMD integer execution. Haswell adds 128-bit integer SIMD execution capabilities to the existing 128-bit FP stack, while re-using the FP bypass network to forward half of the 256-bit results; again saving area and power. This means that Figure 3 is more representative of the logical view of the execution units, rather than the underlying physical implementation. In essence, the FP and SIMD stacks are now identical, forming the lower and upper halves of the data path. Both stacks are 128-bits wide and have the same FP and SIMD execution units.
http://www.realworldtech.com/haswell-cpu/4/
 

PPB

Golden Member
Jul 5, 2013
1,118
168
106
The INTERNET STRONGMAN es our inverse John Freue. Every time he says IPC is lower than expected, IPC goes up. He should keep posting some more :)
 
  • Like
Reactions: CHADBOGA

OrangeKhrush

Senior member
Feb 11, 2017
220
343
96
Juan is going to be the most roasted man on the internet. He better get the KY ready.

Anyways it looks like Ryzen is about 5% off Broadwell and 8% off Skylake clock for clock which is a really good effort.
 

blublub

Member
Jul 19, 2016
135
61
101
Juan is going to be the most roasted man on the internet. He better get the KY ready.

Anyways it looks like Ryzen is about 5% off Broadwell and 8% off Skylake clock for clock which is a really good effort.
What was Juan's IPC prediction ? SB level?
 

dogen1

Senior member
Oct 14, 2014
739
40
91
I have little clue (because i did not touch it yet) but it is the only viable explanation of jump between Haswell and Ivy, don't you think? Well, either that or

http://www.realworldtech.com/haswell-cpu/4/

Haswell added a lot of things other than AVX and FMA instructions.

Scroll down a bit on this page and he details some of them.
http://www.emulators.com/docs/nx36_haswell_262.htm

I don't think Dolphin benefits much (or even can benefit?) from AVX/AVX2, as the gamecube had only 2 wide SIMD.

But it would benefit from the MOVBE and BMI instructions, zero latency register moves, and probably the 4th ALU.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Dolphin doesn't have any MOVBE instructions in it, and the only intruction belonging to BMI(1) is TZCNT.

No 256-bit instructions either.

This is for the public 5.0 binary available at the site.

