diff --git a/ratelimit.go b/ratelimit.go
index bc320f0..876e98a 100644
--- a/ratelimit.go
+++ b/ratelimit.go
@@ -4,13 +4,14 @@ import (
 	"net/http"
 	"strconv"
 	"sync"
+	"sync/atomic"
 	"time"
 )
 
 // RateLimiter holds all ratelimit buckets
 type RateLimiter struct {
 	sync.Mutex
-	global          *Bucket
+	global          *int64
 	buckets         map[string]*Bucket
 	globalRateLimit time.Duration
 }
@@ -20,7 +21,7 @@ func NewRatelimiter() *RateLimiter {
 
 	return &RateLimiter{
 		buckets: make(map[string]*Bucket),
-		global:  &Bucket{Key: "global"},
+		global:  new(int64),
 	}
 }
 
@@ -58,8 +59,10 @@ func (r *RateLimiter) LockBucket(bucketID string) *Bucket {
 	}
 
 	// Check for global ratelimits
-	r.global.Lock()
-	r.global.Unlock()
+	sleepTo := time.Unix(0, atomic.LoadInt64(r.global))
+	if now := time.Now(); now.Before(sleepTo) {
+		time.Sleep(sleepTo.Sub(now))
+	}
 
 	b.remaining--
 	return b
@@ -72,7 +75,7 @@ type Bucket struct {
 	remaining int
 	limit     int
 	reset     time.Time
-	global    *Bucket
+	global    *int64
 }
 
 // Release unlocks the bucket and reads the headers to update the buckets ratelimit info
@@ -89,41 +92,25 @@ func (b *Bucket) Release(headers http.Header) error {
 	global := headers.Get("X-RateLimit-Global")
 	retryAfter := headers.Get("Retry-After")
 
-	// If it's global just keep the main ratelimit mutex locked
-	if global != "" {
-		parsedAfter, err := strconv.Atoi(retryAfter)
-		if err != nil {
-			return err
-		}
-
-		// Lock it in a new goroutine so that this isn't a blocking call
-		go func() {
-			// Make sure if several requests were waiting we don't sleep for n * retry-after
-			// where n is the amount of requests that were going on
-			sleepTo := time.Now().Add(time.Duration(parsedAfter) * time.Millisecond)
-
-			b.global.Lock()
-
-			sleepDuration := sleepTo.Sub(time.Now())
-			if sleepDuration > 0 {
-				time.Sleep(sleepDuration)
-			}
-
-			b.global.Unlock()
-		}()
-
-		return nil
-	}
-
-	// Update reset time if either retry after or reset headers are present
-	// Prefer retryafter because it's more accurate with time sync and whatnot
+	// Update global and per bucket reset time if the proper headers are available
+	// If global is set, then it will block all buckets until after Retry-After
+	// If Retry-After without global is provided it will use that for the new reset
+	// time since it's more accurate than X-RateLimit-Reset.
+	// If Retry-After after is not proided, it will update the reset time from X-RateLimit-Reset
 	if retryAfter != "" {
 		parsedAfter, err := strconv.ParseInt(retryAfter, 10, 64)
 		if err != nil {
 			return err
 		}
-		b.reset = time.Now().Add(time.Duration(parsedAfter) * time.Millisecond)
 
+		resetAt := time.Now().Add(time.Duration(parsedAfter) * time.Millisecond)
+
+		// Lock either this single bucket or all buckets
+		if global != "" {
+			atomic.StoreInt64(b.global, resetAt.UnixNano())
+		} else {
+			b.reset = resetAt
+		}
 	} else if reset != "" {
 		// Calculate the reset time by using the date header returned from discord
 		discordTime, err := http.ParseTime(headers.Get("Date"))