Total / Count / Percentage

T: 309 T: 2380642 T: 100%
---------- ----------- ------------
mov 822813 34.5627
lea 262551 11.0286
call 192047 8.0670
jmp 102408 4.3017
add 96287 4.0446
cmp 94477 3.9686
jz 78576 3.3006
test 70349 2.9550
sub 56230 2.3620
nop 54053 2.2705
pop 52257 2.1951
xor 50990 2.1419
push 45898 1.9280
jnz 42797 1.7977
retn 41115 1.7271
movzx 33818 1.4205
movups 23249 0.9766
inc 21878 0.9190
and 21147 0.8883
movsd 19346 0.8126
jb 18031 0.7574
or 15274 0.6416
shr 13676 0.5745
movaps 11497 0.4829
movsxd 10646 0.4472
sar 9534 0.4005
imul 9081 0.3815
int 8309 0.3490
shl 7208 0.3028
jnb 5981 0.2512
jle 4232 0.1778
ja 3770 0.1584
jl 3262 0.1370
jbe 3219 0.1352
movss 2834 0.1190
movsx 2663 0.1119
neg 2640 0.1109
movq 2638 0.1108
movdqa 2462 0.1034
xorps 2389 0.1004
dec 2336 0.0981
ror 2225 0.0935
movdqu 2081 0.0874
movd 1939 0.0814
cmovnz 1875 0.0788
jge 1776 0.0746
bt 1766 0.0742
jns 1516 0.0637
bts 1490 0.0626
setz 1433 0.0602
paddw 1327 0.0557
mulss 1286 0.0540
rol 1162 0.0488
cdq 1161 0.0488
setnz 1153 0.0484
jg 1122 0.0471
mulsd 1118 0.0470
js 1109 0.0466
btr 1089 0.0457
cmovnb 1039 0.0436
xchg 1005 0.0422
cmovz 979 0.0411
cmovg 913 0.0384
cdqe 895 0.0376
cmovb 887 0.0373
addss 865 0.0363
bswap 712 0.0299
cvtdq2ps 711 0.0299
addsd 701 0.0294
pmaddwd 664 0.0279
paddd 657 0.0276
not 649 0.0273
psubw 627 0.0263
pxor 578 0.0243
cvtdq2pd 524 0.0220
cmova 490 0.0206
cvttsd2si 479 0.0201
punpcklbw 452 0.0190
cvtps2pd 418 0.0176
divsd 412 0.0173
div 409 0.0172
comisd 400 0.0168
subss 396 0.0166
pmullw 391 0.0164
divss 384 0.0161
idiv 373 0.0157
mul 368 0.0155
psrad 346 0.0145
vmovdqa 341 0.0143
subsd 319 0.0134
pavgb 304 0.0128
vpmaddwd 292 0.0123
sbb 280 0.0118
comiss 274 0.0115
cvtsi2ss 256 0.0108
cmovl 253 0.0106
cvtsi2sd 241 0.0101
vpaddd 236 0.0099
setb 232 0.0097
mulpd 223 0.0094
cqo 212 0.0089
movapd 202 0.0085
xadd 201 0.0084
adc 199 0.0084
packuswb 196 0.0082
pand 195 0.0082
paddusw 188 0.0079
vmovaps 184 0.0077
psrldq 182 0.0076
jp 181 0.0076
punpckhbw 178 0.0075
cvttss2si 177 0.0074
psrlw 171 0.0072
vpsrad 157 0.0066
cvtpd2ps 156 0.0066
cmovs 156 0.0066
por 155 0.0065
cmovo 149 0.0063
punpcklwd 147 0.0062
cvtsd2ss 145 0.0061
phaddd 144 0.0060
addpd 143 0.0060
cmovge 137 0.0058
packssdw 137 0.0058
punpckldq 132 0.0055
setnle 126 0.0053
psubd 124 0.0052
subpd 121 0.0051
ucomisd 120 0.0050
pshufb 118 0.0050
setnbe 114 0.0048
vpsubd 112 0.0047
movhps 109 0.0046
psraw 106 0.0045
cmovbe 100 0.0042
pshufd 100 0.0042
pavgusb 95 0.0040
cmovle 93 0.0039
psadbw 92 0.0039
punpckhwd 87 0.0037
setnl 76 0.0032
pmaxsw 74 0.0031
vpackssdw 74 0.0031
punpcklqdq 74 0.0031
stos 73 0.0031
punpckhdq 72 0.0030
psrlq 67 0.0028
cwde 67 0.0028
mulps 64 0.0027
setnb 61 0.0026
ucomiss 61 0.0026
punpckhqdq 60 0.0025
psubb 60 0.0025
andps 59 0.0025
pshuflw 57 0.0024
pmaddubsw 56 0.0024
setl 53 0.0022
vpunpckhwd 52 0.0022
sets 52 0.0022
vpunpcklwd 52 0.0022
pshufw 52 0.0022
psllw 52 0.0022
cvtps2dq 51 0.0021
pmulld 50 0.0021
minps 49 0.0021
cmpxchg 49 0.0021
pcmpgtw 48 0.0020
vmovdqu 47 0.0020
psllq 44 0.0018
psubusb 42 0.0018
cvtss2sd 38 0.0016
unpcklps 37 0.0016
paddsw 36 0.0015
addps 35 0.0015
shufps 34 0.0014
movhlps 34 0.0014
vpsrlw 33 0.0014
maxss 33 0.0014
unpcklpd 31 0.0013
paddb 31 0.0013
minsd 30 0.0013
bsf 30 0.0013
pminsd 30 0.0013
bsr 28 0.0012
pminsw 28 0.0012
pslld 28 0.0012
vpminsw 26 0.0011
setns 26 0.0011
setbe 24 0.0010
vpmaxsw 24 0.0010
vpunpcklqdq 24 0.0010
unpckhpd 24 0.0010
pmovsxwd 24 0.0010
maxsd 23 0.0010
pshufhw 23 0.0010
pextrw 22 0.0009
vpand 21 0.0009
xorpd 20 0.0008
psrld 19 0.0008
pandn 18 0.0008
andpd 18 0.0008
vpaddw 18 0.0008
vmovd 18 0.0008
vpackuswb 17 0.0007
setle 17 0.0007
vpunpckhqdq 16 0.0007
packsswb 16 0.0007
vpunpckldq 16 0.0007
vpunpckhdq 16 0.0007
pmulhrsw 16 0.0007
movlhps 16 0.0007
pabsw 16 0.0007
movhpd 15 0.0006
pslldq 15 0.0006
subps 15 0.0006
vpmullw 14 0.0006
cpuid 14 0.0006
pcmpeqb 14 0.0006
paddq 13 0.0005
vmovq 12 0.0005
vpsraw 12 0.0005
vpaddsw 12 0.0005
pavgw 11 0.0005
orpd 11 0.0005
movs 10 0.0004
minss 9 0.0004
pinsrw 9 0.0004
vpbroadcastb 9 0.0004
pcmpeqd 9 0.0004
outs 9 0.0004
vpsubw 8 0.0003
vpxor 8 0.0003
cvttpd2dq 8 0.0003
vpshufb 8 0.0003
crc32 8 0.0003
sahf 7 0.0003
cmpltpd 7 0.0003
vpcmpeqb 6 0.0003
vpackusdw 6 0.0003
vpavgw 6 0.0003
lods 6 0.0003
packusdw 6 0.0003
pcmpgtd 6 0.0003
pmovsxdq 6 0.0003
vpshuflw 6 0.0003
pmaxsd 5 0.0002
scas 5 0.0002
jrcxz 4 0.0002
cli 4 0.0002
vpsllw 4 0.0002
pmovzxwd 4 0.0002
ins 4 0.0002
in 4 0.0002
movmskpd 4 0.0002
paddusb 4 0.0002
xgetbv 3 0.0001
tzcnt 3 0.0001
stmxcsr 3 0.0001
retf 3 0.0001
vpshufd 3 0.0001
vmovhps 3 0.0001
ldmxcsr 3 0.0001
leave 3 0.0001
fstp 3 0.0001
clc 2 0.0001
pmovmskb 2 0.0001
vpalignr 2 0.0001
pcmpgtb 2 0.0001
lahf 2 0.0001
cvtsd2si 2 0.0001
vpslld 2 0.0001
stc 2 0.0001
std 2 0.0001
cmpunordpd 2 0.0001
cmps 2 0.0001
vpunpcklbw 2 0.0001
enter 2 0.0001
cmpeqpd 2 0.0001
pushfq 2 0.0001
cmovns 2 0.0001
rdtsc 2 0.0001
psubsw 2 0.0001
movmskps 2 0.0001
wait 1 0.0000
psubq 1 0.0000
palignr 1 0.0000
divps 1 0.0000
emms 1 0.0000
cmpnltpd 1 0.0000
hlt 1 0.0000
iret 1 0.0000
xlat 1 0.0000
rcl 1 0.0000
maxps 1 0.0000
fistp 1 0.0000
roundpd 1 0.0000
cmc 1 0.0000
loopne 1 0.0000
popfq 1 0.0000
jo 1 0.0000
fcomp 1 0.0000
loop 1 0.0000
prefetcht0 1 0.0000
fisttp 1 0.0000
shufpd 1 0.0000
cvttps2dq 1 0.0000
cmpnlepd 1 0.0000
cld 1 0.0000
btc 1 0.0000
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
o_Oo_Oo_O



Exclusive ? Are they under NDA or they just got source?
Its another Passmark result that anyone can see as a click bait article, they dont know if turbo was running or not. The other result was removed from passmark, but you can still see it if you had added before it was removed.
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
Again they dont know if turbo was enabled or not. Nor they can be sure about the turbo or clock of the other CPUs...
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
Hey look, this 3.6Ghz I7-4790 is kicking Ryzen ass at ST when both at 3.6Ghz, because it cant read turbo clocks :/

QENBcZd.png


Can we please not fall in those clickbait articles anymore?

For someone that doest get it: Look at the I3 score, you can trust that score, because I3 dont have Turbo and cant be overclocked, the Haswell ST score is petty acurate, the I7 is getting 250 more points, ergo is running turbo enabled to 4Ghz, but the program is unable to read it.
 
Last edited:
  • Like
Reactions: Sweepr

OrangeKhrush

Senior member
Feb 11, 2017
220
343
96
Again they dont know if turbo was enabled or not. Nor they can be sure about the turbo or clock of the other CPUs...

Yes they can, the other cpus have recorded turbos and the 1800X like the other had 1700X which could be firmware disabled. Passmark has a built in CPUID which detects turbo on or off through BIOS. It was a 3.6 baseline validation score.
 

inf64

Diamond Member
Mar 11, 2011
3,684
3,954
136
You need to remember that for the number AMD stated until very recently (40% over XV), SB kind of IPC is pretty accurate ;)
Nobody got the clock right, thou. Neither did I :)
I did some calculations from majord's results and it turns out SB should have 29-30% higher IPC than XV core. He got 61% higher IPC (with AVX2 in the mix) for SKL Vs XV and SKL is around 24-25% faster than SB according to AT's tests. 40% above XV would be 1.4/1.3~=8% above SB level. So yes quite close but still higher. IB is 5.8% above SB so that would still put Zen above even IB. Now if Zen is even faster than what AMD claimed(sandbagged) then it should be approaching Haswell which in turn should be ~52% faster than XV core.
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
Yes they can, the other cpus have recorded turbos and the 1800X like the other had 1700X which could be firmware disabled. Passmark has a built in CPUID which detects turbo on or off through BIOS. It was a 3.6 baseline validation score.

Read my previous post.
 

CatMerc

Golden Member
Jul 16, 2016
1,114
1,149
136
You need to remember that for the number AMD stated until very recently (40% over XV), SB kind of IPC is pretty accurate ;)
Nobody got the clock right, thou. Neither did I :)
I am aware, however he did mention it would be "best case". As in, only in cherry picked scenarios.
 
Status
Not open for further replies.